Revolutionizing Neurological Diagnostics: Integrating 6G Technology with Deep Learning for Enhanced Detection of Multiple Sclerosis and Myelitis

Abstract

This paper proposes an alternative detection framework for multiple sclerosis (MS) and idiopathic acute transverse myelitis (ATM) within the 6G-enabled Internet of Medical Things (IoMT) environment. The developed framework relies on the implementation of a deep learning technique known as Dense Convolutional Networks (DenseNets) in the 6G-enabled IoMT to enhance prediction performance. To validate the performance of DenseNets, we compared it with other deep learning techniques, including Convolutional Neural Networks (CNN) and MobileNet, using real-world datasets. The experimental results show the high performance of DenseNets in predicting MS and ATM compared to other methods, achieving an accuracy of nearly 90%.

Full text

Revolutionizing Neurological Diagnostics:
Integrating 6G Technology with Deep Learning for
Enhanced Detection of Multiple Sclerosis and
Myelitis
1st Rana M. NourEldeen
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
rana.mohamed@gu.edu.eg
2nd Rahma A. Elshshtawy
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
rahma.elshshtawy@gu.edu.eg
3rd Omnia A. Elgohary
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
omnia.ayman@gu.edu.eg
4th Fatma Alzahraa A. Wahba
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
fatma.alzahraa@gu.edu.eg
5th Marco K. Attia
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
marko.kamal@gu.edu.eg
6th Zeinab B. Abo elazm
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
zeinab.bakry@gu.edu.eg
7th Mohamed G. Khattap
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
mohamed.ghareb@gu.edu.eg
8th Hend Galal Eldeen Mohamed Ali Hassan
Radiology and Medical Imaging,
Faculty of Applied Health Sciences,
Galala University
Suez 435611, Egypt
doctor hendgalal@gu.edu.eg
9th Mariam Mostafa
Biomedical Informatics,
Faculty of Computer Science
and Engineering,
Galala University
Suez 435611, Egypt
mariam.abdalwahab@gu.edu.eg
10th Nayra Ibrahim
Biomedical Informatics,
Faculty of Computer Science and
Engineering,
Galala University
Suez 435611, Egypt
nayra.elgazzaz@gu.edu.eg
11th Reem Ramadan
Biomedical Informatics,
Faculty of Computer Science and
Engineering,
Galala University
Suez 435611, Egypt
reem.abdelsalam@gu.edu.eg
12th Yomna Refaat
Biomedical Informatics,
Faculty of Computer Science and
Engineering,
Galala University
Suez 435611, Egypt
yomna.abdelalim@gu.edu.eg
13th Mennatallah Khaled
Biomedical Informatics,
Faculty of Computer Science and
Engineering,
Galala University
Suez 435611, Egypt
mennatallah.ali@gu.edu.eg
14th Ali Maher
Artificial Intelligence Science Program,
Faculty of Computer Science
and Engineering
Galala University
Suez 435611, Egypt
aly.abdelrahman@gu.edu.eg
15th Mohamed Abd Elaziz
Artificial Intelligence Science Program,
Faculty of Computer Science
and Engineering
Galala University
Suez 435611, Egypt
abd el aziz m@yahoo.com
16th Ahmed Gamal Abdellatif Ibrahim
Department of Electronics and Electrical Communications Engineering,
Air Defense College
Alexandria, Egypt
ag.abdellatef@zu.edu.eg

Abstract—This paper proposes an alternative detection frame-
work for multiple sclerosis (MS) and idiopathic acute transverse
myelitis (ATM) within the 6G-enabled Internet of Medical Things
(IoMT) environment. The developed framework relies on the
implementation of a deep learning technique known as Dense
Convolutional Networks (DenseNets) in the 6G-enabled IoMT to
enhance prediction performance. To validate the performance of
DenseNets, we compared it with other deep learning techniques,
including Convolutional Neural Networks (CNN) and MobileNet,
using real-world datasets. The experimental results show the high
performance of DenseNets in predicting MS and ATM compared
to other methods, achieving an accuracy of nearly 90%.
