How sanjaykris20/pathological-myopia works
This repository documents a research pipeline for analyzing pathological myopia in fundus photographs. Public datasets are harmonized, quality-checked, split by group, and preprocessed before a six-model classification benchmark and validation-based selection. The selected Swin-Tiny backbone supports classification, severity, fovea localization, and annotation-dependent segmentation; outputs also support quantitative phenotyping and explainability, followed by final evaluation. The tree contains only a notebook and README, with no source excerpts, so the graph reflects documented workflow relationships rather than inspected implementation calls.
Main entry points: pathological-myopia.ipynb Swin-Tiny encoder
Components 16
Data preparation
- Dataset harmonization
pathological-myopia.ipynb - Quality control
pathological-myopia.ipynb - Group-aware split
pathological-myopia.ipynb - Retinal preprocessing
pathological-myopia.ipynb
Model benchmarking
- Six-model benchmark
pathological-myopia.ipynb - Validation selection
pathological-myopia.ipynb - Controlled optimization
pathological-myopia.ipynb
Multi-task analysis
- Swin-Tiny encoder
pathological-myopia.ipynb - PM classification
pathological-myopia.ipynb - Severity assessment
pathological-myopia.ipynb - Fovea localization
pathological-myopia.ipynb - Lesion segmentation
pathological-myopia.ipynb
Interpretation
- Quantitative phenotyping
pathological-myopia.ipynb - Grad-CAM analysis
pathological-myopia.ipynb - Final evaluation
pathological-myopia.ipynb
Other components
- Public fundus datasets
Connections 20
- Public fundus datasets → Dataset harmonization: harmonizes
README.md - Dataset harmonization → Quality control: quality checks
README.md - Quality control → Group-aware split: splits groups
README.md - Group-aware split → Retinal preprocessing: prepares images
README.md - Retinal preprocessing → Six-model benchmark: feeds benchmark
README.md - Six-model benchmark → Validation selection: reports validation
README.md - Validation selection → Controlled optimization: selects benchmark
README.md - Validation selection → Swin-Tiny encoder: selects backbone
README.md - Retinal preprocessing → Swin-Tiny encoder: feeds images
README.md - Swin-Tiny encoder → PM classification: shares features
README.md - Swin-Tiny encoder → Severity assessment: shares features
README.md - Swin-Tiny encoder → Fovea localization: shares features
README.md - Swin-Tiny encoder → Lesion segmentation: shares features
README.md - Lesion segmentation → Quantitative phenotyping: supplies masks
README.md - Lesion segmentation → Grad-CAM analysis: supports mask analysis
README.md - PM classification → Final evaluation: evaluated in
README.md - Severity assessment → Final evaluation: evaluated in
README.md - Fovea localization → Final evaluation: evaluated in
README.md - Quantitative phenotyping → Final evaluation: reported in
README.md - Grad-CAM analysis → Final evaluation: reported in
README.md