Content uploaded by Subek Sharma
Author content
All content in this area was uploaded by Subek Sharma on Jan 19, 2024
Content may be subject to copyright.
COVER PAGE
TRIBHUVAN UNIVERSITY
INSTITUTE OF ENGINEERING
PASHCHIMANCHAL CAMPUS
LAMACHAUR, POKHARA
A PROJECT PROPOSAL ON
“Anemia detection in pregnant women in Nepal using U-Net.”
SUBMITTED BY:
Kamlesh Ranabhat
Indra Prasad Paneru
Subek Sharma
(PAS076BCT015)
(PAS076BCT014)
(PAS076BCT043)
TABLE OF CONTENTS
TABLE OF CONTENTS.......................................................................................................... i
LIST OF FIGURES................................................................................................................. ii
LIST OF ABBREVIATIONS................................................................................................. iii
CHAPTER 1: INTRODUCTION........................................................................................... 1
1.1 Background.................................................................................................................... 1
1.2 Motivation...................................................................................................................... 2
1.3 Problem Statement......................................................................................................... 2
1.4 Objective........................................................................................................................ 3
1.5 Scope of Project............................................................................................................. 3
1.6 Feasibility Analysis........................................................................................................ 3
1.6.1 Technical Feasibility..............................................................................................4
1.6.2 Operational Feasibility.......................................................................................... 4
1.6.3 Economical Feasibility..........................................................................................4
CHAPTER 2: LITERATURE REVIEW............................................................................... 5
CHAPTER 3: METHODOLOGY..........................................................................................9
3.1 Tools and Technologies Used.........................................................................................9
3.2 Data Collection...............................................................................................................9
3.3 Data Preprocessing.......................................................................................................10
3.4 Data Augmentation...................................................................................................... 10
3.5 Train Test Split............................................................................................................. 10
3.6 Encoder Pathway..........................................................................................................10
3.7 Down-sampling (Max Pooling)....................................................................................10
3.8 Decoder Pathway..........................................................................................................11
3.9 Up-sampling (Up-convolution).................................................................................... 11
3.10 Output Layer...............................................................................................................11
3.11 Loss Function and Optimization.................................................................................11
3.12 Inference and Evaluation............................................................................................11
3.13 Model Evaluation....................................................................................................... 12
CHAPTER 4: DEVELOPMENT AND SOFTWARE REQUIREMENT.........................15
4.1 Development environment........................................................................................... 15
4.2 Programming Language, libraries, and packages.........................................................15
CHAPTER 5: EPILOGUE....................................................................................................16
5.1 Expected output............................................................................................................16
5.2 Budget Analysis........................................................................................................... 16
5.3 Scheduling....................................................................................................................16
CHAPTER 6: REFERENCES.............................................................................................. 17
i
LIST OF ABBREVIATIONS
AI - Artificial Intelligence
AUC - Area Under the Curve
CIELAB - CIE 1976 (L*, a*, b*) Color Space
CLAHE - Contrast-Limited Adaptive Histogram Equalization
CNN - Convolutional Neural Network
GLUDA - Global and Local View Unifying for Detection of Anemia
iNAP - integrated Network Access Point
IoMT - Internet of Medical Things
IoU - Intersection over Union
KNN - k-Nearest Neighbors
LoA - Limits of Agreement
MATLAB - MATrix LABoratory
ML - Machine Learning
MLR - Multiple Linear Regression
PLSR - Partial Least Squares Regression
RBC - Red Blood Cell
RoI - Region of Interest
SGD - Stochastic Gradient Descent
SVM - Support Vector Machine
SVR - Support Vector Regression
iii
CHAPTER 1: INTRODUCTION
1.1 Background
Anemia is a widespread global health issue characterized by a decrease in the
concentration of hemoglobin in the blood, which plays a vital role in carrying oxygen
from the lungs to various tissues and organs in the body. This condition disrupts the
body's ability to deliver oxygen effectively, leading to a range of adverse health
effects. Anemia can affect individuals of all ages and demographics, from children to
the elderly, and can arise due to various factors such as nutritional deficiencies,
chronic diseases, or genetic disorders.