Index Terms—6G networks; Internet of medical things; deep
learning; Multiple Sclerosis and Myelitis.
I. INTRODUCTION
Acute myelopathies encompass a diverse set of neurological
disorders marked by rapid onset spinal cord dysfunction
stemming from various causes, leading to significant acute
and potential long-term disability. These disorders are cate-
gorized by their symptom onset speed and deterioration rate,
with inflammatory myelopathies appearing suddenly within
days, and hereditary types evolving over months, although
some chronic forms can acutely worsen [1]. The severity
and outcomes depend on the nature and extent of spinal
cord damage, with distinct subtypes including inflammatory
myelopathies like multiple sclerosis (MS), idiopathic acute
transverse myelitis (ATM), autoantibodies against aquaporin-4
(AQP4-IgG), and myelin oligodendrocyte glycoprotein (MOG-
IgG) each with specific characteristics and implications for
diagnosis and treatment [2]. Distinguishing among the diverse
types of myelopathies—whether they stem from inflammatory,
infectious, or demyelinating origins—presents a challenge
but is crucial for initiating timely and appropriate treatment
while preventing harm caused by unnecessary procedures
[3]. In 2002, the Transverse Myelitis Consortium Working
Group (TMCWG) established diagnostic criteria for ATM,
emphasizing the need for clinical evidence of spinal cord
dysfunction, neuroimaging to rule out structural causes, and
proof of inflammation through MRI or cerebrospinal fluid
(CSF) analysis, among other exclusions, to classify a case as
definite or possible ATM [4]. However, even with these strict
guidelines, a relatively small percentage of patients are diag-
nosed as having ATM, and some initially diagnosed patients
may later receive different diagnoses, reflecting challenges in
accurately identifying this condition [5], [6].
Diagnosing acute myelopathy begins with a comprehen-
sive neurological evaluation to identify the affected spinal
cord region, followed by Magnetic resonance Imaging (MRI)
to investigate compressive or structural causes. MRI with
gadolinium contrast is the preferred method [7]. ATM typically
shows a central T2 hyperintense lesion extending over more
than two spinal segments and occupying more than two-
thirds of the cord’s cross-sectional area, often with variable
enhancement patterns [8]. On the other hand, MS features
smaller spinal lesions, usually affecting less than two segments
and often located dorsolaterally, with T2 hyperintensity and
contrast enhancement [8]. MS lesions demonstrate a pattern
of dissemination in time and space, with the presence of
oligoclonal bands in CSF further supporting the diagnosis
[9], distinguishing it from ATM through imaging and clinical
follow-up.
Despite high sensitivity, MRI may not always definitively
differentiate between MS and ATM due to overlapping imag-
ing features necessitating supplementary invasive procedures
like CSF analysis for conclusive diagnosis. Approximately
40% of acute transverse myelopathies may remain undemon-
strated on MRI, and no clearly different and specific patterns
have been conclusively identified for each etiology [10],
[11]. This diagnostic ambiguity not only complicates clinical
decision-making but also delays the initiation of appropriate
treatment regimes, highlighting the imperative need for inno-
vative diagnostic approaches.
Recently, Artificial Intelligence (AI) has emerged as a
promising adjunct by potentially enhancing the specificity and
sensitivity of MRI interpretations, thereby facilitating early
and accurate differentiation of these conditions [12], [13].
Additionally, the large quantity of data generated quickly at
a short timescale, raises the problem of efficiently processing
such images in real-time to help the medical field detect these
diseases in their early stages. So, the integration of sixth-
generation (6G) communication solutions with AI heralds
a transformative leap in the diagnostic precision and man-
agement of neurological conditions. This synergy promises
to overcome the limitations of MRI in distinguishing these
disorders by enabling real-time, high-speed data transmis-
sion and advanced AI analytics. The high bandwidth and
ultra-low latency characteristic of 6G facilitate the seamless
transfer of large volumes of imaging data, enhancing the
capabilities of AI algorithms in processing and analyzing
MRI scans with unprecedented speed and accuracy [14]–[16].
