Deep Learning Approaches for the Classification of Keloid Images in the Context of Malignant and Benign Skin Disorders

Abstract

Background/Objectives: Misdiagnosing skin disorders leads to the administration of wrong treatments, sometimes with life-impacting consequences. Deep learning algorithms are becoming more and more used for diagnosis. While many skin cancer/lesion image classification studies focus on datasets containing dermatoscopic images and do not include keloid images, in this study, we focus on diagnosing keloid disorders amongst other skin lesions and combine two publicly available datasets containing non-dermatoscopic images: one dataset with keloid images and one with images of other various benign and malignant skin lesions (melanoma, basal cell carcinoma, squamous cell carcinoma, actinic keratosis, seborrheic keratosis, and nevus). Methods: Different Convolution Neural Network (CNN) models are used to classify these disorders as either malignant or benign, to differentiate keloids amongst different benign skin disorders, and furthermore to differentiate keloids among other similar-looking malignant lesions. To this end, we use the transfer learning technique applied to nine different base models: the VGG16, MobileNet, InceptionV3, DenseNet121, EfficientNetB0, Xception, InceptionRNV2, EfficientNetV2L, and NASNetLarge. We explore and compare the results of these models using performance metrics such as accuracy, precision, recall, F1score, and AUC-ROC. Results: We show that the VGG16 model (after fine-tuning) performs the best in classifying keloid images among other benign and malignant skin lesion images, with the following keloid class performance: an accuracy of 0.985, precision of 1.0, recall of 0.857, F1 score of 0.922 and AUC-ROC value of 0.996. VGG16 also has the best overall average performance (over all classes) in terms of the AUC-ROC and the other performance metrics. Using this model, we further attempt to predict the identification of three new non-dermatoscopic anonymised clinical images, classifying them as either malignant, benign, or keloid, and in the process, we identify some issues related to the collection and processing of such images. Finally, we also show that the DenseNet121 model has the best performance when differentiating keloids from other malignant disorders that have similar clinical presentations. Conclusions: The study emphasised the potential use of deep learning algorithms (and their drawbacks), to identify and classify benign skin disorders such as keloids, which are not usually investigated via these approaches (as opposed to cancers), mainly due to lack of available data.

Full text

Academic Editor: Dechang Chen
Received: 11 February 2025
Revised: 2 March 2025
Accepted: 8 March 2025
Published: 12 March 2025
Citation: Adebayo, O.E.; Chatelain,
B.; Trucu,D.; Eftimie, R. Deep
Learning Approaches for the
Classification of Keloid Images in the
Context of Malignant and Benign Skin
Disorders. Diagnostics 2025,15, 710.
https://doi.org/10.3390/
diagnostics15060710
Copyright: © 2025 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license
(https://creativecommons.org/
licenses/by/4.0/).
Article
Deep Learning Approaches for the Classification of Keloid
Images in the Context of Malignant and Benign Skin Disorders
Olusegun Ekundayo Adebayo 1, Brice Chatelain 2, Dumitru Trucu 3,* and Raluca Eftimie 1,3, *
1Laboratoire de Mathématiques de Besançon, Université Marie et Louis Pasteur, F-25000 Besançon, France;
olusegun.adebayo@univ-fcomte.fr
2Service de Chirurgie Maxillo-Faciale, Stomatologie et Odontologie Hospitalière, CHU Besançon,
F-25000 Besançon, France; bchatelain@chu-besancon.fr
3Division of Mathematics, University of Dundee, Dundee DD1 4HN, UK
*Correspondence: trucu@maths.dundee.ac.uk (D.T.); raluca.eftimie@univ-fcomte.fr (R.E.)
Abstract: Background/Objectives: Misdiagnosing skin disorders leads to the adminis-
tration of wrong treatments, sometimes with life-impacting consequences. Deep learning
algorithms are becoming more and more used for diagnosis. While many skin cancer/lesion
image classification studies focus on datasets containing dermatoscopic images and do not
include keloid images, in this study, we focus on diagnosing keloid disorders amongst other
skin lesions and combine two publicly available datasets containing non-dermatoscopic
images: one dataset with keloid images and one with images of other various benign
and malignant skin lesions (melanoma, basal cell carcinoma, squamous cell carcinoma,
actinic keratosis, seborrheic keratosis, and nevus). Methods: Different Convolution Neural
Network (CNN) models are used to classify these disorders as either malignant or benign,
to differentiate keloids amongst different benign skin disorders, and furthermore to dif-
ferentiate keloids among other similar-looking malignant lesions. To this end, we use the
transfer learning technique applied to nine different base models: the VGG16, MobileNet,
InceptionV3, DenseNet121, EfficientNetB0, Xception, InceptionRNV2, EfficientNetV2L, and
NASNetLarge. We explore and compare the results of these models using performance
metrics such as accuracy, precision, recall, F1
score
, and AUC-ROC. Results: We show that
the VGG16 model (after fine-tuning) performs the best in classifying keloid images among
other benign and malignant skin lesion images, with the following keloid class performance:
an accuracy of 0.985, precision of 1.0, recall of 0.857, F1 score of 0.922 and AUC-ROC value
of 0.996. VGG16 also has the best overall average performance (over all classes) in terms of
the AUC-ROC and the other performance metrics. Using this model, we further attempt
to predict the identification of three new non-dermatoscopic anonymised clinical images,
classifying them as either malignant, benign, or keloid, and in the process, we identify
some issues related to the collection and processing of such images. Finally, we also show
that the DenseNet121 model has the best performance when differentiating keloids from
other malignant disorders that have similar clinical presentations. Conclusions: The study
emphasised the potential use of deep learning algorithms (and their drawbacks), to identify
and classify benign skin disorders such as keloids, which are not usually investigated via
these approaches (as opposed to cancers), mainly due to lack of available data.
Keywords: deep learning; classification; benign skin disorders; malignant skin disorders;
keloid; dermatoscopic images; non-dermatoscopic images; transfer learning; CNN
Diagnostics 2025,15, 710 https://doi.org/10.3390/diagnostics15060710

Diagnostics 2025,15, 710 2 of 30
1. Introduction
Correctly identifying and classifying skin disorders is an important topic in the context
of malignant vs. benign disorders, when patient treatment needs to start very quickly. There
are various clinical examples where patients with certain malignant skin disorders have
been misdiagnosed as benign disorders (because of their benign appearance), and thus they
were not offered the appropriate treatment [
1
–
9
]. Such misdiagnoses are because clinical
images of certain benign disorders are very similar to malignant cancers, and medical
practitioners might not have the appropriate expertise to distinguish them.
In this study, we focus on identifying keloid lesions either as a benign skin disorder
amongst different malignant and benign skin disorders, simply identifying keloids as
keloids, or identifying keloids among other malignant lesions that look similar. Keloids are
a particular type of fibroproliferative skin disorder characterised by increased deposition of
collagen in the area of an initial wound (i.e., a raised “scar”) and the subsequent invasion
of this lesion into the adjacent healthy tissue. This invasion aspect made researchers
compare keloids with cancers [
10
–
13
]. Moreover, due to these differences and similarities
between keloids and skin cancers [
14
], various studies and clinical case reports have
mentioned the misdiagnosis of some skin cancers as keloids due to their similar clinical
presentation; see, for example [
1
,
4
,
5
,
7
,
9
,
15
,
16
]. Also, the misdiagnosis of keloids as tumours
can lead to improper surgical treatment that could lead to worse outcomes (i.e., patient
disfigurement) [
17
]. Even if we acknowledge that such misdiagnoses are rare, there is
the question (and opportunity) of using new computational approaches to improve the
diagnosis of different benign and malignant skin disorders (when histopathology tests are
not immediately available or when expert dermatologists are not immediately available).
We need to emphasise that the classical approach to diagnose various skin disorders
uses dermatoscopy [
18
–
20
]. However, this medical device is perceived to be too complex
and quite expensive to use [
21
,
22
]. Much less expensive options are classical point-and-
shoot cameras or smartphone cameras that produce non-dermatoscopic images. However,
the accuracy of the diagnostic results for various skin disorders (in particular for non-
dermatoscopic images) is still not consistently optimal, and further research is needed to
enhance its reliability and effectiveness in clinical practice [
23
]. With the rise of artificial
intelligence, machine learning and deep learning approaches have been developed and
employed to help clinicians diagnose different skin disorders and to provide faster and
more accurate diagnostic results.
Previous Work
Studies have shown that deep learning approaches perform better than machine
learning approaches when it comes to image classification [
24
–
26
]. Thus, several studies
have used deep learning approaches for biomedical image classification, e.g., lesion clas-
sification [
27
,
28
], breast cancer classification [
29
,
30
], etc., and for classifying benign and
malignant skin disorders [
31
]. However, to the best of our knowledge, there are no studies
that include the classification of keloids as benign or malignant skin disorders, nor do any
studies exist involving their identification among benign and malignant skin lesions using
non-dermatoscopic (clinical) image datasets.
Nevertheless, we note that there are studies that investigate keloids (and other scars)
using machine learning and deep learning approaches. In [
32
], post-thyroidectomy scars
(including keloid) were classified based on their severity. More precisely, the authors
in [
32
] focused on images of post-thyroidectomy scars (which included hypertrophic scars
and keloids), along with clinical features such as age, body mass index (BMI), and scar
symptoms, and they used a convolutional block attention module integrated with a ResNet-
50 model for the image-based severity prediction, achieving an accuracy of 0.733, AUC-ROC

Diagnostics 2025,15, 710 3 of 30
of 0.912, and a recall of 0.733 on an external test set when the images were combined with
clinical data. However, more datasets are needed to improve model generalisation, and
clinical validation is needed to ascertain the performance of the model. In [
33
], a cascaded
vision transformer architecture was proposed for keloid segmentation and identification
by focusing on the blood perfusion rate and growth rates of keloid lesions. The data used
in this study were intensity and blood perfusion images of 150 untreated keloid patients,
with images showing keloids from various body regions like the chest, face, and limbs
obtained via Laser Speckle Contrast Imaging (LSCI), and they achieved an average accuracy
of 0.927. However, more datasets are needed to improve model generalisation, and clinical
validation is needed to ascertain the result of the model.
In addition to these very few keloid-focused studies, there are many more studies that
focus on general tissue lesions (see also Table 1). For example, in [
28
] a novel Generative
Adversarial Networks (GANs) architecture for generating medical images was introduced
and used to generate synthetic liver lesion images (as opposed to the traditional data
augmentation technique). They showed that adding these synthetically generated images
to the original images (containing computed tomography images of 182 liver lesions)
and training the model on these images improved the performance of their convolution
neural network (CNN) model (approx. 7% increase in performance) compared to using the
classical augmentation method available in deep learning, achieving an AUC-ROC of 0.93,
a recall of 0.857, and a specificity of 0.924. A limitation to this study is that GAN-generated
images may not fully capture real variations. Another is that the size of the dataset may
prevent generalisation. In [
27
], a new non-dermatoscopic skin lesion dataset [
34
] (that
has also been used in our current study) was introduced. The dataset included clinical
images and related patient clinical information. The study in [
27
] showed that combining
these patients’ clinical features (i.e., information) of non-dermatoscopic medical data with
their images improved the performance of CNN models. They achieved an accuracy of
0.788
±
0.025, a precision of 0.8
±
0.028, and a recall of 0.788
±
0.025. Despite the promising
result of the proposed model, the model may be too dependent on patient information,
and the size of the data will likely cause the model not to generalise well, especially when
tested on skin lesions found in anatomical regions not present or under-represented in
the training dataset. In [
31
], a novel CNN model named SkinNet-16 was proposed for
accurate early detection of skin lesions using two public dermatoscopic skin lesion images
from [
35
,
36
]. The images were cleaned, feature extraction was done, and the proposed
model was trained with the extracted features. The model achieved an accuracy of 0.992
and a recall of 0.98. The limitation of this study includes the fact that only a binary model
was proposed despite the training dataset containing more that two skin lesions. Also,
more training data might be needed for improved generalisation ability of the model.
In this study, we focus on CNN models, since they are the most widely used models.
We acknowledge that there are other models such as vision transformers (ViTs) [
37
], which
were originally used in natural language processing (NLP) and now are being used for
image classification because they require fewer computations to achieve close enough
performance compared to CNN models. Nevertheless, current studies show that they do
not outperform CNN models except when trained on very large datasets [
38
,
39
]. For this
reason (and because we do not have such large datasets), in this study we decided to focus
exclusively on CNN models. The goal of our current article is threefold:
i.
To train various CNN models to classify keloid images as benign lesions and not as ma-
lignant lesions (and for this, we use a large variety of skin lesion images, from keloids
to melanoma, basal cell carcinoma, squamous cell carcinoma, seborrheic keratosis,
nevus, etc., which could be either malignant or benign);

