This paper adopted a Deep Residual Convolutional Neural Network (ResNet50) architecture for the experiments amongst other discriminative learning techniques to train the model. Using the novel dataset and two publicly available MRI brain datasets, this proposed approach attained a classification accuracy of 86.40\% on the proposed dataset, 93.80% on the Harvard Whole Brain Atlas, and 97.05% accuracy on the School of Biomedical Engineering dataset. Our experimental results significantly demonstrate our proposed framework for Transfer Learning is a potential and effective approach for brain tumour multi-classification tasks.

1. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Deep Transfer Learning for Magnetic Resonance
Image Multi-class Classification
Yusuf Brima
brima.yusuf@aims.ac.rw
Supervised by
Professor Ernest Fokoué
epfeqa@rit.edu
February 9, 2021

2. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Outline
1 Citation
2 Problem
3 Literature Survey
4 Research Goal
5 Proposed Methodology
6 Dataset
7 Experimental Results
8 Conclusion
9 References

3. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Citation
Brima Y., Tushar M.,Kabir U, and Islam T, “Deep Transfer Learning for
Multi-class Brain MRI Classification”, November 2020, in submission to
IEEE/ACM-TCBB

4. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Problem
I Radiologists analyse brain MRI scans

5. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Problem
I Radiologists analyse brain MRI scans
I Painstakingly difficult and time-consuming

6. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Problem
I Radiologists analyse brain MRI scans
I Painstakingly difficult and time-consuming
I Lack of expertise in low-resource societies

7. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
State-of-the-art
I Chaplot et al 2006 [1], Classification of magnetic resonance brain
images using wavelets as input to Support Vector Machine and
Neural Network
I Limitation: Manual feature engineering

8. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
State-of-the-art
I Talo et al 2019 [2], Application of Deep Transfer Learning for
automated brain abnormality classification using MR images
I Limitation: Binary Classifier

9. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Research Goal
I Deep Transfer Learning
I Using
• National Institute of Neuroscience and Hospital (NINS)
• The Harvard Whole Brain Atlas

10. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
Formal Definition
D := {X , P(X)}
X = {x1, x2, x3, . . . , xn}, ∀xi ∈ X

11. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
Formal Definition
For domain D, a task is defined as:
T := {Y , P(Y |X)}
Y = {y1, y2, y3, . . . , yn}, ∀yi ∈ Y

12. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
Formal Definition
Therefore
DS := {XS , P(XS )}
XS = {xS1
, xS2
, xS3
, . . . , xSn
}, ∀xSi
∈ XS
YS = {yS1
, yS2
, yS3
, . . . , ySn
}, ∀ySi
∈ YS
DT := {XT , P(XT )}
XT = {xT1
, xT2
, xT3
, . . . , xTn
}, ∀xTi
∈ XT
YT = {yT1
, yT2
, yT3
, . . . , yTn
}, ∀yTi
∈ YT

13. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
The goal of Transfer Learning
Given
η : X ,→ Y where η ∈ H
∼
X = argmin
η∈H
{L (η(XSi
) 6= YSi
)}
And
RDT
:= P(η(XT ) 6= yT |
∼
X)
η∗
= argmin
η∈H
{RDT
(η(XT ), YT ,
∼
X)}

14. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
Figure 1: High-level formal representation of Transfer Learning

15. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
Figure 2: Proposed System Architecture

16. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Transfer Learning
Figure 3: Deep Transfer Learning Stages

17. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Dataset
Category Total Patients Total Slices
Normal 2 65
Degenerative Disease 8 223
Neoplastic Disease 8 277
Inflammatory Infectious Disease 5 189
Cerebrovascular Disease 15 376
Table 1: Harvard Whole Brain Atlas dataset contains 1133 T2-weighted
contrast-enhanced images

18. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Dataset
Figure 4: Sample Harvard Whole Brain MRI Dataset

19. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Experimental Results
Figure 5: Illustration of top miss-classified images after stage III of fine-tuning

20. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Experimental Results
Figure 6: Error Rates across the four model architectures

21. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Experimental Results
Figure 7: Accuracy across four different model architectures

22. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Experimental Results
Figure 8: Stage III confusion matrix

23. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
Conclusion
Figure 9: Proposed system deployment diagram

24. References
[1] S. Chaplot, L. Patnaik, and N. Jagannathan, “Classification of
magnetic resonance brain images using wavelets as input to support
vector machine and neural network,” Biomedical signal processing
and control, vol. 1, no. 1, pp. 86–92, 2006.
[2] M. Talo, U. B. Baloglu, and U. R. Acharya, “Application of deep
transfer learning for automated brain abnormality classification using
mr images,” Cognitive Systems Research, vol. 54, pp. 176–188,
2019.

25. Citation Problem Literature Survey Research Goal Proposed Methodology Dataset Experimental Results Conclusion References
End
Thank You!