This technological advancement not only aids in the early
and accurate differentiation of acute myelopathies but also
supports telemedicine applications, allowing neurologists to
deliver timely, informed, and personalized patient care.
Little research has focused on differentiating MS from
its mimics, such as Neuromyelitis Optica Spectrum Disorder
(NMOSD). For example, Yoo et al. [17] introduced a hierar-
chical multimodal fusion (HMF) deep learning (DL) model to
distinguish NMOSD from MS using MRI and diffusion tensor
images, achieving an 81.3% diagnostic accuracy. Similarly,
Wang et al. [18] developed a DL model using 2D CNNs
with transfer learning to differentiate NMOSD from MS in
MRI images, outperforming traditional 3D CNN models with
higher accuracy, sensitivity, and specificity over 70%. Another
study by Eshaghi et al. [19] used multi-kernel learning for
the automatic diagnosis of NMO and MS, achieving an 88%
accuracy in differentiating NMO from MS and 84% accuracy
in distinguishing between NMO, MS, and healthy controls.
Despite these advancements, the diagnostic accuracy remains
relatively low. To the best of our knowledge, only one study
has presented a model specifically for classifying MS from
ATM.
The economic feasibility and scalability of deploying a 6G-

AI diagnostic system for neurological conditions are crucial,
particularly in resource-limited settings. High initial invest-
ments for infrastructure and ongoing operational costs must
be considered. The proposed solution leverages fog computing
to reduce costs and latency. A phased deployment approach,
starting in critical regions, can help manage expenses. Overall,
the economic benefits, such as improved patient outcomes and
reduced invasive procedures, must be carefully evaluated.
In this paper, we proposed DL technique to enhance the
prediction of MS and ATM. In general, the data is collected
using devices and will be sent to Fog computing in a 6G
network that will pass the data to the cloud computing layer.
Within this layer, the DL model is learned using the collected
data and then we will distribute the model to preserve it at the
fog computing layer. This will reduce the transmission cost
and can support the way area without a sufficient number of
doctors.
The main contribution of this study can be summarized as
follows:
1) Propose an alternative Neurological Diagnostics Frame-
work in 6G IoMT environment to enhance the prediction
of MS and ATM.
2) Apply Densnet as a DL technique to predict the predic-
tion of MS and ATM.
3) Apply the developed model to the real-world and com-
pare it with other DL models.
The remainder of this paper is structured as follows: Sec-
tion 2 outlines our methodological approach by comparing
three DL models: Convolutional Neural Networks (CNNs),
MobileNet, and Dense Convolutional Networks (DenseNets),
thereby laying the groundwork for our computational strategy.
In Section 3, we introduce our proposed framework that
merges the 6G IoMT with our DL techniques to improve the
diagnosis of MS and ATM, illustrating the synergistic potential
of advanced technologies in healthcare. Section 4 delves
into the results of our experiments, underscoring DenseNet’s
outstanding accuracy in diagnosing MS and ATM, which in
turn confirms our framework’s efficacy. Concluding the paper,
Section 5 recaps our main contributions and explores future
research avenues, particularly the expansion of our 6G-IoMT
model to diagnose additional neurological conditions, ensuring
clarity and conciseness while enhancing overall readability.
II. METHODS
Within this section, we introduce basic information about
CNN, MobileNet, and DenseNet as DL techniques.
A. CNN
In the field of DL, CNNs are widely used, especially
for applications involving medical imaging [20]. For division
neural networks, it’s an amended version of a fully connected
multilayer feed, aimed at identifying local peculiarities to
classify them. The basic architecture of a CNN includes the
input layer, convolutional layer, pooling layer, fully connected
layer, and output layer. The convolutional layer and pooling
layer are the major components in the CNN architecture.
The basic CNN architecture, designed particularly for medical
image classification, is shown in Figure 1. The original images,
set up by several neurons in the input layer, correspond to the
feature dimension input. The convolution filters in the convo-
lutional layer extract features from the image using a fixed step
length. Additionally, each neuron in the convolutional layer is
connected to a group of weights to a specific part of the image
in the top layer. Among the neurons in the convolutional layer,
there is a local connection. By obtaining local features from
the surface of the input layer, feature maps are easily created.