Diagnostics 2025,15, 710 4 of 30
ii.
To train the CNN models to classify images of various skin disorders in three separate
classes: malignant, benign, or keloid;
iii. To train the CNN models to identify keloid images among other malignant lesions that
can mimic keloids, e.g., basal cell carcinoma [
40
,
41
] and squamous cell carcinoma [
42
].
Note also that our focus in this study is on the following:
a. The identification and classification of keloid lesions.
b.
Using non-dermatoscopic (clinical) images.
We aim to train the CNN models such that they can be used in communities where dermato-
scopes are either rare or unavailable (and where dermatologists might not be available).
Table 1. Summary of some previous studies on medical image classification.
Study Dataset Algorithm Category Results Limitation
[28]
Liver lesion
images (CT
scans)
GAN + CNN
Benign vs.
Malignant Liver
Lesions
Improved CNN accuracy
with GAN augmentation
(AUC-ROC = 0.93,
recall = 0.857,
specificity = 0.924)
GAN-generated
images may not
fully capture real
variations; limited
dataset size
[27]
Non-
dermatoscopic
skin lesion
images + clinical
data
Several CNN
classifiers
Skin Cancer
Detection
Accuracy = 0.788 ±0.025,
precision = 0.8 ±0.028,
recall = 0.788 ±0.025
Dependency on
patient metadata;
limited
generalisation
ability
[29]
Mammogram
images tested on
the MIAS and
DDSM
ResNet101 +
Metaheuristic
optimisation
Breast Cancer
Classification
DDSM (Accuracy = 0.986,
recall 0.987) MIAS
(accuracy = 0.992,
recall = 0.979)
High
computational cost;
possible overfitting
[31]HAM10000 +
ISIC dataset
proposed
CNN-based
SkinNet-16
Benign vs.
Malignant Skin
Lesions
Accuracy ≈0.992,
recall ≈0.98
Limited dataset;
only binary
classification
[32]
Post-operative
scar images +
clinical data
Proposed CBAM
+ ResNet50
Scar Severity
Prediction
Accuracy = 0.733,
AUC-ROC = 0.912,
recall = 0.733
Variability in scar
assessment; need
for real-world
validation
[33]
Keloid images
(Laser Speckle
Contrast
Imaging)
Proposed a
cascaded vision
transformer
architecture
Keloid Severity
Evaluation Average accuracy = 0.927
Limited dataset;
requires clinical
validation
The paper is structured as follows. In Section 2, we discuss the data used for the
experiments, the deep learning approaches considered, the data preprocessing steps, the ar-
chitecture of the models, and the performance metrics. In Section 3, we discuss the results of
the classification tasks for the three goals mentioned above using the performance metrics
from Section 2while highlighting the best models. In Section 4, we discuss the general
overview of this study and its limitations. In Section 5, we present the conclusion, and in
Section 6, we conclude with a summary of the results while also comparing them with
some results from the published literature.

Diagnostics 2025,15, 710 5 of 30
2. Materials and Methods
2.1. Data
In this study, we made use of two non-dermatoscopic datasets:
•
The first dataset was taken from [
34
] and was obtained using a smartphone-based
application created to help doctors and medical students collect clinical photos of
skin lesions, as well as clinical information about the patient. Each dataset sample
contains up to 26 attributes: a clinical diagnosis; a picture of the lesion; the patient’s
age; the location of the lesion; if the lesion itches, bleeds, or has bled, hurts, has recently
risen, has altered its pattern, is elevated, etc. The dataset contains 2298 images taken
from 1373 patients between 2018 and 2019, which are available in Portable Network
Graphics (PNGs) in raw format (i.e., just as they were taken). Each image in this
dataset focused on one of the following six common skin disorders (also summarised
in Table 2): melanoma, basal cell carcinoma, squamous cell carcinoma (which are
considered malignant), actinic keratosis, seborrheic keratosis, and nevus (which are
considered benign). We emphasise that 58.4% of the skin lesions and 100% of the
skin cancers in this dataset are histopathology-confirmed (and this is about the same
percentage as for the ISIC [43] dataset). For more information on the data, see [34].
Figure 1i shows an example of some of the images from the first image dataset [
34
]
divided into their respective pathological classification (i.e., either malignant or being),
while Figure 1ii shows the anatomical region of these lesions.
•
The second dataset is taken from [
44
] and consists of different non-dermatoscopic
keloid images; we chose 274 keloid images and cropped them to ensure that these
images are consistent with the (zoomed-in) images in the first dataset. Figure 2shows
a sample of the (keloid) images in this second dataset [44].
Table 2. Pathological classification of the skin disorders image dataset in [
34
], detailing the total
number of images available for each skin disorder, the total images classified as malignant or benign,
and the overall number of images in the dataset.
Diagnostic Class Skin Disease Nbr. of Images Total Nbr. Images
Melanoma 52
Malignant Basal Cell
Carcinoma 845 1089
Squamous Cell
Carcinoma 192
Seborrheic
Keratosis 235
Nevus 244 1209
Benign Actinic Keratosis 730
Total Images 2298
Note that there are studies that have shown that some of the skin disorders classified
as benign in Table 2may become malignant [
45
,
46
]; e.g., in [
45
], the authors showed that
seborrheic keratosis can transform into squamous cell carcinoma, and in [
46
], the authors
reported cases of malignant dermatofibroma. In this study, we categorised each of these
skin disorders as either malignant or benign, as shown in Table 2. Furthermore, we cat-
egorised the keloids as “benign“ for the first classification task and as “keloids” for the
second classification task.

Diagnostics 2025,15, 710 6 of 30
Melanoma Squamous Cell Carcinoma
Seborrheic Keratosis
Actinic Keratosis
Basal cell carcinoma
Nevus
(a)
(b)
(i)
(ii)
Figure 1. (i) A sample of the image dataset in [
34
] (the PAD dataset) for (a) malignant and (b) benign
skin cancers. (ii) The anatomical regions where these skin lesions were found, as specified in [34].
Figure 2. A sample of the cropped keloid images from the Kaggle dataset (see [
44
]). Note that this
keloid dataset did not contain any information about anatomical regions where the keloid lesions
were found.
2.2. Deep Learning Approaches
While the classical dense (i.e., fully connected) layers were the first used in deep
learning, they have been largely replaced by convolutional neural networks (CNNs, also
known as covnets), as they have been shown to perform better than densely connected
layers [
24
]. One of the reasons for which convolutional layers tend to perform better is
because they can learn local patterns from their input feature space compared to dense
layers that learn global patterns. Also, the patterns learned by a convolutional layer are
translation-invariant, while those of dense layers are not [24].

Diagnostics 2025,15, 710 7 of 30
Due to the high computational cost of training new algorithms with high predictive
performance, we decided to use transfer learning approaches for nine different already-
published deep learning models—VGG16 [
47
], InceptionV3 [
48
], DensNet121 [
49
], Mo-
bileNet [
50
], EfficientNetB0 [
51
], Xception [
52
], InceptionRNV2 [
53
], EfficientNetV2-L [
54
],
and NASNet-L [
55
]—which have been shown to perform well on biomedical image clas-
sification [
56
–
64
]. These models were pretrained on the Imagenet database, and in this
study, we used them as base models. In the beginning, the weights of each base model
were frozen, and the models were only trained on an additional dense input layer with
1024 neurons after a 2D global average pooling was done, followed by a unit dense output
layer with a sigmoid activation function, as appropriate for binary classification. For a
detailed description of these models, see Section 2.2.2 below.
2.2.1. Data Preprocessing
In the first step, we loaded images available in the first dataset [
34
], which sums
up to a total of 2298 coloured images with pixel values ranging between 0 and 255. We
then recategorised the datasets based on the “diagnostic“ feature in relation to each image
identified with their “im_id” feature. Next, we split the dataset into train, validation,
and test sub-sets using stratified shuffle split from scikit learn library to preserve the
percentage of the classes as in the original dataset (i.e., this makes sure the split data
contain the proportionate percentage of the skin disorders when compared to the original
dataset). We also applied the same split method to the second dataset (containing only
keloid lesions) [
44
] and then added them together. The dataset was then split into train
(80%) and test (20%) sets, while 90% of the train set was used for training, and 10% was
used to validate the model. On the whole dataset, we applied the following standard
preprocessing rules: (i) we rescaled each pixel value for each image matrix to lie between
0 and 1; (ii) we resized each image to a target size of our choice (i.e., 128
×
128). We then
categorised each skin disorder as malignant or benign: melanoma, basal cell carcinoma and
squamous cell carcinoma were classified as malignant [
65
] (a total of 1089 such images),
while seborrheic keratosis, nevus, and actinic keratosis were classified as benign [
66
] (a
total of 1209 such images).
Since the dataset was imbalanced and small, we first applied random oversampling
on the train set to tackle the class imbalance issue and increase the dataset. We also applied
data augmentation techniques, e.g., image flip, rotation, zoom, shift, and shear, to a random
sample of each class in the original train dataset to introduce some variability and also
increase the size. It is important to note that the oversampling and data augmentation
methods detailed above were only applied to the train sets, while we kept the original
images in the validation and test sets.
2.2.2. Model Details
Building an efficient custom image classification CNN model from scratch with high
accuracy is computationally demanding. Also, insufficient data might lead to inaccurate
categorisation, and CNN training could take a while to converge [
67
]. Hence, most im-
age classification tasks employ transfer learning methods, where already-trained models
(considered base models) are reused on new datasets. Figure 3shows the transfer learning
framework used in this study: we removed the top layers containing the classification
layers and replaced them with a hidden classification block (containing a dense layer(s)
followed by an output layer depending on the number of classes).

Diagnostics 2025,15, 710 8 of 30
Pretrained
CNN Hidden layers Prediction
(Sigmoid)
Input
"Benign"
Figure 3. The transfer learning framework.
Below, we introduce the models considered in this study:
1.
VGG16: VGG16 was proposed in [
47
] by the Visual Geometry Group (VGG) at
the University of Oxford. It consists of 13 convolutional layers (with 3
×
3 filter
sizes), 5 max-pooling layers, and 3 dense (i.e., fully connected) layers including the
output layer. The rectified linear unit (ReLU) activation function is used for each
convolutional and dense layer except for the output layer, which uses the “softmax”
activation function. Each convolutional layer use a 3
×
3 convolutional filter with a
stride of 1 and same padding, which makes the resulting feature maps retain the same
spatial dimensions as the input. The convolutional layers are stacked on each other,
with the number of input filters doubled after each max-pooling layer (with a stride
of 2). The depth of the convolutional layers is also increased monotonically. Figure 4
and Table 3show a summary of the VGG16 architecture.
2 x Conv
Filter size: 3x3
Stride: 1
with padding
Max Pool
Filter size: 2x2
Stride: 2
2 x Conv
Filter size: 3x3
Stride: 1
with padding
Max Pool
Filter size: 2x2
Stride: 2
Max Pool
Filter size: 2x2
Stride: 2
3 x Conv
Filter size: 3x3
Stride: 1
with padding
3 x Conv
Filter size: 3x3
Stride: 1
with padding
Max Pool
Filter size: 2x2
Stride: 2
3 x Conv
Filter size: 3x3
Stride: 1
with padding
Max Pool
Filter size: 2x2
Stride: 2
Flatten 2 x Dense
Output Dense
+ Softmax
Classification Block
Figure 4. VGG16 architecture [
49
] highlighting the classification block, which is usually replaced
during transfer learning.
Table 3. Classical VGG16 architecture.
Layers Output Size VGG16
Convolution 224 ×224 7 ×7 conv
Pooling 112 ×112 2 ×2 max pool
Convolution 56 ×56 3 ×3 conv
Pooling 28 ×28 2 ×2 max pool
Convolution 14 ×14 3 ×3 conv
Pooling 7 ×7 2 ×2 max pool
Fully Connected 4096 Fully connected layer
Fully Connected 4096 Fully connected layer
Output 1000 (classes) Output layer with
softmax activation

