metadata

tags:
  - image-classification
  - medical-imaging
  - cataract
  - resnet18
  - pytorch
  - interpretability
license: mit
datasets:
  - other
language: en

Cataract Detection ResNet18 Model

Trained on ~1170 clinical images (binary: cataract vs non-cataract).

Architecture: ResNet18
Image Size: 224x224
Train Accuracy: 100%
Test Accuracy: 97.16%
Use: Baseline model for severity grading project

Future Steps

Grad-CAM to interpret focus region (pupil/lens)
Region-based unsupervised feature extraction
Clustering for cataract severity grading