Among different population groups, pregnant women represent a particularly
vulnerable group when it comes to anemia. During pregnancy, a woman's
physiological demands increase substantially to support both her own health and that
of the developing fetus. The body's need for additional oxygen and nutrients places
significant strain on the production of red blood cells, leading to an increased risk of
anemia.
Classification of Anemia is based on its severity, which helps in guiding appropriate
treatment and management strategies:
Mild Anemia: Hemoglobin levels are slightly below the normal range (10.0-10.9
g/dl). Individuals with mild anemia may not exhibit significant symptoms, but early
detection and intervention are crucial to prevent progression to more severe stages.
Moderate Anemia: Hemoglobin levels are moderately decreased (7.0-9.9 g/dl).
Individuals with moderate anemia may experience symptoms such as fatigue,
weakness, and pallor, indicating the need for medical attention and appropriate
management.
Severe Anemia: Hemoglobin levels are significantly below the normal range (< 7
g/dl). Severe anemia can lead to pronounced symptoms, including extreme fatigue,
shortness of breath, dizziness, and increased vulnerability to infections.
The implications of anemia during pregnancy are far-reaching. Anemia can lead to
complications such as premature delivery, low birth weight, and increased maternal
morbidity and mortality. Hence, early detection and classification of anemia in
pregnant women are of paramount importance. This facilitates timely interventions,
1
including iron and folic acid supplementation, as well as dietary adjustments, to
ensure optimal outcomes for both the mother and the developing fetus.
1.2 Motivation
Current anemia detection methods in Nepal rely on invasive laboratory techniques,
which are costly, time-consuming, and stressful for patients, especially those with
chronic anemia. Moreover, these methods may expose healthcare professionals to
potential blood-transmitted diseases. In resource-constrained regions, the immobility
of standard clinical methods further inconveniences patients who must travel long
distances to access laboratories.
The pursuit of non-invasive alternatives for anemia diagnosis, such as qualitative
correlations between anemia and subjective assessments of pallor in various regions,
highlights the necessity for better alternatives in low-income areas. The development
of a mobile application for anemia detection holds the promise of transforming
healthcare access, allowing pregnant women across diverse regions of Nepal to
undergo non-invasive, accessible, and timely screening.
By implementing this integrated approach, we aim to revolutionize anemia detection
during pregnancy, ensuring early identification of at-risk individuals and providing
timely interventions to safeguard the health and well-being of mothers and their
unborn children. Through this project, we aspire to make a significant positive impact
on maternal and fetal health outcomes in Nepal, reducing the burden of
anemia-related complications and enhancing overall healthcare accessibility and
equity.
1.3 Problem Statement
Anemia in pregnant women is a significant health challenge in Nepal, impacting
maternal and fetal well-being. The lack of an efficient and systematic approach for
early detection and classification hinders timely interventions and appropriate
management. A comprehensive strategy is needed to address this issue and improve
maternal and fetal health outcomes through modern technologies and
community-based initiatives. Empowering healthcare providers and pregnant women
with tools and knowledge will facilitate prompt identification and tailored
2
management strategies. Bridging these gaps will reduce the burden of anemia-related
complications and contribute to better health for expectant mothers and their unborn
children across diverse regions of Nepal.
1.4 Objective
1. To use a U-Net architecture-based deep learning model that can effectively
detect anemia in pregnant women using images of finger nails, conjunctiva and
palm.
2. Categorization of participant to anemic, polycythemia and healthy categories
1.5 Scope of Project
The scope of this project includes:
1. Collection and curation of a diverse dataset of images of finger nails,
conjunctiva and palm labeled with different stages of anemia.
2. Implementation and training of the U-Net architecture for image segmentation
and classification tasks.
3. Evaluation of the model's performance using appropriate metrics to assess its
accuracy and effectiveness.