Diagnostics 2025,15, 710 9 of 30
2.
DenseNet121: This model, which was proposed in [
49
], is a feedforward network
that connects all layers to all other layers in a feedforward fashion. It comprises
121 layers and consists of densely connected convolutional layers within dense blocks,
promoting feature reuse and gradient flow. The model also consists of transition
layers which help to control the growth of complexity between blocks. Figure 5and
Table 4show a summary of the DenseNet121 architecture.
Input
Prediction
Dense Block 2
Linear
Conv
Dense Block 1
Pooling
Conv
Conv
Pooling
"Benign"
Pooling
Dense Block 3
Figure 5. An example of a deep DenseNet architecture with three dense blocks [
49
], and with a keloid
image as input. The layers between two adjacent blocks are referred to as transition layers.
Table 4. Classical DenseNet121 architecture.
Layers Output Size DenseNet121
Convolution 112 ×112 7 ×7 conv
Pooling 56 ×56 3 ×3 max pool
Dense Block 56 ×56 6 layers
Transition Layer 28 ×28 1 ×1 conv, 2 ×2 avg pool
Dense Block 28 ×28 12 layers
Transition Layer 14 ×14 1 ×1 conv, 2 ×2 avg pool
Dense Block 14 ×14 24 layers
Transition Layer 7 ×7 1 ×1 conv, 2 ×2 avg pool
Dense Block 7 ×7 16 layers
Pooling 1 ×1 7 ×7 avg pool
Fully Connected 1024 Fully connected layer
Output 1000 (classes) Output layer with
softmax activation
3.
InceptionV3: InceptionV3 is an extension of GoogleNet (a model that has been
shown to demonstrate strong classification performance in some biological appli-
cations [
68
,
69
]). InceptionV3 uses the inception model to minimise the number of
parameters required to be trained, thus lowering the computational cost by concate-
nating numerous convolutional filters of various sizes into a new filter. Figure 6and
Table 5summarise the InceptionV3 architecture.
4.
MobileNet: The MobileNet model uses depthwise separable convolutions (a form of
factorised convolutions that factorise a classical convolution into a depthwise convolu-
tion and a 1
×
1 convolution called a pointwise convolution) to reduce computational
cost and model size. In one step of these convolutions, the depthwise convolution
applies a single filter per input channel, and the pointwise convolution applies a 1
×
1
convolution to combine the outputs of the depthwise convolution. For more details
on MobileNet, see [50]. Table 6summarises the MobileNet architecture.

Diagnostics 2025,15, 710 10 of 30
Conv
Filter size: 3x3
Stride: 2
Conv
Filter size: 3x3
Stride: 1
Conv + padding
Filter size: 3x3
Stride: 1
Max Pool
Filter size: 3x3
Stride: 2
Conv
Filter size: 3x3
Stride: 2
Conv
Filter size: 3x3
Stride: 1
Conv
Filter size: 3x3
Stride: 1
Inception
Model 1
x 3
Inception
Model 2
x 5
Inception
Model 3
x 2
Avg Pool
Filter size: 8x8
Stride: 0
Linear
Logits
Figure 6. Summary of Inception-V3 model architecture (see also Table 5. For explicit details on the
structure of Inception Model 1, Inception Model 2, and Inception Model 3, see [48].
Table 5. Classical InceptionV3 architecture.
Layers Output Size InceptionV3
Convolution 149 ×149 3 ×3 conv
Convolution 147 ×147 3 ×3 conv
Convolution 147 ×147 3 ×3 conv
Pooling 73 ×73 3 ×3 max pool
Convolution 71 ×71 3 ×3 conv
uses the Convolution 35 ×35 3 ×3 conv
Inception Module 35 ×35 Inception A
Inception Module 17 ×17 Inception B
Inception Module 8 ×8 Inception C
Pooling 1 ×1 8 ×8 avg pool
Fully Connected 2048 Fully connected layer
Output 1000 (classes) Output layer with
softmax activation
Table 6. MobileNet architecture.
Layers Output Size MobileNet
Convolution 112 ×112 3 ×3 conv
Depthwise Separable Conv 112 ×112
3
×
3 depthwise, 1
×
1 pointwise
Pooling 56 ×56 2 ×2 max pool
Depthwise Separable Conv 56 ×56
3
×
3 depthwise, 1
×
1 pointwise
Depthwise Separable Conv 28 ×28
3
×
3 depthwise, 1
×
1 pointwise
Pooling 14 ×14 2 ×2 max pool
Depthwise Separable Conv 14 ×14
3
×
3 depthwise, 1
×
1 pointwise
Depthwise Separable Conv 7 ×7
3
×
3 depthwise, 1
×
1 pointwise
Pooling 1 ×1 7 ×7 avg pool
Fully Connected 1024 Fully connected layer
Output 1000 (classes) Output layer with
softmax activation
5.
EfficientNetB0: EfficientNet is a family of CNNs that proposed a new scaling method
after carefully identifying that better accuracy could be achieved when the depth,
width, and resolution of the network are carefully balanced. The proposed scaling

Diagnostics 2025,15, 710 11 of 30
approach uses a simple compound coefficient to uniformly scale the depth, width, and
resolution to reduce the computational cost and the model size. EfficientNet consists
of 8 models in the range
B
0
–B
7, where the
B
1
–B
7 models are scaled increasingly
from the baseline
B
0 model using different compound coefficients. The baseline
model EfficientNetB0 is based on mobile inverted bottleneck convolution (MBConv)
blocks [
70
]. In this study, we focus only on the baseline model (EfficientNetB0) due to
limited computational resources. For more details on EfficientNet, see [
51
]. Table 7
summarises the EfficientNetB0 architecture.
Table 7. EfficientNet-B0 architecture.
Layers Output Size EfficientNet-B0
Convolution 112 ×112 3 ×3 conv, 32 filters, stride 2
MBConv1 112 ×112 3 ×3 depthwise, 16 filters
MBConv6 56 ×56 3 ×3 depthwise, 24 filters, stride 2
MBConv6 28 ×28 5 ×5 depthwise, 40 filters, stride 2
MBConv6 14 ×14 3 ×3 depthwise, 80 filters, stride 2
MBConv6 14 ×14 5 ×5 depthwise, 112 filters
MBConv6 7 ×7 5 ×5 depthwise, 192 filters, stride 2
MBConv6 7 ×7 3 ×3 depthwise, 320 filters
Convolution 7 ×7 1 ×1 conv, 1280 filters
Pooling 1 ×1 Global average pooling
Fully Connected 1280 Fully connected layer
Output 1000 (classes) Output layer with softmax activation
6.
Xception: The Xception model uses depthwise separable convolutions like MobileNets
and the Inception models described above. However, this model is based on the
hypothesis that cross-channels correlations and spatial correlations mappings in the
feature maps of CNNs can be decoupled entirely. This is a stronger assumption than
that of the Inception models, and hence the name “Xception” that derived from the
phrase “Extreme Inception”. For more details on the Xception model, see [
52
]. Table 8
summarises the Xception architecture.
Table 8. Xception architecture.
Layers Output Size Xception
Convolution 149 ×149 3 ×3 conv, 32 filters, stride 2
Convolution 147 ×147 3 ×3 conv, 64 filters
Entry Flow 73 ×73 3 ×SeparableConv layers, 128 filters, stride 2
Entry Flow 37 ×37 3 ×SeparableConv layers, 256 filters, stride 2
Entry Flow 19 ×19 3 ×SeparableConv layers, 728 filters, stride 2
Middle Flow 19 ×19 8 ×(3 ×SeparableConv layers, 728 filters)
Exit Flow 10 ×10 3 ×SeparableConv layers, 1024 filters, stride 2
Exit Flow 10 ×10 3 ×SeparableConv layers, 2048 filters
Pooling 1 ×1 Global average pooling
Fully Connected 2048 Fully connected layer
Output 1000
(classes) Output layer with softmax activation
7.
InceptionRNV2: This is the second verion of the InceptionResNet model. It is largely
similar to the Inception model already described above (see InceptionV3). However,
in InceptionResNet (i.e., a residual version of the Inception model), a cheaper Inception
block is used preceded by a filter expansion layer which scales up the dimension of
filter bank, hence cushioning the effect of the the dimensionality reduction caused

Diagnostics 2025,15, 710 12 of 30
by the Inception block. Another distinction between the InceptionResNet models
and the vanilla Inception models is that batch normalisation is applied only to the
standard layers not the summations, hereby increasing the overall number of Inception
blocks. In this study, we used the second version of the InceptionResNet model.
For more details on the InceptionResNet model, see [
53
]. Table 9summarises the
InceptionResNetV2 architecture.
Table 9. InceptionResNetV2 architecture.
Layers Output Size InceptionResNetV2
Convolution 149 ×149 3 ×3 conv, 32 filters, stride 2
Convolution 147 ×147 3 ×3 conv, 32 filters
Convolution 73 ×73 3 ×3 conv, 64 filters, stride 2
Inception-ResNet-A 35 ×35 5 ×Inception-ResNet-A modules
Reduction-A 17 ×17 Transition layer (pooling, conv)
Inception-ResNet-B 17 ×17 10 ×Inception-ResNet-B modules
Reduction-B 8 ×8 Transition layer (pooling, conv)
Inception-ResNet-C 8 ×8 5 ×Inception-ResNet-C modules
Convolution 8 ×8 1 ×1 conv, 1536 filters
Pooling 1 ×1 Global average pooling
Fully Connected 1536 Fully connected layer
Output 1000
(classes) Output layer with softmax activation
8.
EfficientNetV2-L: This is an extension of the EfficientNet models described above
(see 5) with faster training time, because it uses about half the number of parameters
used in EfficientNet. In this new family of models, the scaling approach introduced
in [
51
] was combined with a training-aware neural architecture search (NAS) to
optimise the training speed and number of parameters. Unlike the original Effi-
cientNet, which uses depthwise separable convolutions, the fused-MBConv block
fuses the initial pointwise and depthwise convolutions, reducing the computational
cost. As opposed to the vanilla EfficientNet, EfficienNetV2 utilises both MBConv
and the fused-MBConv [
71
] in the early layers. This new family comprises three
variants (i.e., small, medium, and large) based on their size and performance. In this
study, we made use of the largest variant (i.e., EfficientNetV2-L). For more details on
EfficientNetV2, see [54]. Table 10 summarises the EfficientNetV2-L architecture.
Table 10. EfficientNetV2-L architecture
Layers Output Size EfficientNetV2-L
Convolution 224 ×224 3 ×3 conv, 32 filters, stride 2
MBConv1 112 ×112 3 ×3 depthwise, 32 filters
MBConv4 56 ×56 3 ×3 depthwise, 64 filters, stride 2
Fused-MBConv4 28 ×28 3 ×3 fused conv, 128 filters, stride 2
Fused-MBConv6 14 ×14 3 ×3 fused conv, 256 filters, stride 2
MBConv6 7 ×7 3 ×3 depthwise, 512 filters, stride 2
MBConv6 7 ×7 3 ×3 depthwise, 1280 filters
Convolution 7 ×7 1 ×1 conv, 1280 filters
Pooling 1 ×1 Global average pooling
Fully Connected 1280 Fully connected layer
Output 1000
(classes) Output layer with softmax activation