This effective extraction of local image features enables the
convolutional filter of the current layer to function efficiently.
To achieve the feature extraction function, the convolutional
layer performs convolution operations by inspecting the input
data using its internal convolution process. Extracting local
features from the input layer’s surface allows for the creation
of feature maps, which in turn enables the current layer’s
convolution filter to effectively capture local image features.
The convolutional layer performs convolution operations by
examining the input data with its internal convolution process
to carry out the feature extraction function.
Xi
j=f
X
i∈Mj
Xl−1
i×wl
ij +bl
j

(1)
To achieve spatial invariance, the pooling layer performs
feature selection and information filtering by decreasing the
resolution of the feature map. Rapidly reducing the matrix
size, pooling operations like max pooling and average pooling
decrease the network’s parameters and accelerate computation.
Additionally, downsampling the input data helps mitigate
overlapping issues.
Xi
j=Bl
j×d(Xl−1
i) + bl
j)(2)
The flexibility of the network model is enhanced by facil-
itating the completion of image categorization and recogni-
tion tasks through the use of convolutional layers, pooling
layers, and local category discrimination information. The
fully connected layer transforms the sample label space into
the obtained ’distributed feature representation.’ Moreover, the
commonly applied ReLU activation function follows the fully
connected layer to enhance the performance of convolutional
neural networks.
B. MobileNet
Algorithms like depth-wise separable convolution, as de-
scribed by MobileNet [21], are used to further strengthen
CNN designs, as shown in Figure 2. Depth-wise separable
convolution reduces the total number of parameters needed
for the model without losing computational performance or
accuracy by breaking down traditional convolution into depth-
wise and point-wise convolution. Additionally, the problem of
limited receptive field size can be addressed by using dilation
convolution in neural networks, which allows the learning
of characteristics at multiple scales and levels. Increased

Fig. 1: Structure of CNN model.
computation enhances the ability to extract intrinsic image
features while maintaining spatial resolution and global data
by expanding the receptive field of convolutional kernels.
Furthermore, by employing linear algebra operations like the
Sigmoid function, the enhanced MobileNet model improves
depth-wise separable convolutional blocks. This enhancement
aids in preserving channel data and improves classification and
recognition precision. When assessing the performance of DL
models in tasks such as classification, variables like accuracy,
sensitivity, and specificity are applied. These measures are
essential for determining how well various algorithms and
variable options perform categorization tasks. In summary,
modifications in CNN design, such as MobileNet—which
combines techniques like dilation convolution and depth-
wise separable convolution—along with the application of
effective evaluation metrics, have a significant positive impact
on improving the accuracy and efficiency of medical image
categorization and recognition applications.
Fig. 2: Structure of the MobileNet.
C. Dense Convolutional Networks (DenseNets)
DenseNets [22] are characterized by their densely connected
architecture. Unlike standard CNNs, which usually consist of
sequentially stacked layers, DenseNets have a more complex
design.
In general, the fundamental unit in DenseNets is the dense
block. Each dense block comprises multiple convolutional
layers stacked together (see Figure 3). Crucially, every layer
within a dense block receives feature maps from all preceding
layers within the same block. The feature maps from different
layers are concatenated along the channel dimension. This
design choice encourages feature reuse and facilitates the flow
of information across layers. By directly connecting each layer
to all subsequent layers, DenseNets mitigate the vanishing-
gradient problem, allowing gradients to flow more effectively
during backpropagation.
Transition layers serve as connectors between dense blocks.
They consist of (1) Batch Normalization Layer: Normalizes
feature maps to stabilize training. (2) 1x1 Convolutional Layer:
Reduces the number of channels, controlling computational
complexity. (3) Average Pooling: Reduces spatial dimensions
while preserving essential information.
Fig. 3: Structure of basic DenseNet.