4. Optimization of the model by fine-tuning its hyperparameters and architecture.
5. Development of a user-friendly interface to allow general publics for easily
accessible anemia detection via a mobile application.
By successfully completing this project, we aim to provide healthcare professionals
with a reliable and efficient tool for early detection and classification of anemia in
pregnant women, contributing to improved patient health care.
1.6 Feasibility Analysis
After studying and analyzing the required functionalities of the systems, the next task
is to do a feasibility study for the project. It is said that “All Projects are feasible given
unlimited resources and times”. However, both resources and times are limited in
reality. The project should adhere to the time and make efficient use of the available
resources. The proposed solution should satisfy all the user requirements and should
be flexible enough to allow for future changes based on new requirements.
3
The following areas are covered by the feasibility study: -
1. Technical Feasibility
2. Operational Feasibility
3. Economical Feasibility
1.6.1 Technical Feasibility
The development of the project is possible as most of the tools used for the
development of this project are open source and easily available. The team members
have a proper understanding of the programming languages used in this project. Also,
the libraries and frameworks used in this project are available on the internet. This has
made the technical aspect of the project quite smooth.
1.6.2 Operational Feasibility
The user interface will be designed to be user-friendly, ensuring that users can easily
access the system. The system will utilize an iterative process to continuously
enhance the model's predictive quality with each iteration. As a result, the product
will pioneer the field of early stage diagnosis and classification of anemia in pregnant
women.
1.6.3 Economical Feasibility
The various management tools that will be used in the project include tools like
version management tools (e.g., git). They work well with their basic free version
used but we may have to use a paid version of Google Collaboratory to use GPU for
model training, so there will not be any complexity in using these tools for the
development of our project. For the deployment and hosting Microsoft Azure will be
used.`
4
CHAPTER 2: LITERATURE REVIEW
The rapid advancement of artificial intelligence (AI) in the field of healthcare has
shown great promise in improving diagnostic accuracy and patient outcomes. Within
this context, one critical area of focus is the detection of anemia in pregnant women
using machine learning. Anemia during pregnancy poses significant health risks to
both the expectant mother and the developing fetus, making easy,early and accurate
detection crucial for timely interventions. By harnessing the power of AI and ML
algorithms, researchers and healthcare professionals aim to enhance the efficiency and
precision of anemia diagnosis, leading to improved prenatal care and better overall
maternal and neonatal health. In the last few years, AI applications within the domain
of health and diagnosis, with a specific focus on ML-based approaches, have been
developed.
Ghosal et al. (2021) propose a novel hybrid approach, named iNAP, which leverages
IoMT technology to address this crucial healthcare challenge. The iNAP model
utilizes eye and fingernail images, with a specific focus on the regions of interest
(RoI): conjunctiva and fingernails. By employing color spectroscopy analysis on these
extracted portions, the dominant color is extracted, enabling the accurate prediction of
blood hemoglobin levels. A notable feature of the iNAP approach is its
smartphone-based implementation, offering a convenient and accessible solution for
users that employs an adaptive K-means clustering algorithm, further enhancing the
accuracy of the detection process. According to the evaluation results, the iNAP
model demonstrates a remarkable accuracy of +- 0.33 g/dl and a sensitivity of 90%
for 99 patients, indicating its potential as a reliable non-invasive tool for anemia and
polycythemia diagnosis in the IoMT era.
Asare et al. (2023) proposed a mobile application based model for anemia detection
using conjunctiva images of eyes. The model implies YOLO v5 algorithm for RoI
with which accuracy, sensitivity and specificity were 92.5%, 90% and 95%
respectively. In comparison to SVM 78.90%, Decision Tree 73.91% and CNN
77.58%, YOLO v5 algorithm had a better accuracy for anemia detection using
conjunctiva images of eyes.
