1. Deep Learning Framework-based
Automated Multi-class Diagnosis for
Neurological Disorders
Syed Saad Azhar Ali1, Khuhed Memon2, Norashikin Yahya2, Karim Asif Sattar1, Sami El Ferik1
1Research Center for Smart Mobility & Logistics, King Fahad University of Petroleum and
Minerals, KSA
2Centre for Intelligent Signal and Imaging Research (HICoE-CISIR), Department of Electrical
and Electronic Engineering, Universiti Teknologi PETRONAS, Malaysia
email: saadazhar@ieee.org, khuhed_22000210, norashikin_yahya@utp.edu.my,
karimas@kfupm.edu.sa, selferik@kfupm.edu.sa
2. Presentation outline
1
• Background
• Introduction
• Motivation
• Methodology
• MRI varieties
• Disease pool and dataset
• DL architecture
• Results
• Way forward
• Conclusion
4. Introduction
3
Neurological/ Neurodegenerative disorders
Among the top 3 killers[i]
Cause severe disabilities - eventual death [1]
Increasing incidence in older population [1]
Diagnosis
Diagnostic techniques rely on patient symptoms [2]
Symptomatic diagnoses are subjective [2]
Recent Computer Aided Diagnosis methods assist
neurologists
AI based approaches – dominant [1][2]
[i] https://www.who.int/news-room/fact-sheets/detail/the-top-10-causes-of-death
5. Motivation
4
• Early diagnosis may help [1][2]
• Diagnosis by manual examination of MRIs needs:
• Highly skilled and experienced experts
• Other tests to reach a concrete conclusion
• Time
• Brain MRI images possess very similar features, specially in case of
neurodegenerative disorders [3][4]
• Some neurological disorders in their very initial stages might get
overlooked [2]
• Result - delay in preemptive measures to decelerate disease progression
• Intelligent systems for identification of diseases can play a pivotal role in
such scenarios [5]
6. Recent Diagnostic Advancements
5
• ML vs DL!
• DL has shown greater efficiency for Brain MRI
classification [6][7]
• ML and DL in Medical Diagnosis [6]
• Promising results
• Artificial Intelligence (AI) – The Revolution!
[1]
• Machine Learning (ML) and Deep Learning (DL)
• CAD
• SVM and CNN – widely used [7]
• CAD extended to tools development
• Found few [8][9]
7. 6
Literature Survey
Alzheimer’s Disease (AD)
Tumor
Dementia
Schizophrenia
Mild Cognitive Impairment (MCI)
Parkinson’s Disease (PD)
Multiple Sclerosis (MS)
Meningioma
ADHD
Cancer
Bipolar Disorder
Stroke
Keywords: MRI Neurological
Disorders Machine Learning Deep
Learning Computer Aided Diagnosis
80%
18%
2%
MRI AND AI BASED NEUROLOGICAL DISORDERS
DIAGNOSIS - RESEARCH AREAS
Individual disease based Research
Research catering 2-4 diseases/ sub-types
Research catering more than 4 diseases (Unified model)
8. Prior Art Highlights
7
• Top 2 researched diseases: Alzheimer’s Disease (AD) and Tumors
• Majority of work caters Single Disease Diagnosis
• Small fraction deals with Multiple Disease Diagnosis
• Available CAD research holds the following limitations:
• Small dataset for training
• Questionable Accuracy of inference
• Low Reliability
• Few diagnoseable diseases
• Very few Tools designed to assist neurologists
• These limitations can be mitigated in a unified, comprehensive, efficient
and reliable framework, which can assist in diagnosis of multiple
neurological disorders, using MRI modality
9. Methodology
8
1
2
3
Dataset Consolidation &
Labelling
Data Investigation for types of
MRI images
Deep Learning Model Training
& Testing
Unified Framework for Multi-
Class Classification
App/ Tool Development for
Real-Time Diagnosis from MRI
Phases
Research
Development
11. Most common diseases (questionable accuracy in the existing algorithms)
Diseases not used in Multiple disease CAD
Based on available datasets
Neurodegenerative Diseases
The pool can be extended to cater:
• ADHD
• Etc.
Disease pool
10
DL Model(s)
Parkinson’s
Multiple
Sclerosis
Traumatic
Brain Injury
Stroke
Tumor/
types
Alzheimer’s/
stages
Other MRI diagnosable Diseases/ conditions
The pool can be extended to cater:
• Schizophrenia
• Autism
• Etc.