Diagnostics 2025,15, 710 13 of 30
9.
NASNet-L: This model was designed using a proposed search space called NAS
(that enables transferability) to find optimal network architectures. It utilises rein-
forcement learning to explore a preset search space of architectures while optimising
performance and efficiency and also using a regularisation technique called “Sched-
uledDropPath”. It has different variants, NASNet-A, B, and C, tailored for different
use cases (where A is the most accurate and was designed to deliver high performance,
while B and C provide a trade-off between efficiency and accuracy, with B being more
accurate than C). The model also includes large and small versions depending on
the resources available. The large model (i.e., NASNet-L) is particularly effective
for high-performance, while the small model (i.e., NASNetMobile) is particularly
effective for resource-constrained environments like mobile devices. In this study, we
used NASNet-L. For more details on NASNet models, see [
55
]. Table 11 summarises
the NASNet-L architecture.
Table 11. NASNet-L architecture.
Layers Output Size NASNetLarge
Convolution 224 ×224 3 ×3 conv, 96 filters, stride 2
Normal Cell 112 ×112 5 ×Normal cells, 168 filters
Reduction Cell 56 ×56 Transition layer (pooling, conv)
Normal Cell 56 ×56 5 ×Normal cells, 336 filters
Reduction Cell 28 ×28 Transition layer (pooling, conv)
Normal Cell 28 ×28 5 ×Normal cells, 672 filters
Reduction Cell 14 ×14 Transition layer (pooling, conv)
Normal Cell 14 ×14 5 ×Normal cells, 1344 filters
Reduction Cell 7 ×7 Transition layer (pooling, conv)
Normal Cell 7 ×7 5 ×Normal cells, 4032 filters
Pooling 1 ×1 Global average pooling
Fully Connected 4032 Fully connected layer
Output 1000
(classes) Output layer with softmax activation
Generally in transfer learning, we replace the classical output layer of each pretrained
model (i.e., 1000 output units) with the number of target classes
nc
that our classification
tasks require: usually nc−1 for binary classification and ncfor multiclass classification.
In this experiment, we used the pretrained models as feature extractors and built a
new neural network on top of them for our specific classification tasks: one binary class
classification (benign vs. malignant lesion) and one multiclass classification (here with
three classes: benign vs. malignant vs. keloid). We began the process by first initialising a
base model (i.e., VGG16, DenseNet121, InceptionV3, MobileNet, EfficientNetB0, Xception,
InceptionRNV2, EfficientNetV2-L, or NASNet-L), with weights pretrained on the ImageNet
dataset. As already explained above, we excluded the fully connected layers at the top
of the network (with output unit 1000), and specified the input shape parameter, which
defines the shape of the input data (which we chose to be 128
×
128
×
3 for our tasks, where
“3” corresponds to the number of channels in the images, as given by the RGB format). It is
important to note that images were in the RGBA format (i.e., 4 channels) and were then
converted to the RGB format before use for uniformity in the input shape.
Subsequently, we froze the weights of all layers in the pretrained model (i.e., we
did not train them on the new data) and added these base layers to a new output block.
This new output block is a Sequential model to which the pretrained model is added.
We incorporated the following into this new model: a GlobalAveragePooling2D layer (to
reduce the spatial dimensions of the output from the pretrained model), a dense layer

Diagnostics 2025,15, 710 14 of 30
with 1024 units, a ReLU activation function (to introduce non-linearity and learn high-level
features), and a dropout layer with a rate of 0.5 included to prevent overfitting (by randomly
setting half of the input units to zero during training). Lastly, a dense output layer with 3
(or 1) unit(s) and a softmax (or sigmoid) activation function were added for the multiclass
(or binary) classification task, where 3 corresponds to the number of target classes for the
multiclass classification, and 1 corresponds to binary classification (i.e., two target classes).
We made use of the binary cross-entropy loss function for the binary classification task
and the categorical cross-entropy loss function for the multiclasss task. Throughout these
experiments, we made use of the Adam optimiser with a learning rate equal to 0.0001 for
model training without fine-tuning and a rate of 0.00001 for model training with fine-tuning.
We trained all the CNN models over 50 epochs with a batch size of 32.
When training the models, we considered the following cases regarding the base
models and the train dataset:
Base model:
1.
Freezing the base model (i.e., training only the weights and biases of the added model).
2.
Fine-tuning the base model by unfreezing all layers of the base model (i.e., training
the weights and biases of the base model alongside the added model).
Train dataset:
1.
Training the model on the original train data after splitting it into train, validation,
and test datasets.
2. Training the model on an oversampled train dataset.
3. Training the model on an augmented train dataset.
Note that for each category of the train dataset considered above, we also considered
the two cases as regarded the state of the base model above.
2.2.3. Evaluation Metrics
For the classification of images with the different models, we employed the following
evaluation metrics that are given in terms of true positive (TP) values, true negative (TN)
values, false positive (FP) values, and false negative (FN) values: accuracy, precision, recall,
F1
score
, AUC-ROC and the confusion matrix. For more details on the these performance
metrics, see Appendix A. Note that each model was evaluated only on the out-of-sample
dataset, i.e., the test dataset, which had not been exposed to the model at all during the
training procedure.
3. Results
Here, we discuss the results obtained after applying the transfer learning technique
using the pretrained CNN models as base models. We start with a binary classification in
Section 3.1, followed by multiclass classifications in Sections 3.2 and 3.4. Regarding the
results in the tables below, we note that the values obtained for some of the pretrained
models improved as we fine-tuned them.
3.1. Binary Classification: Benign vs. Malignant Lesions
First, we present the performance of the CNN models trained on the original dataset
(i.e., before oversampling or applying data augmentation) on the test dataset. Table 12
shows the performance of the CNN models before fine-tuning the base models, while
Table 13 shows their performances after fine-tuning (i.e., training all layers of the base
models alongside the output layers).

Diagnostics 2025,15, 710 15 of 30
Table 12. Accuracy, precision, recall and F1 score for the pretrained models presented above (rows
1–9
). Here, the models were trained and validated on all raw skin lesion data (before any oversam-
pling or data augmentation; see Section 2.1), and before any fine-tuning of the pretrained models. In
the “Accuracy” column, we show in bold the largest value indicating the best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.7893 0.7642 0.7297 0.7465 0.8809
MobileNet 0.8008 0.7287 0.8468 0.7833 0.8776
DenseNet121 0.7969 0.7589 0.7658 0.7623 0.8841
InceptionV3 0.7356 0.6721 0.7387 0.7039 0.8123
EfficientNetB0 0.5747 0.0 0.0 0.0 0.5608
Xception 0.7635 0.7333 0.6937 0.7130 0.8280
InceptionRNV2 0.7165 0.6761 0.6396 0.6574 0.8131
EfficientNetV2L 0.5754 0.0 0.0 0.0 0.5943
NASNetLarge 0.6743 0.625 0.5856 0.6047 0.7370
In Table 12, we observe that MobileNet outperformed the rest of the models in ac-
curacy, recall, and F1
score
. This was followed by DenseNet121 on the same metrics, but
this one yielded the highest AUC-ROC score. VGG16 had the highest precision and only
performed worse than DenseNet121 in terms of AUC-ROC measure. EfficientNetB0 and
EfficientNetV2L had zero precision, recall, and F1
scores
; also, their accuracy and AUC
scores were low and close to each other, making them the worse models. The next-worse
models were the NASNetLarge and the Inception models (i.e., InceptionV3 and Incep-
tionRNV2). InceptionV3, however, performed better than VGGG16 in terms of recall rate.
In Table 13, we see that after fine-tuning, VGG16 outperformed the rest of the models
in accuracy, recall, F1
score
, and AUC-ROC. DenseNet121 had the highest precision rate
and the second-best in AUC-ROC score. We also see that after fine-tuning, the overall
performance of MobileNet reduced for almost all metrics except in terms of precision,
where it increased slightly. In general, the performance of the other models seems to have
increased after this fine-tuning, with the exception of MobileNet. Also, we see that the
precision, recall, and F1
scores
of EfficientNetB0 and EfficientNetV2L went from zero to
over 77%, implying that their initial weights and biases (i.e., initialised from ImageNet)
performed really poorly on the test dataset. Now, the EfficientNetV2L had the second
best accuracy and recall score, while the NASNetLarge model had the highest recall score.
The Xception model seems to have also performed well, ranking second best in terms
of precision.
Table 13. Accuracy, precision, recall, F1 score, and AUC of the pretrained CNN models (rows 1–9).
Here, the models were trained and validated on all raw skin lesion data (before any oversampling or
data augmentation; see Section 2.1), after fine-tuning the pretrained base models. In the “Accuracy”
column, we show in bold the largest value indicating the best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8467 0.8142 0.8288 0.8214 0.9274
MobileNet 0.7816 0.7455 0.7387 0.7421 0.8633
DenseNet121 0.8238 0.8495 0.7117 0.7745 0.9039
InceptionV3 0.8122 0.7719 0.7930 0.7822 0.8880
EfficientNetB0 0.8008 0.7706 0.7568 0.7636 0.8793
Xception 0.8161 0.8316 0.7117 0.7670 0.8958
InceptionRNV2 0.8161 0.7838 0.7838 0.7838 0.8793
EfficientNetV2L 0.8314 0.7815 0.8378 0.8097 0.8869
NASNetLarge 0.7203 0.6173 0.9009 0.7326 0.8501

Diagnostics 2025,15, 710 16 of 30
Next, we trained the CNN models on the oversampled data (where we randomly
increased the number of images for the classes with a lower number of images by adding
copies of these randomly selected images). In Table 14, we show their performance on the
test dataset without fine-tuning (i.e., without training the layers of the base models), while
in Table 15, we show their performance after fine-tuning (i.e., training all layers of the base
models). We see from Table 14 that VGG16 had the highest accuracy, and MobileNet had the
highest recall and F1
score
while sharing the same accuracy with DenseNet121. VGG16 had
the highest AUC-ROC, followed by DenseNet121 and MobileNet. EfficientNetB0 had the
lowest recall, F1
score
, and AUC-ROC score, while EfficientNetV2L had the lowest accuracy
and precision.
Table 14. Accuracy, precision, recall, F1 score, and AUC of the pretrained CNN models (rows 1–9)
trained on oversampled train dataset (to address the issue of imbalanced data classes; see Table 1) on
the test data. Here, the pretrained base models were not fine-tuned. In the “Accuracy” column, we
show in bold the largest value indicating the best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.7969 0.7900 0.7117 0.7488 0.8787
MobileNet 0.7893 0.7154 0.8378 0.7718 0.8674
DenseNet121 0.7893 0.7188 0.8288 0.7699 0.8742
InceptionV3 0.7203 0.6301 0.8288 0.7160 0.8001
EfficientNetB0 0.5785 0.5785 0.2162 0.3038 0.5615
Xception 0.7548 0.7156 0.7027 0.7091 0.8031
InceptionRNV2 0.7471 0.7064 0.6937 0.7000 0.8250
EfficientNetV2L 0.5402 0.4764 0.8198 0.6026 0.6011
NASNetLarge 0.6935 0.6449 0.6216 0.6330 0.7815
Table 15. Accuracy, precision, recall, F1score, and AUC of the transfer learning models (rows 1–9) on
the test dataset. Here, the models were trained on oversampled train dataset, and the pretrained base
models were fine-tuned. In the “Accuracy” column, we show in bold the largest value indicating the
best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8237 0.7982 0.7839 0.7909 0.9255
MobileNet 0.7931 0.7822 0.7117 0.7453 0.8668
DenseNet121 0.8084 0.8020 0.7297 0.7642 0.8917
InceptionV3 0.7854 0.7391 0.7658 0.7522 0.8641
EfficientNetB0 0.8391 0.8224 0.7928 0.8073 0.8978
Xception 0.8352 0.8269 0.7748 0.8000 0.9031
InceptionRNV2 0.8276 0.8000 0.7928 0.7964 0.8889
EfficientNetV2L 0.8199 0.7963 0.7748 0.7854 0.8591
NASNetLarge 0.7050 0.5966 0.9459 0.7317 0.8762
After fine-tuning (Table 15), we see that VGG16 outperformed the other models only
in AUC-ROC. EfficientNetB0 had the highest accuracy and the third best AUC score, while
NASNetLarge had again the highest recall score. Hence, overall, VGG16 seems to have the
highest ability to distinguish between malignant and benign skin lesions while possessing
a good trade-off between precision and recall (see the high F1score).
Finally, we trained the CNN models on the augmented data (obtained by applying
rotations, flips, zoom-ins, and shears to randomly selected images), and we present their
performance on the test data. This data augmentation approach not only increases the
number of images available for training (to solve the class imbalance issue) but also