III. PROP OS ED M OD EL 6G IOMT FR AM EW OR K
In this section, we present a detailed methodology of the
developed 6G IoMT Framework for the detection of MS and
ATM. This framework leverages DenseNets and integrates
them within a 6G-enabled IoMT environment to enhance
prediction performance. The proposed framework is illustrated
in Figure 4.
The first step in our methodology involves data collection.
Medical images of MS and ATM are acquired from medical
imaging devices. These images are then processed at the edge
computing layer, which includes fog computing and Multi-
access Edge Computing (MEC) servers. This layer performs
preliminary computations, reducing latency and bandwidth
usage. Aggregated data from various edge devices are sub-
sequently transmitted to the cloud computing layer through
the 6G network. In the cloud computing layer, the DenseNet
model is trained using the medical images. Once trained, the
DenseNet model is deployed back to the fog computing layer
to enable real-time predictions. Medical professionals can use
the deployed model to perform diagnostics on new medical
images, receiving predictions with high accuracy.
Additionally, our model offers another service: if an expert
aims to predict the outcome for a current patient, the test
pattern in the API forecasting tools is used. By leveraging both
fog and cloud computing, the framework ensures efficient data
transmission and facilitates model deployment across diverse
healthcare settings.

Fig. 4: The suggested 6G-enabled IoMT framework diagram.
The model’s performance is evaluated using metrics such as
accuracy, F1-score, precision, and recall to ensure its reliability
in clinical settings. The framework also includes a forecasting
and continuous learning component. Diagnostic results are
integrated with forecasting models to predict the progression
of MS and ATM. The model is continuously updated with new
data to improve its diagnostic capabilities over time.
While the proposed 6G IoMT Framework is currently
conceptual, its future implementation will involve real-world
testing and validation to refine the integration of 6G technol-
ogy with DL models in medical diagnostics.
IV. EXPERIMENTAL RESULTS
A. Dataset Description
This study utilized a novel dataset of 2,746 MR images,
approved by Fırat University’s Ethics Committee, featuring
patients with myelitis but not MS, MS patients, and healthy
controls. The myelitis cases showed lesions exceeding three
vertebral lengths, while MS lesions were generally shorter
than two vertebral segments and located posterolaterally.
The control group exhibited no spinal lesions. Due to lesion
size variability, the number of images per subject varied,
with 4 to 6 images for sagittal views and 6 to 10 for
axial views, as depicted in Fig. 5. The dataset, stored in
JPG format and detailed further in Table. I, includes 150
healthy controls, 128 MS patients, and 131 myelitis patients,
segmented in both axial and sagittal planes for analysis.
This valuable dataset is accessible for research purposes at
(https://www.kaggle.com/datasets/turkertuncer/myelitis3classes
(accessed on 13 March 2024)) [13].
TABLE I: Properties of the collected dataset.
Class Male Female Total Age (Mean ±sd) Number of MRIs
Myelitis 63 68 131 35.7±5.22 667
MS 55 73 128 33.4±6.24 706
Control 75 75 150 34.4±3.16 1373
(a) (b) (c)
(d) (e) (f)
Fig. 5: Sample slices from the collected dataset, demonstrating
typical features in each category. (a), (d) Myelitis; (b), (e)
Multiple Sclerosis (MS); (c), (f) Control group.
B. Results and Discussion
In this section, we discuss the performance comparison of
the DenseNet model with other DL techniques, specifically
CNN and MobileNet, based on key metrics: accuracy, preci-
sion, recall, and F1-score. The implementation of those DL
techniques is determined based on the original references.
Table II presents the detailed performance metrics for each
model.