5
Sevani et al. (2018) present a method for anemia detection based on conjunctiva
images, leveraging the k-means algorithm. The approach involves extracting
dominant color clusters from the conjunctiva images and comparing them to existing
data in the database to draw conclusions about the detection result. The model
achieves an accuracy of 90% in its evaluations. However, a major limitation lies in the
small dataset size, comprising only 10 testing data samples.
Asare et al. (2023) employ various machine learning models, including CNN
(AlexNet) with ReLU activation, K-Nearest Neighbors (k=2), Naive Bayes, Support
Vector Machine (cost=100, epsilon=1.10), and Decision Tree (tree=100), for the
detection of iron deficiency anemia. The study utilizes palm images, which are subject
to image augmentation using CIELAB space operations before being processed
through each of the algorithms. Remarkably high accuracies are achieved, with the
CNN, KNN, Naive Bayes, SVM, and Decision Tree models yielding 99.92%,
99.92%, 99.96%, 96.34%, and 99.29% accuracy, respectively. The results highlight
the efficiency of these machine learning approaches in detecting iron deficiency
anemia, with the SVM model displaying the highest overall accuracy.
Alyosify et al. (2023) develop a smartphone-based method for estimating clinical
signs of anemia and noninvasive hemoglobin level measurement. The study utilizes
images of finger nail, palm, and conjunctiva captured by the mobile phone. For
estimation, the authors employ MLR (Multiple Linear Regression), PLSR (Partial
Least Squares Regression), and SVR (Support Vector Regression) algorithms. The
proposed approach demonstrates a notable accuracy, with a 95% limit of agreement
(LOA) within +-1 g/dl, showcasing its potential as a reliable and noninvasive tool for
assessing anemia and estimating hemoglobin levels in a mobile phone environment.
Dimauro et al. (2019) focus on non-invasive anemia detection by automatically
segmenting crucial sections of the conjunctiva. The researchers utilized the OpenCV
library and employed the a*LAB (a channel extraction) method, followed by image
blurring to reduce noise. For further processing, they applied the OTSU algorithm to
create a binary image and employed the Suzuki algorithm for edge extraction. The
6
study centered on image segmentation, with the aim of identifying relevant sections of
the conjunctiva that could aid in the detection of anemia.
El-kenawy et al. (2023) employed the CLAHE algorithm for image enhancement. The
study utilized an attention-based Unet architecture for the detection and segmentation
task. Comparing the results, the proposed model achieved an impressive accuracy of
98% with a small standard deviation of 1.18, outperforming the traditional Unet
model (95.1% +- 3.6) and the Unet++ model (95.93% +- 2.63). The attention-based
Unet demonstrated its potential in accurately identifying and segmenting COVID-19
infections in medical images.
Appiahene et al. (2023) investigated the detection of iron deficiency anemia using
medical images with machine learning algorithms. The researchers employed Naive
Bayes, Support Vector Machine (SVM), and Convolutional Neural Network (CNN)
for this task. The results indicated impressive model accuracies, with Naive Bayes
achieving 99.06%, SVM 96.34%, and CNN 99.92%.
Zheng et al. (2021) introduced a novel algorithm framework named GLUDA (Global
and Local View Unifying for Detection of Anemia) to enhance intelligent anemia
detection using conjunctiva images. The researchers utilized the efficientNet
architecture for their two-branch neural networks. The GLUDA approach achieved
promising results, with accuracy, sensitivity, and specificity values of 83.13%,
87.23%, and 81.42%, respectively. By combining global and local views of
conjunctiva images, the proposed method demonstrated potential in improving the
accuracy of anemia detection, providing valuable insights for the development of
advanced diagnostic tools in the field of healthcare.
