Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 26

AUTOMATED DETECTION

OF ABNORMALITIES IN MEDICAL IMAGING

FYP FINAL EVALUATION


Team Members Supervisor
Ali Iqbal Khan Ms. Sumaiyah Zahid
Ahmad Azaan Farooqui Co-Supervisor
Mujtaba Ahmed Dr. Muhammad Farrukh Shahid 1
Problem Statement

"Addressing the urgent need for automated disease


detection and in-depth medical imaging analysis
through cutting-edge AI techniques."

2
INTRODUCTION
In our Research Project, we're delving into the transformative potential of advanced AI
models such as LLM and CNN to revolutionize the diagnosis of chest X-ray, MRI, and CT
scan abnormalities. Our aim is to drive improved healthcare outcomes through enhanced
diagnostic processes.

● Development of a CNN Model for Automated Abnormality Detection

● Performance Evaluation of CNN in Medical Imaging Abnormality Detection

● NLP-Powered Medical Reports from CNN Model Outcomes

● Integrating AI for Enhanced Imaging Analysis and Diagnoses

3
Presentation Roadmap

01 02 03 04
Literature Methodology Market Timeline
Review Scope
Base Papers and dataset System Diagram Potential Opportunities Gantt Chart and Work
details and Project Workflow Distribution

4
LITERATURE REVIEW

5
[1] ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-
Supervised Classification and Localization of Common Thorax Diseases (Wang et
al. ,2017)

Advantages:
-Large-Scale Chest X-ray Database. Future work:
-Weakly-Supervised Multi-Label Classification. -Expansion of Disease Classes in ChestX-ray8
-Integration with Patient History Data
Limitations: -Enhanced Radiological Report Generation
-Challenges in Disease Localization.
-Unbalanced Data Distribution Methodology:
-Labeling Disease Names by Text Mining
Target: -Quality Control on Disease Labeling
-Address the challenge of utilizing the vast amount -Common Thoracic Disease Detection and
of chest X-ray data for the development of highly Localization
precise CAD systems using DL methods. -Unified DCNN Framework

Main findings: Data Set:


-Quantitative Performance on Pathology Dataset has a size of 45GB with 112,000 Chest
Classification. X-ray images labeled with disease from more
-ResNet-50 with highest accuracy among other than 30,000 unique patients
models(AlexNet,GoogLeNet,VGGNet-16 ).

6
Fig.1 The overall flow-chart of our unified DCNN framework and disease localization process.

7
[3] CovidExpert: A Triplet Siamese Neural Network framework for the detection of
COVID-19 (CT-Scans) (Ornob et al. ,2023).

Advantages: Future Work:


-Efficient Multi-Class Classification - Reduce Training Dataset Size
-Enhanced Prediction Diversity -Include Different CT Views(sagittal, coronal)
-Outperforms CNN Models -Incorporate Patient Data
-High Medical Applicability
Methodology:
Limitations: - Data Preprocessing
-Lengthy Training Time - Train on pre-trained CNN models
-Limited to Axial CT Views - Triplet Siamese Network
-Ignores Patient Data - Evaluation Metrics
- Analysis
Target:
-Expedite COVID-19 Diagnosis Data Set:
-High Accuracy in Classification The dataset comprises 7 public datasets with a
total size of 2GB, containing 6893 non covid,
Main Findings: 7593 covid and 2618 CAP pictures.
- ROC score of 99.9%, specificity of 99.36%,
sensitivity of 98.72%, and overall accuracy of
98.719%.
- Proposed model significantly improves overall
performance and stability. 8
Fig. 4 Proposed Model

9
[2] Computer-Aided Brain Tumor Diagnosis: Performance Evaluation of Deep Learner
CNN using Augmented Brain MRI (Asma Naseer et al. ,2021).

Advantages: Future Work:


- Pioneering CNN Approach for Brain Tumor - Explore Alternative CNN Architectures
Diagnosis - Enhance Data Augmentation
- High Accuracy with Data Augmentation and - Incorporate Additional Datasets
Regularization - Continual Dataset Updates

Limitations: Methodology:
- Small Training Dataset's Impact - Data Preprocessing
- Performance on Limited Sample Datasets - Gradual CNN Model Training
- Data Augmentation Techniques
Target: - CNN Architecture Components
- Computer-Aided MRI-Based Brain Tumor
Diagnosis Data Set:
The dataset comprises 7 datasets with a total
Main Findings: size of 1307MB, containing a collection of
- Consistent Accuracy Rates > 90% 27,733 images
- Outperforms Existing Systems

10
Fig.2 Models Implemented in paper

Fig.3 Proposed CNN


11
Literature Review

12
METHODOLOGY

13
14
Overall System Architecture
Medical Image Data Set

Data Preprocessing
of Image & Text

CNN(Abnormality Transformer
Detection) base NLP

Combine CNN &


Transformer

Frontend Interface
(HTML/CSS,Django)

User Interaction
(View reports & results)

15
16
Models Used
Labels Proposed Models in Implemented Models
Papers

Adaptive Model - EfficientNetB0 with


Transfer Learning

Chest X-Ray AlexNet,GoogleNet,VG Resnet50 with Transfer


G-16 and Resnet50 Learning

Brain MRI Authors Proposed EfficientNetB0 with


CNN Transfer Learning

CT-Scan Proposed Ensemble ResNet50 with


Model Transfer Learning

17
Accuracy Analysis
99.00% 98.01% 98.81% 98.72%
100% 96.49%

90%
81.41%
80%
71.00%
70%

60%

50%

40%

30%

20%

10%

0%
Adaptive Model Brain MRI Lungs CT-Scan Chest X-Ray

Our Model Base Paper Model


18
Limitations in Our Project
Data
Imbalance

Repeated
Limited
Data
Data
Repositories

Limitations

No Error
Handling
Lack of Medical yet
Supervision
Not enough
Computational
resources

19
01. and Results
Testing
Enter a subtitle here if you need it

20
Image Uploading

21
Image Result

22
TIMELINE
FYP - I
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15

Project Scope and


Objectives

Gather datasets and


research papers

Literature Review &


proposal

Research on Base
Papers

FYP-1 Defense
Presentation

Django Environment
Setup & Frontend

Testing CNN Model

SRS/SDS
documentation

Final Defense
Preparation 23
TIMELINE
FYP - II
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 Week 11 Week 12 Week 13 Week 14 Week 15

Testing on other CNN


models

Start NLP Model


Development for
Reports
Data Augmentation (if
needed)

Integration of CNN
and NLP models

Testing on integrated
models on frontend

Complete Frontend
development

Complete Backend
development

QA Testing

Final Documentation
and Report 24
Tools And Technologies

25
REFERENCES
[1]Wang et al. (2017) ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and
Localization of Common Thorax Diseases. IEEE CVPR, 1-10
(https://openaccess.thecvf.com/content_cvpr_2017/papers/Wang_ChestX-ray8_Hospital-Scale_Chest_CVPR_2017_paper.pdf)

[2]Naseer et al. (2021) Computer-Aided Brain Tumor Diagnosis: Performance Evaluation of Deep Learner CNN Using Augmented
Brain MRI. International Journal of biomedical Imaging (https://www.hindawi.com/journals/ijbi/2021/5513500/)

[3]Ornob et al. (2023) CovidExpert: A Triplet Siamese Neural Network framework for the detection of COVID-19. Informatics in
Medicine Unlocked (https://arxiv.org/ftp/arxiv/papers/2302/2302.09004.pdf)

Thank You
26

You might also like