From those results, we can observe that the DenseNet
model outperformed both CNN and MobileNet across all
evaluated metrics. The accuracy of DenseNet reached 90.22%,
which is notably higher than CNN’s 82.97% and MobileNet’s
88.77%. This indicates that DenseNet provides a more reliable
prediction for diagnosing MS and ATM. Similarly, in terms
of precision, DenseNet achieved 88.84%, compared to CNN’s
80.99% and MobileNet’s 88.25%. Higher precision suggests

that DenseNet is more effective in correctly identifying true
positive cases, reducing the likelihood of false positives. The
recall metric, which measures the ability of the model to
identify all relevant cases, also favored DenseNet with a value
of 88.61%. In comparison, CNN and MobileNet achieved
81.16% and 87.45%, respectively. This higher recall value
for DenseNet indicates its superior sensitivity in detecting
actual positive cases of MS and ATM. Finally, the F1-score,
which provides a balance between precision and recall, was
highest for DenseNet at 88.72%, while CNN and MobileNet
scored 81.08% and 87.82%, respectively. This further confirms
DenseNet’s robust performance in predicting these neurologi-
cal conditions.
Examining the confusion matrices for each model, as il-
lustrated in Figure 6, provides a more detailed understanding
of their classification performance. The confusion matrices
reveal that DenseNet significantly reduces misclassifications
compared to CNN and MobileNet, leading to higher precision,
recall, and overall accuracy.
TABLE II: Performance Metrics Comparison of CNN, Mo-
bileNet, and DenseNet Models in Diagnosing MS and ATM
Measure CNN MobileNet DenseNet
Accuracy 82.97% 88.77% 90.22%
Precision 80.99% 88.25% 88.84%
Recall 81.16% 87.45% 88.61%
F1-score 81.08% 87.82% 88.72%
To further contextualize our findings, we compared the
performance of our proposed methodology with previous
studies, as shown in Table III. This comparison highlights
the significant improvements our approach offers over past
methodologies. For instance, 3D CNN [23] achieved an accu-
racy of 71.1%, and the hierarchical multimodal fusion model
[17] reached 81%. Gray matter volumes with random forest
[24] showed an accuracy of 80%, while the multi-kernel learn-
ing model [19] achieved 84%. Our DenseNet-based approach
not only outperformed these studies but also consistently
provided higher accuracy across different evaluation metrics.
This underscores the value of integrating 6G technology with
advanced DL models in the IoMT framework for neurological
diagnostics.
Overall, the superior performance of DenseNet across all
metrics, coupled with its highest accuracy compared to previ-
ous studies, demonstrates its potential effectiveness for future
integration in a 6G-enabled IoMT environment for detecting
MS and ATM. Although we have not yet integrated the 6G-
enabled IoMT environment with our model, our results suggest
that such an integration could enable faster and more accurate
diagnostics, which are crucial for timely and appropriate
treatment interventions. These results validate the potential of
our proposed framework to revolutionize neurological diag-
nostics, providing a strong foundation for future research and
development in this area.
(a) CNN
(b) MobileNet
(c) DenseNet
Fig. 6: Confusion matrices of the models: (a) CNN, (b)
MobileNet, and (c) DenseNet, showcasing the classification
performance and misclassifications for each model.

TABLE III: Comparison of Accuracy Between Previous Stud-
ies and the Proposed DenseNet-Based Methodology for Neu-
rological Diagnostics.
Previous Studies Accuracy
Kim et al. [23] 71.1%
Yoo et al. [17] 81%
Wang et al. [18] 75%
Eshaghi et al. [24] 80%
Eshaghi et al [19] 84%
Amin et al. [25] 78%
Huang et al. [26] 81.4%
Seok et al. [12] 76.1%
Proposed Methodology 90.22%
V. CONCLUSION AND FUTURE WORKS
In this paper, we developed an alternative neurological
diagnostic technique for the detection of MS and ATM within
the 6G network. The developed model depends on applying the
DenseNet model to enhance the prediction process. To validate
the performance of the developed model, a set of real-world
datasets was used. Additionally, we conducted comparisons
with other DL techniques, namely CNN and MobileNet. From
the results, the DenseNet has established its performance
among the other techniques.
Besides the promising results, we can extend the developed
model’s neurological diagnostic technique in a 6G-IoMT en-
vironment to other diseases.
REFERENCES
[1] E. Sechi and E. P. Flanagan, “Evaluation and management of acute
myelopathy,” in Seminars in Neurology, vol. 41, no. 05. Thieme
Medical Publishers, Inc., 2021, pp. 511–529.