Diagnostics 2025,15, 710 17 of 30
introduces variability in the train dataset. In Table 16, we present the performance of these
models (trained on the augmented dataset) before fine-tuning. Here, we see that MobileNet
outperformed the other models in all metrics except recall, and it possesses the best trade-
off between precision and recall. In contrast, EfficientNetB0 had the worst performance on
all metrics.
Finally, in Table 17 we show the performance these CNN models (trained on aug-
mented data) after the fine-tuning of base models. We see that VGG16 outperformed the
other models in accuracy, precision, and AUC-ROC followed by EfficientNetV2L, which
yielded the same accuracy, the highest F1
score
, and was the second best in terms of recall
and AUC-ROC.
Table 16. Accuracy, precision, recall, F1 score, and AUC of the transfer learning models (rows 1–9)
on the test dataset. Here, we trained all models on augmented train datasets (but before fine-tuning
of the base models). In the “Accuracy” column, we show in bold the largest value indicating the
best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8084 0.8081 0.7207 0.7619 0.8808
MobileNet 0.8276 0.8367 0.7387 0.7847 0.8965
DenseNet121 0.8084 0.7798 0.7658 0.7727 0.8962
InceptionV3 0.7165 0.6529 0.7117 0.6810 0.7974
EfficientNetB0 0.5057 0.4505 0.7387 0.5597 0.5575
Xception 0.7356 0.6875 0.6937 0.6906 0.8186
InceptionRNV2 0.7356 0.7100 0.6396 0.6730 0.8191
EfficientNetV2L 0.5556 0.4868 0.8288 0.6133 0.6845
NASNetLarge 0.7011 0.6774 0.5676 0.6176 0.7694
Overall, VGG16 showed the best performance on test data after fine-tuning when
the model was trained on raw training data (i.e, Table 13), oversampled training data
(i.e., Table 15), and augmented train data (i.e., Table 17).
Table 17. Accuracy, precision, recall, F1 score, and AUC for the transfer learning models (rows 1–9)
on the test dataset. Here, we trained these models on augmented train dataset and fine-tuned the
base models. In the “Accuracy” column, we show in bold the largest value indicating the best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8506 0.8600 0.7748 0.8152 0.9205
MobileNet 0.7854 0.7570 0.7297 0.7431 0.8728
DenseNet121 0.8467 0.8318 0.8018 0.8165 0.9101
InceptionV3 0.8352 0.8469 0.7477 0.7943 0.9168
EfficientNetB0 0.8199 0.7909 0.7838 0.7873 0.9020
Xception 0.8276 0.8113 0.7748 0.7926 0.9052
InceptionRNV2 0.8200 0.8137 0.7477 0.7793 0.8976
EfficientNetV2L 0.8506 0.8461 0.7928 0.8186 0.9197
NASNetLarge 0.8391 0.8556 0.7477 0.7981 0.9026
In Figure 7, we present the confusion matrices and AUC-ROC curves of the two best
models: (a,b) show the VGG16 model trained on the original dataset and fine-tuned; (c,d)
show the VGG16 model trained on the augmented dataset and fine-tuned.

Diagnostics 2025,15, 710 18 of 30
(a)
(c) (d)
(b)
Figure 7. The (a) confusion matrix and (b) AUC-ROC curve for VGG16 model trained and vali-
dated on the original train and validation datasets, respectively (while fine-tuning the base model),
and tested on the out-of-sample data (i.e., test data). (c) The confusion matrix and (d) AUC-ROC
curve for VGG16 model trained on augmented training data (while fine-tuning the base model).
3.2. Identifying Keloids as Keloids and Not Only as Benign vs. Malignant Skin Disorders
In what follows, we proceedd to train the models above not only to identify keloids
as “benign“ lesions but also to be able to distinctively identify keloids as “keloids” among
the various images of benign and malignant skin lesions. To this end, we trained the
models again (using transfer learning while maintaining the architecture for multiclass
classification, as described in Section 2.2.2) on the two datasets (i.e., the malignant vs.
benign lesions, as classified in Table 2and the keloid lesions, as shown in Figure 2), and
fine-tuned them.
We present their performances when trained on the original training dataset, oversam-
pled training dataset, and augmented training dataset:
(a)
Original training data: In Table 18, we present the performance of the 10 models
considered in this study (on test data) when they were trained for 50 epochs on the
original training dataset validated on the original validation dataset (while fine-tuning
all layers of the base models). We see here that VGG16 equally outperformed the rest
of the model on all metrics, followed by Xception and DensNet121, respectively.
(b)
Oversampled training data: In Table 19, we present the performance of the trained
model on oversampled data. Here, we see that the performance of each of the models
improved in comparison to Table 18. Here again, VGG16 performed better than the
rest of the models on all metrics followed by Xception (with a slightly lower AUC that
DenseNet121) and DenseNet121.
(c)
Augmented training data: In an attempt to further improve the result of the models,
instead of random oversampling, we applied data augmentation (such as rotations,
flips, zooms, etc.) to increase the training dataset to introduce variability and to help
the model generalise better. In Table 20, we see an increase in the performance of all
the models as expected in comparison to Tables 18 and 19. We see again that VGG16

Diagnostics 2025,15, 710 19 of 30
outperformed the rest of the models on all metrics followed by EfficientNetV2L and
InceptionV3, which both had lower AUC ROC scores compared to DenseNet121,
which had the second best AUC ROC score, while its other performances ranked
below InceptionV3.
We emphasise here that out of the pretrained CNN models used, MobileNet model
was the worst-performing model on the test data considered in this study.
Table 18. Accuracy, precision, recall, F1
score
, and AUC of the models (rows 1–9) on the test dataset.
Here, we trained and validated these models on the original train and validation datasets and fine-
tuned the base models. In the “Accuracy” column, we show in bold the highest value indicating the
best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8276 0.8301 0.8276 0.8276 0.9376
MobileNet 0.7471 0.7534 0.7471 0.7480 0.8914
DenseNet121 0.7931 0.7942 0.7931 0.7922 0.9305
InceptionV3 0.7816 0.7841 0.7816 0.7823 0.9095
EfficientNetB0 0.7624 0.7663 0.7624 0.7632 0.8925
Xception 0.8199 0.8245 0.8199 0.8202 0.9309
InceptionRNV2 0.7854 0.7854 0.7854 0.7854 0.9091
EfficientNetV2L 0.7778 0.7843 0.7778 0.7743 0.9111
NASNetLarge 0.6743 0.7266 0.6743 0.6685 0.8665
Table 19. Accuracy, precision, recall, F1
score
, and AUC of the models on the test dataset. Here, we
trained these models on oversampled train dataset and validated them on the original validation
datasets. The base models were fine-tuned. In the “Accuracy” column, we show in bold the largest
value indicating the best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8506 0.8530 0.8506 0.8502 0.9427
MobileNet 0.7625 0.7650 0.7625 0.7634 0.9032
DenseNet121 0.8084 0.8101 0.8084 0.8091 0.9295
InceptionV3 0.8008 0.8056 0.8008 0.8015 0.9205
EfficientNetB0 0.7969 0.7989 0.7969 0.7969 0.9148
Xception 0.8199 0.8239 0.8199 0.8205 0.9282
InceptionRNV2 0.8007 0.8036 0.8008 0.8008 0.9215
EfficientNetV2L 0.7969 0.7983 0.7969 0.7973 0.9179
NASNetLarge 0.7050 0.7222 0.7050 0.7061 0.8728
Table 20. Accuracy, precision, recall, F1
score
, and AUC of the models (rows 1–9) on the test dataset.
The models were trained on the augmented train dataset and validated on the original validation
datasets, with the base models fine-tuned.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8774 0.8813 0.8774 0.8778 0.9519
MobileNet 0.8046 0.8045 0.8046 0.8046 0.9240
DenseNet121 0.8391 0.8399 0.8391 0.8394 0.9452
InceptionV3 0.8467 0.8474 0.8467 0.8470 0.9403
EfficientNetB0 0.8429 0.8422 0.8429 0.8424 0.9406
Xception 0.8046 0.8072 0.8046 0.8046 0.9334
InceptionRNV2 0.8199 0.8222 0.8199 0.8206 0.9359
EfficientNetV2L 0.8467 0.8502 0.8467 0.8470 0.9322
NASNetLarge 0.8391 0.8410 0.8391 0.8394 0.9365

Diagnostics 2025,15, 710 20 of 30
Finally, we show in Figure 8the confusion matrix and AUC-ROC curves for the best
model: the VGG16 model trained on the augmented data.
(a) (b)
Figure 8. The (a) confusion matrix and (b) AUC-ROC curve for VGG16 trained on the augmented
train data validated on the original validation dataset (while fine-tuning the base model) and tested
on the test data.
From the confusion matrix and the AUC-ROC curves, we see that the model’s ability
to distinguish keloids from the other classes is better than its ability to distinguish the other
classes (i.e., the green curve in Figure 8b), though it also performed excellently well in
distinguishing malignant and benign skin disorders individually, with its AUCs equal to
0.9294 and 0.9289, respectively.
3.3. New Test Data: Clinical Images
To conclude the testing of our algorithm, we have finally used three new clinical
anonymised images (see Figure 9a provided by B. Chatelain). To classify them, we used the
our proposed algorithms (i.e., the VGG16 model trained to classify malignant vs. benign
images, as well as the VGG16 model trained to classify malignant vs. benign vs. keloid im-
ages). To this end, we zoomed-in the original images (see Figure 9b) to be consistent with the
other images in the datasets. We can see that the first two images, i.e., Figure 9b(i,ii), have
been correctly identified in both cases as malignant. In contrast, the third zoomed-in image
(i.e., Figure 9b(iii)) has been incorrectly identified as “benign“ in one case and as “keloid”
in the second case.
Since trained clinicians can diagnose the lesions from a distance, we emphasise that
we also tried to classify the original images (non-zoomed-in; see Figure 9a). In that case, we
obtained a correct result with the binary algorithm for Figure 9a(i,iii). In contrast, with the
multiclass algorithm, we obtained a correct prediction for Figure 9a(i,iii) but an incorrect
result for Figure 9a(ii).
Remark 1. There were a few issues with the classification of the new images in Figure 9. First,
the original images were taken at a distance, and to use the previous algorithms, we had to crop
them so we could zoom-in to focus on the lesions (since the images we trained our algorithm on were
focused; see Figure 1i). But this zoom-in led to blurred images.
Second, the trained data were relatively small and might not contain all possible types of lesions
and anatomical regions where such lesions can occur (e.g., the eye or close to the eye; see Figure 1).
Note that these aspects can impact the performance of the model, as exemplified in Figure 10, where
the algorithm classified the lesion as “malignant“ (see Figure 10a) when the lower part of the iris
and the sclera is obvious in the zoom-in while classifying the lesion as “benign” (see Figure 10b)
when only the lower part of the sclera is obvious in the zoom-in.

Diagnostics 2025,15, 710 21 of 30
(b)
(a) (i) (ii) (iii)
Binary
Classification
Benign: 0.000
Malignant: 1.000
Benign: 0.0000
Malignant: 1.0000
Benign: 0.0000
Malignant: 1.0000
Multiclass
Classification
Benign: 0.000
Malignant: 1.000
Keloid: 0.0000
Benign: 0.0000
Malignant:
0.0000
Keloid:
1.0000
Benign: 0.0000
Malignant: 0
.9988
Keloid:
0.0012
(i) (ii) (iii)
Binary
Classification
Benign: 0.0000
Malignant: 1.0000
Benign: 0.0000
Malignant: 1.0000
Benign: 0.9478
Malignant: 0.0522
Multiclass
Classification
Benign: 0.0851
Malignant: 0.9149
Keloid: 0.0000
Benign: 0.0000
Malignant: 1.0000
Keloid: 0.0000
Benign: 0.0000
Malignant: 0.0000
Keloid: 1.0000
Figure 9. The result of our proposed binary model and multiclass model on (a) three anonymised
clinical images of skin lesions (denoted by (i–iii)) and (b) their zoom-ins. The first rows of the tables
in (a,b) show their class probabilities when tested with the proposed binary model, while the second
rows of (a,b) show their class probabilities when tested with the proposed multiclass model.
Benign: 0.0005
Malignant: 0.9995
Benign: 0.9972
Malignant: 0.0028
(a)
(b)
Figure 10. Result of the binary algorithm on two different levels of zoom-ins for the anonymised
image in Figure 9a(iii). More precisely, in (a) we show an approx. 40% zoom-in and in (b) an approx
70% zoom-in of the original image in Figure 9a(iii).
3.4. Keloid vs. Similar-Looking Malignant Lesions
As mentioned in the Introduction, previous studies have shown that malignant lesions
such as basal cell carcinoma and squamous cell carcinoma are sometimes misdiagnosed as
keloid [
72
–
74
]. In this section, we trained the algorithms to be able to differentiate between
these skin lesions, and hence, we restricted ourselves only to the basal cell carcinoma
and squamous cell carcinoma images in [
34
], as well as to the keloid dataset from [
44
]
for training, validation, and testing. We returned to all 10 models and trained them
on augmented data as well as fine-tuned them (since this approach produced the best
performance results). The results are presented in Table 21. We see here that DenseNet121
outperformed the rest of the models except in terms of the AUC score, where it is the
fourth-best model. The best AUC score was given by the VGG16 model, followed by