Peksi et al. (2021) focused on the classification of anemia using digital images of
nails and palms with the Naive Bayes method. The study involved converting the
images into the YCbCr color space, which facilitated segmentation and extraction of
color features. The color features were then classified using the Naive Bayes
classifier. The proposed approach achieved an accuracy of 92.3%, demonstrating the
potential of using digital images of nails and palms, along with the Naive Bayes
method, as a promising tool for non-invasive anemia classification
7
Suner et al. (2021) conducted a prospective convenience sample study in emergency
department patients of academic teaching hospitals, where the team developed a
method to predict anemia and estimate hemoglobin concentration using a smartphone
camera. They collected images from 142 patients in phase 1 to train the model and
then tested and validated it on images from 202 other patients in phase 2. The model
achieved accuracy, sensitivity, and specificity rates of 82.9%, 90.7%, and 73.3%,
respectively. To process the images, they utilized a RAW image processing algorithm
in MATLAB.
Asare et al. (2023) proposed a smartphone application approach for the detection of
anemia using conjunctiva images. The application was developed using React Native.
It utilized the smartphone's selfie camera to capture conjunctiva images, which were
then sent to a fast server for processing. The server extracted the Region of Interest
(RoI) and used CNN, Logistic Regression, Classification, and Gaussian Blur
algorithms to predict the Red Blood Cell (RBC) count and Hemoglobin levels. Based
on the predictions, patients were categorized as either anemic or non-anemic, offering
a convenient and accessible tool for anemia detection through smartphone technology.
8
CHAPTER 3: METHODOLOGY
The U-Net architecture has proven to be highly effective in image segmentation tasks,
including the detection and classification of anemia in medical images. By leveraging
its unique U-shaped design and skip connections, U-Net can learn both contextual and
spatial information, making it a valuable tool for accurate and precise medical image
analysis.
3.1 Tools and Technologies Used
We have used the following libraries, frameworks, and development environment for
developing our system:
I. Jupyter Notebook
II. Tensorflow
III. Keras
IV. Scikit-Learn
V. Numpy
VI. Pandas
VII. Matplotlib
VIII. Seaborn
IX. Django
X. Google Collab
3.2 Data Collection
We will use a data collection system to collect the datasets manually visiting different
hospitals and medical colleges, and use their information to survey the patients to take the
required data which includes images of finger nails, conjunctiva and palm. We also release a
form to gather patient information(age, gender, region) for analysis purposes. We require
about 1000 data from different patients for this project.
If possible we will propose a collaboration with the Ministry of Health and Population
(MoHP) for assisting us with the dataset collection.
9
3.3 Data Preprocessing
Data preprocessing is a vital step in preparing images of finger nails, conjunctiva and
palm for the Anemia detection in pregnant women in Nepal using the U-Net. It
includes techniques such as enhancing image quality, standardizing intensity values,
resizing images to a uniform resolution and filters in the dataset. These preprocessing
steps ensure consistent input, enhance model performance, and increase the diversity
of the dataset.
3.4 Data Augmentation
To improve the model's generalization and robustness, data augmentation techniques
can be applied to the training dataset. Common augmentations include rotations, flips,
scaling, and brightness adjustments, effectively increasing the diversity of the training
set.
3.5 Train Test Split
The dataset is divided into training, validation, and testing subsets. The training set is
used to update the model's parameters during training, while the validation set helps
monitor performance and prevent overfitting. The testing set evaluates the model's
final performance on unseen data.
3.6 Encoder Pathway
The U-Net architecture starts with a series of convolutional layers in the encoder
pathway. These layers are responsible for capturing high-level features and context
from the input images. As the architecture is based on a fully convolutional network,
the spatial information is preserved throughout the layers.
3.7 Down-sampling (Max Pooling)
The encoder performs down-sampling by using max pooling layers, reducing the
spatial dimensions of the feature maps while retaining the essential information. This
helps in capturing broader context and global patterns in the image.
10
3.8 Decoder Pathway
The decoder pathway is designed to up-sample the feature maps back to the original
image resolution. It uses up-convolutional layers or transposed convolutions to
achieve this. The skip connections are the unique aspect of U-Net, which helps in
fusing the low-level spatial information from the encoder pathway with the high-level
context from the decoder pathway. This enhances the model's localization capability.
