B-Yassine88/PathMNIST-XAI-Lightweight-Explainable-CNN-for-Medical-Imaging
🔍 Enhance medical imaging with a lightweight CNN model that offers over 91% accuracy and integrated explainability for better clinical trust.
What's novel
🔍 Enhance medical imaging with a lightweight CNN model that offers over 91% accuracy and integrated explainability for better clinical trust.
Code Analysis
2 files read · 2 roundsA single monolithic Python script that trains a small CNN on the PathMNIST pathology dataset, evaluates it, runs Integrated Gradients on 500 test samples, and stores attribution maps in a SQLite database.
Strengths
The training loop, model definition, and Integrated Gradients integration are functionally correct and demonstrate a working end-to-end pipeline for explainable medical image classification. The use of Captum for attribution and SQLite for persistence, while simple, shows a coherent (if minimal) workflow.
Weaknesses
The README describes a full GUI application with installers, upload buttons, and cross-platform deployment — none of which exist in the code. The entire project is one unstructured script with no modularity, no tests, no dependency management, no error handling, and no separation of training/inference/explainability concerns.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
13
Contributors
2
Files
8
Active weeks
5
Repository
Language
Python
Stars
1
Forks
1
License
MIT