Diagnostics 2025,15, 710 22 of 30
InceptionNetRNV2 and EfficientNetV2L. Moreover, the results in this table show that
VGG16 is in top three algorithms that can differentiate between keloids and other similar-
looking skin lesions.
Table 21. Accuracy, precision, recall, F1
score
, and AUC of the models on the test dataset. The models
were trained on the augmented train dataset and validated on the original validation datasets,
with the base models fine-tuned. In the “Accuracy” column, we show in bold the largest value
indicating the best model.
Model Accuracy Precision Recall F1score AUC
VGG16 0.8647 0.8674 0.8647 0.8378 0.8982
MobileNet 0.8346 0.8045 0.8346 0.8100 0.8727
DenseNet121 0.9098 0.9110 0.9098 0.8972 0.8872
InceptionV3 0.8797 0.8956 0.8797 0.8559 0.8848
EfficientNetB0 0.8421 0.8172 0.8421 0.8133 0.8867
Xception 0.8346 0.8045 0.8346 0.8100 0.8836
InceptionRNV2 0.8647 0.8521 0.8647 0.8447 0.8976
EfficientNetV2L 0.8571 0.8343 0.8571 0.8344 0.8942
NASNetLarge 0.8571 0.8551 0.8571 0.8297 0.8819
Finally, in Figure 11 we show the confusion matrix of the performance of DensNet121,
as well as its AUC-ROC curve and score.
(a) (b)
Figure 11. The (a) confusion matrix and (b) AUC-ROC curve for VGG16 trained on the augmented
train data, validated on the original validation dataset (while fine-tuning the base model), and tested
on the test data.
4. Discussion and Research Limitation
4.1. Discussion
In this study, we trained nine classical base CNN models (VGG16, MobileNet,
DenseNet121, InceptionV3, EfficientNetB0, Xception, InceptionRNV2, EfficientNetV2-L,
and NASNet-L) to perform three classification tasks using two publicly available image
datasets. The first classification task was to train the models to classify keloids, melanomas,
basal cell carcinomas, squamous cell carcinomas, seborrheic keratosis, nevus, and actinic
keratosis as either “malignant“ or “benign” skin lesions. The second task was to train the
models to classify these skin disorders as “malignant“, “benign”, or “keloid“, since we
wished to distinguish the keloids from other benign lesions. The third classification task
was to distinguish keloids from other malignant look-alikes, which sometimes have been
misdiagnosed. For the three classification tasks, we employed transfer learning techniques
(see Figure 3) to compensate for insufficient image data in biomedicine. It is important

Diagnostics 2025,15, 710 23 of 30
to emphasise that the images considered in this study were non-dermatoscopic clinical
images (as given by the PAD dataset [
34
] for benign and malignant lesions and by the
Kaggle dataset [
44
] for keloid lesions), as we aimed to train these models such that they
can be used in communities where dermatoscopes are either rare or unavailable and where
patients might not be able to have appointments with specialist dermatologists. Note that
most of the studies on benign-vs.-malignant classification of skin lesions use the more
popular and much larger ISIC dataset of dermatoscopic images from [
75
,
76
] or HAM10000
data [77].
We trained the models on the original train data, an oversampled training data,
and augmented training data, and we validated and tested them on the original validation
dataset and the test dataset (with this one never being exposed to the models during
training). For model performance, we used the following classical metrics: accuracy,
precision, recall, F1score, AUC-ROC, and confusion matrix.
4.2. Research Limitations
As much as our proposed model performed well on the test data, we admit it had
some limitations when tested on some anonymised clinical images obtained from Dr. Brice
Chatelain, as shown in Figure 10. Some of the limitations of this study, especially those of
the proposed model, have been highlighted in Remark 1. In what follows, we highlight our
perceived limitations of this study, including those already highlighted in Remark 1:
•
The image dataset used in this study contained non-dermatoscopic (clinical) images
of keloids, some malignant skin cancers/lesions, and other benign lesions (including
melanomas, basal cell carcinomas, squamous cell carcinomas, seborrheic keratosis,
nevus, and actinic keratosis); hence, it may not perform well when tested on other skin
lesions not present in training data or dermatoscopic images of skin lesions present in
training data.
•
In addition, as we mentioned in Section 2.1, not all data we used were pathologically
validated (especially the non-cancerous lesions), which could have impacted the
classification results we obtained.
•
In Remark 1, we highlighted that the models performed poorly on images taken at a
long distance from the skin lesion, as the models were trained on images focused on
the skin lesions. Also, zoomed-in images of the same original pictures led to blurred
images and possible misclassification.
•
Lastly, as previously mentioned in Remark 1, the trained data were relatively small
in number and might not contain all possible anatomical regions where such lesions
can occur (e.g., the eye or close to the eye; see Figure 1). Note that these aspects can
impact the performance of the model, as exemplified in Figure 10, where the algorithm
classified the lesion as “malignant” (see Figure 10a) when the lower part of the iris
and the sclera is obvious in the zoom-in, while classifying the lesion as “benign“ (see
Figure 10b) when only the lower part of the sclera is obvious in the zoom-in.
5. Conclusions
The results showed that in first two classification tasks, VGG16 outperformed the
rest of the models after fine-tuning (see Tables 13 and 20), while for the third classification
task, DenseNet121 outperformed the rest of the models. The VGG16 model (i.e., the best
models for the first two classification tasks) was also tested on three new clinical images
(graciously provided by B. Chatelain), which were not available on the public datasets
used for training/validation/testing. We showed that VGG16 performed well on the
binary classification of the original images but not on the classification of the zoom-in
images (where at least one image was misclassified). This is probably the result of a

Diagnostics 2025,15, 710 24 of 30
lack of similar data (from similar anatomical regions) used for training. Hence, more
image data available for training will likely improve the results. Moreover, combining the
clinical features of each skin lesion with the images of these lesions has been reported to
improve the performance of CNN models [
27
]. We will consider such an approach in the
future (knowing that at this moment only the benign/malignant dataset [
34
] also contains
clinical features; the keloid dataset [
44
] does not contain such detailed information). As
mentioned in the Introduction, we also plan to further explore the use of vision transformers
(ViTs), as they show potential in image classification, since they use fewer computational
resources compared to CNN models, even though they do not currently outperform the
CNN models [
39
], especially when trained on small datasets. We also hope to delve into
mechanistic learning approaches, where we can leverage knowledge-driven modelling
with data-driven modelling to improve the interpretability of such models [
78
]. We will
also explore the ablation study of each considered architecture and further investigate
how feature selection/extraction using algorithms such as principal component analysis
(PCA), minimum redundancy maximum relevance (mRMR), autoencoder (AE), linear
discriminant analysis (LDA), etc., could improve model performance and computational
efficiency. We will also explore how the use of Generative Adversarial Network (GAN) for
data augmentation [28] affects the performance of the models.
6. Comparison with Published Literature
As we mentioned in the Introduction, we are not aware of any studies classifying
keloid images among benign and malignant skin lesion images. However, some stud-
ies classify other skin lesions using dermatoscopic
data [79,80]
or non-dermatoscopic
data [27,81]
. To position our study within the published literature, we compared our
results with those in studies that equally classify the skin lesions as either malignant or
benign. In [
79
], a CNN-based model was trained on the publicly available dermoscopic
datasets of skin lesions
[77,82,83]
to classify benign vs malignant skin lesions and returned
an accuracy of 0.854
±
0.032, a precision of 0.936
±
0.017, and a recall of 0.88
±
0.048.
The same study [
79
] also considered a machine learning-based approach and arrived at an
accuracy of
0.738 ±0.011
, a precision of 0.854
±
0.01, and a recall of 0.811
±
0.013. In [
80
],
a deep learning (DL) model was built using a public dermatoscopic dataset from [77] and
a Kaggle dataset from the ISIC archive [
35
], and the researchers obtained an accuracy of
0.88, a precision of 0.93, a recall of 0.83, and an F1 score of 0.88. Our proposed model
for the binary classification based on the VGG16 model (see first row in Table 17) yielded
an accuracy of 0.85, a precision of 0.86, and a recall of 0.77. It is important to emphasise
that the models in [
79
,
80
] were trained on dermatoscopic image datasets (which are more
available publicly, thanks to the ISIC and HAM10000 datasets, which, when combined,
have over 400,000 images of different skin lesions as of 2024). In this study, our focus was
to train a deep learning model on non-dermatoscopic images of skin lesions to provide
ways of diagnosing these skin lesions using just phone camera pictures (clinical images)
in areas where dermatoscopes are not available or are too expensive to consider. To the
best of our knowledge, very few machine learning and deep learning models have been
used to classify skin lesions using non-dermatoscopic images [
27
,
81
], with [
34
] (i.e., the
dataset used in [
27
]) being the only one available publicly. While [
27
] considered similar
datasets, we emphasise that we added a keloid dataset from Kaggle [
44
]. In [
27
], the clinical
images and the clinical features provided in the metadata were combined, while in this
study, we only considered the clinical images, and the addition of the clinical features
will be a focus of future work. The performance of the multiclass classification model
in [
27
], when clinical images were combined with clinical features (i.e., patient information),
yielded an accuracy of 0.788
±
0.025, precision of 0.800
±
0.028, a recall of 0.788
±
0.025,

Diagnostics 2025,15, 710 25 of 30
and an F1 of 0.790
±
0.027; our proposed model for the binary classification (i.e., benign
vs. malignant) yielded an accuracy of 0.851, a precision of 0.860, a recall of 0.775, and an
F1 of 0.815; the multiclass VGG-based model (i.e., benign vs. malignant vs. keloid; see
Table 20), yielded an accuracy of 0.877, a precision of 0.881, a recall of 0.877, and an F1 of
0.878; and finally, the multiclass DenseNet-based model (i.e., keloid vs. basal cell carcinoma
vs. squamous cell carcinoma; see Table 21), yielded an accuracy of 0.910, a precision of 0.911,
a recall of 0.910, and an F1 of 0.897. In Table 22, we present a summary of the comparisons
with previous work.
It is evident that for a model based on less than 3000 non-dermatoscopic images, our
model has better performance, with an
≈
9% increase in performance compared to [
27
], as
seen in Table 22. We recognise that our models were trained for fewer classes (
i.e
., two and
three classes) compared to the six-class models in [
27
], which might explain the difference
in the results.
Table 22. Comparing accuracy, precision, and recall of our proposed models for the three classification
tasks with result of previous works.
Authors Dataset Category Accuracy Precision Recall
Brutti et al.
[79]
Dermatoscopic image
dataset from [77,82,83]
Benign vs.
Malignant 0.854 ±0.032 0.936 ±0.017 0.88 ±0.048
Bechelli and
Delhommelle
[80]
Dermatoscopic image
dataset from [35,77]
Benign vs.
Malignant 0.88 0.93 0.83
Pacheco and
Krohling [27]
Non-dermatoscopic
image dataset from [34]
Multiple (six) skin
lesion class 0.788 ±0.025 0.8 ±0.028 0.79 ±0.027
Udrea et al.
[81]
Non-dermatoscopic
image dataset (private) Unspecified unavailable unavailable 0.951
Benign vs.
Malignant; 0.851 0.860 0.775
Ours
Non-dermatoscopic
image dataset from [34]
and [44]
Benign vs.
Malignant vs.
Keloid;
0.877 0.881 0.877
Keloid vs.
BCC vs. SCC 0.91 0.911 0.91
Author Contributions: Conceptualisation, O.E.A., D.T. and R.E.; software, O.E.A.; writing—original
draft preparation, O.E.A., D.T. and R.E.; writing—review and editing, O.E.A., D.T. and R.E.; resources,
B.C., supervision, R.E. and D.T.; project administration, R.E. and D.T.; funding acquisition, R.E. All
authors have read and agreed to the published version of the manuscript.
Funding: R.E. and O.E.A. acknowledge funding from a French Agence Nationale de Recherche
(ANR) grant number ANR-21-CE45-0025-01. O.E.A also acknowledges the 2024 international mobility
funding (from the University of Franche Comte research commission) for his visit to the University
of Dundee, United Kingdom, where this project was conceptualised.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: The code for this project is available at https://github.com/
OEAdebayo/skin-project.
Acknowledgments: We acknowledge useful discussions with Charlee Nardin and Thomas Lihoreau
from CHRU Jean Minjoz, Besançon.