[2] S. L. Chiriboga and E. P. Flanagan, “Myelitis and other autoimmune
myelopathies,” CONTINUUM: Lifelong Learning in Neurology, vol. 27,
no. 1, pp. 62–92, 2021.
[3] T. W. West, C. Hess, and B. A. Cree, “Acute transverse myelitis:
demyelinating, inflammatory, and infectious myelopathies,” in Seminars
in neurology, vol. 32, no. 02. Thieme Medical Publishers, 2012, pp.
097–113.
[4] T. M. C. W. Group*, “Proposed diagnostic criteria and nosology of acute
transverse myelitis,” Neurology, vol. 59, no. 4, pp. 499–505, 2002.
[5] J. De Seze, C. Lanctin, C. Lebrun, I. Malikova, C. Papeix,
S. Wiertlewski, J. Pelletier, O. Gout, C. Clerc, C. Moreau et al.,
“Idiopathic acute transverse myelitis: application of the recent diagnostic
criteria,” Neurology, vol. 65, no. 12, pp. 1950–1953, 2005.
[6] J. Bruna, S. Mart´
ınez-Y´
elamos, A. Mart´
ınez-Y´
elamos, F. Rubio, and
T. Arbizu, “Idiopathic acute transverse myelitis: a clinical study and
prognostic markers in 45 cases,” Multiple Sclerosis Journal, vol. 12,
no. 2, pp. 169–173, 2006.
[7] T. W. West, “Transverse myelitis—a review of the presentation, diagno-
sis, and initial management,” Discovery medicine, vol. 16, no. 88, pp.
167–177, 2013.
[8] C. Goh, P. M. Desmond, and P. M. Phal, “Mri in transverse myelitis,”
Journal of Magnetic Resonance Imaging, vol. 40, no. 6, pp. 1267–1279,
2014.
[9] A. J. Thompson, B. L. Banwell, F. Barkhof, W. M. Carroll, T. Coetzee,
G. Comi, J. Correale, F. Fazekas, M. Filippi, M. S. Freedman et al., “Di-
agnosis of multiple sclerosis: 2017 revisions of the mcdonald criteria,”
The Lancet Neurology, vol. 17, no. 2, pp. 162–173, 2018.
[10] S. Weidauer, E. Hattingen, and C. T. Arendt, “Cervical myelitis: a
practical approach to its differential diagnosis on mr imaging,” in R¨
oFo-
Fortschritte auf dem Gebiet der R¨
ontgenstrahlen und der bildgebenden
Verfahren. Georg Thieme Verlag KG, 2023.
[11] G. Scotti and S. Gerevini, “Diagnosis and differential diagnosis of acute
transverse myelopathy. the role of neuroradiological investigations and
review of the literature,” Neurological Sciences, vol. 22, pp. S69–S73,
2001.
[12] J. M. Seok, W. Cho, Y. H. Chung, H. Ju, S. T. Kim, J.-K. Seong, and
J.-H. Min, “Differentiation between multiple sclerosis and neuromyelitis
optica spectrum disorder using a deep learning model,” Scientific reports,
vol. 13, no. 1, p. 11625, 2023.
[13] S. Tatli, G. Macin, I. Tasci, B. Tasci, P. D. Barua, M. Baygin, T. Tuncer,
S. Dogan, E. J. Ciaccio, and U. R. Acharya, “Transfer-transfer model
with msnet: An automated accurate multiple sclerosis and myelitis de-
tection system,” Expert Systems with Applications, vol. 236, p. 121314,
2024.
[14] M. A. Elaziz, A. Dahou, A. Mabrouk, R. A. Ibrahim, and A. O. Aseeri,
“Medical image classifications for 6g iot-enabled smart health systems,”
Diagnostics, vol. 13, no. 5, p. 834, 2023.
[15] W. Wang, F. Liu, X. Zhi, T. Zhang, and C. Huang, “An integrated deep
learning algorithm for detecting lung nodules with low-dose ct and its
application in 6g-enabled internet of medical things,” IEEE Internet of
Things Journal, vol. 8, no. 7, pp. 5274–5284, 2020.