12. Dataset
11
Disease
Source
Subjects
Files
Orientation
Brain Tumor
Kaggle
1500 Tumor, 1500 Normal
JPEG images
Axial
PD
Neurocon and Taowu
36 Normal, 47 PD
NIfTI
Axial
Platform
Train/Val/Test
Images
MATLAB R2022b
80/ 10/ 10 percent split
2220 Normal, 940 PD and 1500 Tumor. (20 deep brain
slices extracted from Neurocon and Taowu NIfTI volumes)
13. Dataset for Brain Tumor
12
https://www.kaggle.com/ahmedhamada0/brain-tumor-detection
14. Dataset for Parkinson’s Disease
13
http://fcon_1000.projects.nitrc.org/indi/retro/parkinsons.html
Control
PD
15. Hardware, software and DL architecture
ResNet50
Total learnables: 23.5M
Layers: 177
Input: 224 x 224
Output: Image Classification
DL architecture:
14
Windows 11 Home Version 22H2
MATLAB Version: 9.13.0.2166757
(R2022b) Update 4
Python 3.10.7
6GB NVIDIA GeForce GTX 1660 Ti and
12GB NVIDIA GeForce RTX 2080 Ti
Android Studio Chipmunk
2021.2.1
Infinix X690B
OS Android 10
TensorFlow 2.8.0
16. Crux of the story
Deep Learning
Features Differential Diagnosis
PD, tumor or normal
…
Currently a three-class system,
which can later be extended to
handle multiple neurological
disorders.
DL model to analyze
input MRI
15
Computer Aided Differential
Diagnosis
18. Results
17
Brain MRI with Tumor
correctly classified
Brain MRI without Tumor
correctly classified
Android Studio Chipmunk 2021.2.1
Infinix X690B
OS Android 10
TensorFlow 2.8.0
19. Way forward
Incorporation of XAI to
build confidence of
medical experts in CAD
Expanding disease pool
and training dataset
Deployment of the
developed tool in
hospitals for testing and
feedback
18
20. Expanding labelled dataset
19
Clinical Diagnosis
Add MRI and Label to database for model tuning
Please Input MRI
Image
Please state
actual Clinical
Diagnosis:
Inference
Parkinson’s
21. Conclusion
20
• CAD for neurological disorders has proven to be of immense
importance in recent years.
• This research is a humble endeavor to create a generic
framework capable of handling varying MRI types and inferring
diagnoses using the pipeline of DL/ML models.
• Even with the very limited MRI data fed into the training of the
model, it was able to distinguish between tumor, PD and normal
classes with above chance accuracy.
• It is anticipated that the proposed algorithm (to distinguish
between multiple neurological disorders) will perform
satisfactorily with an acceptable accuracy after validation from
radiologists.
22. References
21
1) Tautan, Alexandra-Maria, Bogdan Ionescu, and Emiliano Santarnecchi. "Artificial Intelligence in
Neurodegenerative Diseases: A Review of Available Tools with a Focus on Machine Learning
Techniques." Artificial Intelligence in Medicine (2021): 102081
2) Nemoto, Kiyotaka, et al. "Differentiating dementia with Lewy bodies and Alzheimer's disease by deep
learning to structural MRI." Journal of Neuroimaging 31.3 (2021): 579-587.
3) Avants, Brian B., et al. "Symmetric diffeomorphic image registration with cross-correlation: evaluating
automated labeling of elderly and neurodegenerative brain." Medical image analysis 12.1 (2008): 26-41.
4) Tomson, Torbjörn. "Excess mortality in epilepsy in developing countries." The Lancet. Neurology 5.10
(2006): 804-805.
5) Miotto, Riccardo, et al. "Deep learning for healthcare: review, opportunities and challenges." Briefings in
bioinformatics 19.6 (2018): 1236-1246.
6) Greenspan, Hayit, Bram Van Ginneken, and Ronald M. Summers. "Guest editorial deep learning in medical
imaging: Overview and future promise of an exciting new technique." IEEE transactions on medical
imaging 35.5 (2016): 1153-1159.
7) Rashid, Mamoon, Harjeet Singh, and Vishal Goyal. "The use of machine learning and deep learning
algorithms in functional magnetic resonance imaging—A systematic review." Expert Systems 37.6 (2020):
e12644.
8) Deepa, B., et al. "Pattern Descriptors Orientation and MAP Firefly Algorithm Based Brain Pathology
Classification Using Hybridized Machine Learning Algorithm." IEEE Access 10 (2021): 3848-3863.
9) Helaly, Hadeer A., Mahmoud Badawy, and Amira Y. Haikal. "Deep learning approach for early detection of
Alzheimer’s disease." Cognitive computation (2021): 1-17.
23. Acknowledgement
22
The authors would like to acknowledge the support provided by the Deanship of Research Oversight
and Coordination (DROC) at King Fahd University of Petroleum and Minerals (KFUPM) for funding this
work through project No. EC221016 and partial support from Centre for Smart Mobility and Logistics at
King Fahd University of Petroleum and Minerals. We would also like to thank Dr. Shahabuddin Siddiqui,
Diagnostic Radiology, Pakistan Institute of Medical Sciences Hospital, Islamabad, and Dr. Danesh Kumar
Kella, Cardiologist, Piedmont Healthcare, Atlanta Georgia USA for their valuable input in this research.