Diagnostics 2025,15, 710 26 of 30
Conflicts of Interest: The authors declare no conflicts of interest. The funders had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or
in the decision to publish the results.
Appendix A. Details of the Performance Metrics
As explained in Section 2.2.3, we made use of the following performance metrics:
•
Accuracy, which measures the proportion of correct predictions made by the model
defined as
accuracy =TP+TN
TP +FP +TN +FN. (A1)
•
Precision, which measures the ratio of
TP
predictions among all positive predictions
(TP and FP), is given by
precision =TP
TP +FP. (A2)
This metric indicates how many of the predicted positive cases are correctly identified.
•
Recall (also known as sensitivity or TP rate), which measures the proportion of cor-
rectly predicted positive cases (TP) out of all actual positive cases (TP + FN), is given by
recall =TP
TP +FN. (A3)
In other words, it denotes how many of the actual positive cases were correctly
identified by the model. Note that using precision and recall separately might not
provide the full picture, since these metrics are subjective of the classification task at
hand. Thus, it is better to use metrics that combine both precision and recall.
•
A metric that combines the precision and recall is the F1
score
. It provides a good
evaluation of model performance and is defined below as
F1score =2precision ×recall
precision +recall. (A4)
It will not account for imbalance in the dataset.
•
We can also evaluate the performance of each model using the Area Under the Receiver
Operating Characteristic Curve (AUC-ROC). The AUC-ROC measures the ability of
a classifier to differentiate between the positive and negative classes across different
threshold values, while the ROC curve itself is a plot of the FP rate (
x
axis) against the
TP rate (i.e., the recall on the yaxis).
•
The confusion matrix is another (visual) metric that describes the performance of a
classification model by comparing the actual values with the predicted values.
Appendix B. Hardware and Software Specification
We implemented our models on a high-performance computing (HPC) cluster with
compute nodes equipped with dual Intel Xeon Silver 4314 CPUs, where each CPU has
16 cores (totaling 32 cores per node) with varying node RAM capacity and operates at a base
frequency of 2.4 GHz, with a maximum turbo frequency of 3.4 GHz. The software stack
included Ubuntu 20.04 LTS OS, Python 3.12, and Keras 3.3 with a TensorFlow 2.16.1 backend.
The training jobs were submitted via Slurm 19.05.5, and the jobs were assigned based on
node availability. The jobs were run without GPU acceleration, relying entirely on CPU-
based training. The rest of the software dependencies can be found in the requirements.txt
file available in the GitHub link provided in the “Data Availability Statement” above.

Diagnostics 2025,15, 710 27 of 30
References
1.
Behera, D.; Debata, I.; Mohapatra, D.; Agarwal, A. Appearances can be deceptive: An intriguing case of a keloidal mass. Indian J.
Dermatol. 2022,67, 629. [CrossRef] [PubMed]
2.
Grant-Kels, J.M.; Bason, E.T.; Grin, C.M. The misdiagnosis of malignant melanoma. J. Am. Acad. Dermatol. 1999,40, 539–548.
[CrossRef] [PubMed]
3.
Hwang, S.-M.; Pan, H.-C.; Hwang, M.-K.; Kim, M.-W.; Lee, J.-S. Malignant skin tumor misdiagnosed as a benign skin lesion. Arch.
Craniofac. Surg. 2016,17, 86–89. [CrossRef]
4.
Jia, J.; Wang, M.; Song, L.; Feng, Y. A melanotic malignant melanoma presenting as a keloid. A case report. Medicine 2017,
96, e9047. [CrossRef] [PubMed]
5. Nicholas, R.S.; Stodell, M. An important case. BMJ Case Rep. 2014,2014, bcr2014203600. [CrossRef]
6.
Sondermann, W.; Zimmer, L.; Schadendorf, D.; Roesch, A.; Klode, J.; Dissemond, J. Initial misdiagnosis of melanoma located on
the foot is associated with poorer prognosis. Medicine 2016,95, e4332. [CrossRef]
7.
Ucak, M. A rare case of misdiagnosis: Recurrence of dermatofibrosarcoma protuberans that was treated surgically as keloid. Med.
Arch. 2018,72, 74–75. [CrossRef]
8.
Wang, Y.; Yang, F.; Wu, X.L.; Liu, F.; Jin, R.; Gu, C.; Ni, T.; Wang, X.; Yang, Q.; Luo, X.; et al. On the diagnosis and treatment of
cutaneous and soft tissue tumours misdiagnosed as scars: Lessons from four cases. Int. Wound J. 2019,16, 793–799. [CrossRef]
9.
Newcomer J.B.; Durbin A.; Wilson C. Cutaneous Metastasis of Lung Adenocarcinoma: Initially Mimicking and Misdiagnosed as
Keloids. Cureus 2022,14, e27285. [CrossRef]
10.
Niessen, F.B.; Spauwen, P.H.; Schalkwijk, J.; Kon, M. On the nature of hypertrophic scars and keloids: A review. Plast. Reconstr.
Surg. 1999,104, 1435–1458. [CrossRef]
11.
Vincent, A.S.; Phan, T.T.; Mukhopadhyay, A.; Lim, H.Y.; Halliwell, B.; Wong, K.P. Human skin keloid fibroblasts display
bioenergetics of cancer cells. J. Investig. Dermatol. 2008,128, 702–709. [CrossRef] [PubMed]
12.
Gauglitz, G.G.; Korting, H.C.; Pavicic, T.; Ruzicka, T.; Jeschke, M.G. Hypertrophic scarring and keloids: Pathomechanisms and
current and emerging treatment strategies. Mol. Med. 2011,17, 113–125. [CrossRef] [PubMed]
13.
Satish, L.; Lyons-Weiler, J.; Hebda, P.A.; Wells, A. Gene expression patterns in isolated keloid fibroblasts. Wound Repair Regen.
2006,14, 463–470. [CrossRef]
14.
Eftimie, R.; Rollin, G.; Adebayo, O.; Urcun, S.; Bordas, S.P.A. Modelling keloids dynamics: A brief review and new mathematical
perspectives. Bull. Math. Biol. 2023,85, 117. [CrossRef]
15.
Bostanci, S.; Akay, B.N.; Alizada, M.; Okçu, A.H. Cutaneous leiomyosarcoma misdiagnosed as a keloid: A case report with
dermatoscopy. Turkderm.-Turk. Arch. Dermatol. Venereol. 2023,57, 154–156. [CrossRef]
16.
Portugal, E.H.; Alves, J.C.R.R.; Da Fonseca, R.P.L.; De Souza Andrade, J.; De Melo Alemida, A.C.; Araujo, I.C.; Pereira, N.A.; Da
Silva, R.L.F. Dermatofibrosarcoma protuberans misdiagnosed as keloid and treated with triamcinolone acetonid. Rev. Bras. Cir.
Plast. 2016,31, 82–87. [CrossRef]
17.
Tiong, W.; Basiron, N. Challenging Diagnosis of a Rare Case of Spontaneous Keloid Scar. J. Med. Cases 2014,5, 466–469. [CrossRef]
18.
Lallas, A.; Paschou, E.; Manoli, S.M.; Papageorgiou, C.; Spyridis, I.; Liopyris, K.; Bobos, M.; Moutsoudis, A.; Lazaridou, E.; Apalla,
Z. Dermatoscopy of melanoma according to type, anatomic site and stage. Ital. J. Dermatol. Venereol. 2021,156, 274–288. [CrossRef]
19. Puig, S.; Cecilia, N.; Malvehy, J. Dermoscopic criteria and basal cell carcinoma. G. Ital. Dermatol. Venereol. 2012,147, 135–140.
20.
Sgouros, D.; Theofili, M.; Zafeiropoulou, T.; Lallas, A.; Apalla, Z.; Zaras, A.; Liopyris, K.; Pappa, G.; Polychronaki, E.; Kousta,
F.; et al. Dermoscopy of Actinic Keratosis: Is There a True Differentiation between Non-Pigmented and Pigmented Lesions? J.
Clin. Med. 2023,12, 1063. [CrossRef]
21.
Fee, J.A.; McGrady, F.P.; Hart, N.D. Dermoscopy use in primary care: A qualitative study with general practitioners. BMC Primary
Care 2022,23, 47. [CrossRef] [PubMed]
22.
Forsea, A.M.; Tschandl, P.; Zalaudek, I.; Del Marmol, V.; Soyer, H.P.; Eurodermoscopy Working Group; Argenziano, G.; Geller,
A.C. The impact of dermoscopy on melanoma detection in the practice of dermatologists in Europe: Results of a pan-European
survey. J. Eur. Acad. Dermatol. Venereol. 2017,31, 1148–1156. [CrossRef] [PubMed]
23.
Skvara, H.; Teban, L.; Fiebiger, M.; Binder, M.; Kittler, H. Limitations of dermoscopy in the recognition of melanoma. Arch.
Dermatol. 2005,141, 155–160. [CrossRef] [PubMed]
24. Chollet, F. Deep Learning with Python; Manning Publications: Shelter Island, NY, USA, 2018.
25. Lecun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015,521, 436. [CrossRef] [PubMed]
26.
Jo, T.; Nho, K.; Saykin, A. Deep Learning in Alzheimer’s Disease: Diagnostic Classification and Prognostic Prediction Using
Neuroimaging Data. Front. Aging Neurosci. 2019,11, 220. [CrossRef]
27.
Pacheco, A.G.C.; Krohling, R.A. The impact of patient clinical information on automated skin cancer detection. Comput. Biol. Med.
2020,116, 103545. [CrossRef]
28.
Frid-Adar, M.; Diamant, I.; Klang, E.; Amitai, M.; Goldberger, J.; Greenspan, H. GAN-based synthetic medical image augmentation
for increased CNN performance in liver lesion classification. Neurocomputing 2018,321, 321–331. [CrossRef]