[16] M. Abd Elaziz, A. Mabrouk, A. Dahou, S. A. Chelloug et al., “Medical
image classification utilizing ensemble learning and levy flight-based
honey badger algorithm on 6g-enabled internet of things,” Computa-
tional Intelligence and Neuroscience, vol. 2022, 2022.
[17] Y. Yoo, L. Y. Tang, S.-H. Kim, H. J. Kim, L. E. Lee, D. K. Li,
S. Kolind, A. Traboulsee, and R. Tam, “Hierarchical multimodal fusion
of deep-learned lesion and tissue integrity features in brain mris for
distinguishing neuromyelitis optica from multiple sclerosis,” in Medical
Image Computing and Computer Assisted Intervention- MICCAI 2017:
20th International Conference, Quebec City, QC, Canada, September
11-13, 2017, Proceedings, Part III 20. Springer, 2017, pp. 480–488.
[18] Z. Wang, Z. Yu, Y. Wang, H. Zhang, Y. Luo, L. Shi, Y. Wang, and
C. Guo, “3d compressed convolutional neural network differentiates
neuromyelitis optical spectrum disorders from multiple sclerosis using
automated white matter hyperintensities segmentations,” Frontiers in
Physiology, vol. 11, p. 612928, 2020.
[19] A. Eshaghi, S. Riyahi-Alam, R. Saeedi, T. Roostaei, A. Nazeri, A. Agh-
saei, R. Doosti, H. Ganjgahi, B. Bodini, A. Shakourirad et al., “Clas-
sification algorithms with multi-modal data fusion could accurately
distinguish neuromyelitis optica from multiple sclerosis,” NeuroImage:
Clinical, vol. 7, pp. 306–314, 2015.
[20] H. Raja, M. U. Akram, A. Shaukat, S. A. Khan, N. Alghamdi, S. G.
Khawaja, and N. Nazir, “Extraction of retinal layers through convolution
neural network (cnn) in an oct image for glaucoma diagnosis,” Journal
of Digital Imaging, vol. 33, pp. 1428–1442, 2020.
[21] W. Sae-Lim, W. Wettayaprasit, and P. Aiyarak, “Convolutional neural
networks using mobilenet for skin lesion classification,” in 2019 16th
international joint conference on computer science and software engi-
neering (JCSSE). IEEE, 2019, pp. 242–247.
[22] T. Zhou, X. Ye, H. Lu, X. Zheng, S. Qiu, Y. Liu et al., “Dense
convolutional network and its application in medical image analysis,”
BioMed Research International, vol. 2022, 2022.
[23] H. Kim, Y. Lee, Y.-H. Kim, Y.-M. Lim, J. S. Lee, J. Woo, S.-K. Jang,
Y. J. Oh, H. W. Kim, E.-J. Lee et al., “Deep learning-based method
to differentiate neuromyelitis optica spectrum disorder from multiple
sclerosis,” Frontiers in Neurology, vol. 11, p. 599042, 2020.
[24] A. Eshaghi, V. Wottschel, R. Cortese, M. Calabrese, M. A. Sahraian,
A. J. Thompson, D. C. Alexander, and O. Ciccarelli, “Gray matter mri
differentiates neuromyelitis optica from multiple sclerosis using random
forest,” Neurology, vol. 87, no. 23, pp. 2463–2470, 2016.
[25] M. Amin, K. Nakamura, and D. Ontaneda, “Differentiating multiple
sclerosis from non-specific white matter changes using a convolutional
neural network image classification model,” Multiple Sclerosis and
Related Disorders, vol. 82, p. 105420, 2024.
[26] C. Huang, W. Chen, B. Liu, R. Yu, X. Chen, F. Tang, J. Liu, and
W. Lu, “Transformer-based deep-learning algorithm for discriminating
demyelinating diseases of the central nervous system with neuroimag-
ing,” Frontiers in immunology, vol. 13, p. 897959, 2022.