3.9 Up-sampling (Up-convolution)
The decoder pathway performs up-sampling to restore the spatial resolution to the
original dimensions. The skip connections from the encoder pathway are concatenated
with the up-convolutional layers, allowing the model to access both local and global
information simultaneously.
3.10 Output Layer
At the end of the decoder pathway, the final output layer uses a suitable activation
function to generate the segmentation mask. The mask represents the probability of
each pixel belonging to a specific class, such as anemia or non-anemia.
3.11 Loss Function and Optimization
For the task of image segmentation, a common loss function is the Dice coefficient or
Soft Dice Loss, which measures the overlap between the predicted mask and the
ground truth mask. The model is trained using backpropagation and optimization
algorithms like stochastic gradient descent (SGD) or Adam to minimize the loss and
update the model parameters.
3.12 Inference and Evaluation
After training, the trained U-Net model can be used for inference on new, unseen
images. The model segments the input images and generates a probability map
indicating the presence of anemia. These probability maps can be thresholded to
obtain binary segmentation masks, which are then evaluated using metrics such as
Intersection over Union (IoU) or Dice coefficient to assess the model's performance.
11
3.13 Model Evaluation
Model evaluation involves assessing the performance of the U-Net architecture for
anemia diagnosis and classification. This is done using metrics like Sensitivity,
Specificity, Accuracy and Intersection over Union (IoU) or Dice coefficient to assess
the model's performance and to measure its ability to identify positive and negative
cases. The AUC-ROC curve assesses its discrimination ability, while cross-validation
estimates performance on unseen data. Interpretability analysis techniques provide
insights into the model's decision-making process. Through these evaluations, we can
determine how well the U-Net model performs in diagnosing and classifying anemia
in pregnant women in Nepal.
Along with the model evaluation with metrics, model validation using real data with
collaboration with medical specialists will be done.
Figure 5.2 :Proposed method for image segmentation using UNet
12
Figure 5.2 : Proposed architecture for model processing using RoI from UNet
13
Figure 5.3:Proposed workflow of mobile application
14
CHAPTER 4: DEVELOPMENT AND SOFTWARE
REQUIREMENT
The various tools used for the implementation of this project are listed below:
4.1 Development environment
I. Jupyter Notebook, Google Colaboratory (for model building and training)
II. Git and GitHub (for version controlling)
III. Visual Studio (for application development)
4.2 Programming Language, libraries, and packages
I. Python (base programming language)
II. Pandas, Matplotlib, Seaborn (for data manipulation and analysis)
III. NumPy (for numerical computation)
IV. TensorFlow, Keras, Scikit Learn (for data preprocessing and model building)
V. Django(Python web framework for building web apis)
VI. Flutter(Mobile application)
15
CHAPTER 5: EPILOGUE
5.1 Expected output
The expected output of the proposed project on Anemia detection in pregnant women
in Nepal using the U-Net architecture which is a trained model which specializes in
biomedical image segmentation and classification, capable of detecting different
stages of the disease using the images of finger nails, conjunctiva and palm. The
model will provide diagnostic predictions for each input image, indicating the
presence and severity of Anemia in pregnant women. The output may include a
classification label or a probability score indicating the likelihood of each severity
grade.
5.2 Budget Analysis
So the cost accrued for the preparation of the final project may include the expenses
in following:
●GPU Rental Cost : NRS. 12,000
●Research Cost : NRS. 15,000
●Deployment Cost: NRS. 30,000
The consideration of the above expenses indicate that the project’s gross expenses
may not exceed NRS. 57,000.
5.3 Scheduling
The scheduling of the project has been done as per the following Gantt chart:
Figure 5.3 : Gantt chart
16
CHAPTER 6: REFERENCES
[1] J. Asare, P. Appiahene, E. Arthur, S. Korankye, S. Afrifa, and E. Donkoh,
"Detection of anemia using conjunctiva images: A smartphone application approach,"
Medicine in Novel Technology and Devices, vol. 18, p. 100237, 2023. [Online].