Diagnostics 2025,15, 710 28 of 30
29.
Thirumalaisamy, S.; Thangavilou, K.; Rajadurai, H.; Saidani, O.; Alturki, N.; Mathivanan, S.K.; Jayagopal, P.; Gochhait, S. Breast
Cancer Classification Using Synthesized Deep Learning Model with Metaheuristic Optimization Algorithm. Diagnostics 2023,
13, 2925. [CrossRef]
30.
Houfani, D.; Slatnia, S.; Remadna, I.; Zerhouni, N.; Saouli, H. Breast cancer classification using machine learning techniques: A
comparative study. Med. Technol. J. 2020,4, 535–544. [CrossRef]
31.
Ghosh, P.; Azam, S.; Quadir, R.; Karim, A.; Shamrat, F.M.J.M.; Bhowmik, S.K.; Jonkman, M.; Hasib, K.M.; Ahmed, K. SkinNet-16:
A deep learning approach to identify benign and malignant skin lesions. Front. Oncol. 2022,12, 535–544. [CrossRef]
32.
Kim, J.; Oh, I.; Lee, Y.N.; Kim, J.; Lee, H.J. Predicting the severity of postoperative scars using artificial intelligence based on
images and clinical data. Sci. Rep. 2023,13, 13448. [CrossRef]
33.
Li, S.; Wang, H.; Xiao, Y.; Zhang, M.; Yu, N.; Zeng, A.; Wang, X. A Workflow for Computer-Aided Evaluation of Keloid Based on
Laser Speckle Contrast Imaging and Deep Learning. J. Pers. Med. 2022,12, 981. [CrossRef] [PubMed]
34.
Pacheco, A.G.C.; Lima, G.R.; Salomão, A.S.; Krohling, B.; Biral, I.P.; Angelo, G.G.; Alves, F.C.R.; Esgario, J.G.M.; Simora, A.C.;
Castro, P.B.C.; et al. PAD-UFES-20: A skin lesion dataset composed of patient data and clinical images collected from smartphones.
Mendeley Data 2020,V1. [CrossRef] [PubMed]
35.
Kaggle. Skin Cancer: Malignant vs. Benign. Available online: https://www.kaggle.com/datasets/fanconic/skin-
cancermalignant-vs-benign (accessed on 1 November 2021).
36.
Skin Cancer MNIST: HAM10000. Available online: https://www.kaggle.com/datasets/kmader/skin-cancer-mnist-ham10000
?fbclid=IwAR0ZHT4XR1ZBC8fHdPtVkQRlOF0XQIyJNSiG1rohE4rYb7K8vkKx9eVpSjg (accessed on 5 June 2022).
37.
Lu, J. Convolutional neural network and vision transformer for image classification. Appl. Comput. Eng. 2023,5, 104–108.
[CrossRef]
38.
Khan, S.; Naseer, M.; Hayat, M.; Zamir, S.W.; Khan, F.S.; Shah, M. Transformers in Vision: A Survey. Assoc. Comput. Mach. 2022,
54, 10s. [CrossRef]
39.
Steiner, A.; Kolesnikov, A.; Zhai, X.; Wightman, R.; Uszkoreit, J.; Beyer, L. How to train your ViT? Data, Augmentation, and
Regularization in Vision Transformers. arXiv 2022, arXiv:2106.10270.
40.
Acar, E.M.; Kilitci, A.; Kaya, Z.; Abbas, O.L.; Kemeriz, F. Basal cell carcinoma presenting as an excoriated cicatricial plaque: A
case report. Dermatol. Sin. 2018,36, 211–213. [CrossRef]
41. Lewis, J. Keloidal basal cell carcinoma. Am. J. Dermatopathol. 2007,29, 485. [CrossRef]
42.
Mizuno, H.; Uysal, A.C.; Koike, S.; Hyakusoku, H. Squamous cell carcinoma of the auricle arising from keloid after radium
needle therapy. J. Craniofac. Surg. 2006,17, 360–362. [CrossRef]
43.
Gutman, D.; Codella, N.C.F.; Celebi, E.; Helba, B.; Marchetti, M.; Mishra, N.; Halpern, A. Skin lesion analysis toward melanoma
detection: A challenge at the International Symposium on Biomedical Imaging (ISBI) 2016, hosted by the International Skin
Imaging Collaboration (ISIC). arXiv 2016, arXiv:1605.01397.
44.
Kaggle. Keloid Scars vs Hypertrophic Scars. Available online: https://www.kaggle.com/datasets/quratulainislam/keloid-vs-
scar (accessed on 16 February 2024).
45.
Cimpean, I.; Theate, I.; Vanhooteghem, O. Seborrheic keratosis evolution into squamous cell carcinoma: A truly modified
sun-related tumor? A case report and review of the literature. Dermatol. Rep. 2019,11, 7999. [CrossRef]
46. Mentzel, T.; Wiesner, T.; Cerroni, L.; Hantschke, M.; Kutzner, H.; Rütten, A.; Häberle, M.; Bisceglia, M.; Chibon, F.; Coindre, J.M.
Malignant dermatofibroma: Clinicopathological, immunohistochemical, and molecular analysis of seven cases. Mod. Pathol. 2013,
26, 256–267. [CrossRef] [PubMed]
47.
Simonyan, K.; Zisserman, A. Very Deep Convolutional Networks for Large-Scale Image Recognition. In Proceedings of the
International Conference on Learning Representations, San Diego, CA, USA, 7–9 May 2015.
48.
Szegedy, C.; Vanhoucke, V.; Ioffe, S.; Shlens, J.; Wojna, Z. Rethinking the Inception Architecture for Computer Vision. In
Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 27–30 June
2016; pp. 2818–2826.
49.
Huang, G.; Liu, Z.; Van Der Maaten, L.; Weinberger, K.Q. Densely Connected Convolutional Networks. In Proceedings of the
2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21–26 July 2017; pp. 2261–2269.
50.
Howard, A.G.; Zhu, M.; Chen, B.; Kalenichenko, D.; Wang, W.; Weyand, T.; Andreetto, M.; Adam, H. MobileNets: Efficient
Convolutional Neural Networks for Mobile Vision Applications. arXiv 2017, arXiv:1704.04861.
51. Tan, M.; Le, Q.V. EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. arXiv 2020, arXiv:1905.11946.
52. Chollet, F. Xception: Deep Learning with Depthwise Separable Convolutions. arXiv 2017, arXiv:1610.02357.
53.
Szegedy, C.; Ioffe, S.; Vanhoucke, V.; Alemi, A. Inception-v4, Inception-ResNet and the Impact of Residual Connections on
Learning. arXiv 2016, arXiv:1602.07261. [CrossRef]
54. Tan, M.; Le, Q.V. EfficientNetV2: Smaller Models and Faster Training. arXiv 2021, arXiv:2104.00298.
55.
Zoph, B.; Vasudevan, V.; Shlens, J.; Le, Q.V. Learning Transferable Architectures for Scalable Image Recognition. arXiv 2018,
arXiv:1707.07012.

Diagnostics 2025,15, 710 29 of 30
56.
Tejada, M.; Espinoza, R.; Hinojosa, L. Use of Xception Architecture for the Classification of Skin Lesions. Int. J. Comput. Vis. 2024,
12, 145–160.
57.
Kurniawan, V.A.T.; Niswary, E.C.; Chandranegara, D.R. Brain Tumor Classification Using InceptionResNetV2 and Transfer
Learning. J. Comput. Biol. Med. 2024,18, 98–110.
58.
Liu, D.; Wang, W.; Wu, X.; Yang, J. EfficientNetV2 Model for Breast Cancer Histopathological Image Classification. IEEE Int. Conf.
Electr. Eng. Comput. Sci. 2022,7, 384–387.
59.
Rahman, M.A.; Bazgir, E.; Hossain, S.M.S.; Maniruzzaman, M. Skin Cancer Classification Using NASNet. Int. J. Sci. Res. Arch.
2024,11, 23–36.
60.
Ismail, M.; Nisa, F.; Ahmad, R. Utilising VGG16 of Convolutional Neural Network for Brain Tumor and Alzheimer ’s Classification.
J. Adv. Med. Sci. 2024,10, 123–135.
61.
Costa, M.; Sousa, J.; Almeida, P. Classification of X-ray Images for Detection of Childhood Pneumonia using MobileNet. IEEE
Trans. Med. Imaging 2020,39, 1123–1135.
62.
Mulya, R.F.; Utami, E.; Ariatmanto, D. Classification of Acute Lymphoblastic Leukemia based on White Blood Cell Images using
InceptionV3 Model. J. RESTI (Rekayasa Sist. Dan Teknol. Inf.) 2023,7, 5182. [CrossRef]
63.
Zerouaoui, H.; Idri, A. Classifying Breast Cytological Images using Deep Learning Architectures. Proc. Int. Conf. Data Sci. Artif.
Intell. 2022, 557–564. [CrossRef]
64.
Sait, N.A.; Kathirvelan, J. Detection and classification of progressive supranuclear palsy from MRI images using deep learning. J.
Appl. Res. Technol. 2024,22, 2228. [CrossRef]
65.
Feller, L.; Khammissa, R.A.G.; Kramer, B.; Altini, M.; Lemmer, J. Basal cell carcinoma, squamous cell carcinoma and melanoma of
the head and face. Head Face Med. 2016,12, 11. [CrossRef]
66.
Udri
s
,
toiu, A.L.; Stanca, A.E.; Ghenea, A.E.; Vasile, C.M.; Popescu, M.; Udri
s
,
toiu,
S
,
.C.; Iacob, A.V.; Castravete, S.; Gruionu, L.G.;
Gruionu, G. Skin diseases classification using deep learning methods. Curr. Health Sci. J. 2020,46, 136–140.
67.
Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet classification with deep convolutional neural networks. Commun. ACM 2017,
60, 84–90. [CrossRef]
68.
Shin, H.C.; Roth, H.R.; Gao, M.; Lu, L. Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures,
Dataset Characteristics and Transfer Learning. IEEE Trans. Med. Imaging 2016,35, 1285–1298. [CrossRef]
69.
Kumar, A.; Kim, J.; Lyndon, D.; Fulham, M.; Feng, D. An Ensemble of Fine-Tuned Convolutional Neural Networks for Medical
Image Classification. IEEE J. Biomed. Health Inform. 2017,21, 31–40. [CrossRef] [PubMed]
70.
Ioffe, S.; Szegedy, C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. arXiv 2015,
arXiv:1502.03167.
71.
Gupta, S.; Tan, M. EfficientNet-EdgeTPU: Creating Accelerator-Optimized Neural Networks with AutoML. 2019. Available online:
https://research.google/blog/efficientnet-edgetpu-creating-accelerator-optimized-neural- networks-with- automl/ (accessed on
7 January 2025).
72.
Goder, M.; Kornhaber, R.; Bordoni, D.; Winkler, E.; Haik, J.; Tessone, A. Cutaneous basal cell carcinoma arising within a keloid
scar: A case report. OncoTargets Ther. 2016,9, 4793–4796.
73.
Majumder, A.; Srivastava, S.; Ranjan, P. Squamous cell carcinoma arising in a keloid scar. Med. J. Armed Forces India 2019,75,
222–224. [CrossRef]
74.
Ogawa, R.; Yoshitatsu, S.; Yoshida, K.; Miyashita, T. Is radiation therapy for keloids acceptable? The risk of radiation-induced
carcinogenesis. Plast. Reconstr. Surg. 2009,124, 1196–1201. [CrossRef]
75.
Kaggle. Skin Cancer ISIC. Available online: https://www.kaggle.com/datasets/nodoubttome/skin-cancer9-classesisic/data
(accessed on 8 February 2024).
76.
International Skin Imaging Collaboration (ISIC). ISIC Gallery. Available online: https://gallery.isic-archive.com (accessed on 25
October 2024).
77.
Tschandl, P.; Rosendahl, C.; Kittler, H. The HAM10000 dataset, a large collection of multi-source dermoscopic images of common
pigmented skin lesions. Sci. Data 2018,5, 180161. [CrossRef]
78.
Metzcar, J.; Jutzeler, C.R.; Macklin, P.; Köhn-Luque, A.; Brüningk, S.C. A review of mechanistic learning in mathematical oncology.
Front. Immunol. 2024,15, 1363144. [CrossRef]
79.
Brutti, F.; La Rosa, F.; Lazzeri, L.; Benvenuti, C.; Bagnoni, G.; Massi, D.; Laurino, M. Artificial Intelligence Algorithms for Benign
vs. Malignant Dermoscopic Skin Lesion Image Classification. Bioengineering 2023,10, 11322. [CrossRef]
80.
Bechelli, S.; Delhommelle, J. Machine learning and deep learning algorithms for skin cancer classification from dermoscopic
images. Bioengineering 2022,9, 97. [CrossRef]
81.
Udrea, A.; Mitra, G.; Costea, D.; Noels, E.; Wakkee, M.; Siegel, D.M.; de Carvalho, T.M.; Nijsten, T. Accuracy of a smartphone
application for triage of skin lesions based on machine learning algorithms. J. Eur. Acad. Dermatol. Venereol. 2020,34, 648–655.
[CrossRef]

Diagnostics 2025,15, 710 30 of 30
82.
Codella, N.; Rotemberg, V.; Tschandl, P.; Celebi, M.E.; Dusza, S.; Gutman, D.; Helba, B.; Kalloo, A.; Liopyris, K.; Marchetti, M.; et al.
Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (ISIC).
arXiv 2019, arXiv:1902.03368.
83.
Codella, N.C.; Gutman, D.; Celebi, M.E.; Helba, B.; Marchetti, M.A.; Dusza, S.W.; Kalloo, A.; Liopyris, K.; Mishra, N.; Kittler,
H.; et al. Skin lesion analysis toward melanoma detection: A challenge at the 2017 international symposium on biomedical
imaging (ISBI), hosted by the international skin imaging collaboration (ISIC). In Proceedings of the 2018 IEEE 15th International
Symposium on Biomedical Imaging (ISBI 2018), Washington, DC, USA, 4–7 April 2018; pp. 168–172.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.