Available: https://doi.org/10.1016/j.medntd.2023.100237
[2] N. Sevani, F. Fredicia, and G. Persulessy, "Detection of anemia based on
conjunctiva pallor level using k-means algorithm," in IOP Conference Series:
Materials Science and Engineering, vol. 420, p. 012101, N. J. EditorLastName & M.
K. EditorLastName (Eds.), IOP Publishing, 2018. [Online]. Available:
https://doi.org/10.1088/1757-899X/420/1/012101
[3] J. Asare, P. Appiahene, E. Donkoh, and G. Dimauro, "Iron Deficiency Anemia
Detection using Machine Learning Models: A Comparative Study of Fingernails,
Palm and Conjunctiva of the Eye Images," [Preprint], Authorea, 2023. [Online].
Available: https://doi.org/10.22541/au.167570558.82410707/v1
[4] M. Alyosify, M. Nashat, N. Essam, N. Ahmed, M. Abdelraouf, and K. Hamdy,
"Estimate Clinical Signs of Anemia and Noninvasive Hemoglobin Level
Measurement in a Mobile Phone Environment," [Data Set], Figshare, 2023. [Online].
Available: https://doi.org/10.6084/m9.figshare.21901218
[5] G. Dimauro, L. Baldari, D. Caivano, G. Colucci, and F. Girardi, "Automatic
segmentation of relevant sections of the conjunctiva for non-invasive anemia
detection," 2019.
[6] E. S. El-kenawy, E. Khodadadi, and F. M. Talaat, "Automated Detection and
Segmentation of COVID-19 Infection using Machine Learning," Journal of Artificial
Intelligence and Metaheuristics, vol. 3, pp. 28-37, 2023. [Online]. Available:
https://doi.org/10.54216/JAIM.030203
17
[7] P. Appiahene, J. Asare, E. Donkoh, G. Dimauro, and R. Maglietta, "Detection of
iron deficiency anemia by medical images: a comparative study of machine learning
algorithms," BioData Mining, vol. 16, 2023. [Online]. Available: DOI:
10.1186/s13040-023-00319-z
[8] L. Zheng, S. Liu, S. Tian, J. Guo, X. Wang, X. Liao, and J. Hong, "Enhancing
Intelligent Anemia Detection via Unifying Global and Local Views of Conjunctiva
Image with Two-Branch Neural Networks," 2021. [Online]. Available: DOI:
10.21203/rs.3.rs-1170958/v1
[9] N. Peksi, B. Yuwono, and M. F. Yanu, "Classification of Anemia with Digital
Images of Nails and Palms using the Naive Bayes Method," Telematika, vol. 18, p.
118, 2021. [Online]. Available: DOI: 10.31315/telematika.v18i1.4587.
[10] S. Suner, J. Rayner, I. U. Ozturan, G. Hogan, C. Meehan, A. Chambers, J. Baird,
and G. Jay, "Prediction of anemia and estimation of hemoglobin concentration using a
smartphone camera," PloS one, vol. 16, p. e0253495, 2021. [Online]. Available: DOI:
10.1371/journal.pone.0253495.
[11] J. Asare, P. Appiahene, E. Arthur, S. Korankye, S. Afrifa, and E. Donkoh,
"Detection of anemia using conjunctiva images: A smartphone application approach,"
Medicine in Novel Technology and Devices, vol. 18, p. 100237, 2023. [Online].
Available: DOI: 10.1016/j.medntd.2023.100237.
[12] S. Ghosal, D. Das, V. Udutalapally, and P. Wasnik, "iNAP: A Hybrid Approach
for NonInvasive Anemia-Polycythemia Detection in the IoMT," 2021. [Online].
Available: DOI: 10.1145/3503466.
18
[13] World Health Organization, "Anaemia in women and children," Global Health
Observatory(GHO)data.[Online].Available:https://www.who.int/data/gho/data/themes
/topics/anaemia_in_women_and_children
19
