How sifatahammed/diabetic-retinopathy-classification-using-deep-learning works
This repository is a notebook-based research project comparing deep-learning approaches to diabetic-retinopathy severity classification from fundus images. Its documented workflow starts with the Kaggle dataset, uses a preprocessing notebook, then evaluates five model experiments using accuracy and Cohen’s kappa. The README identifies the models and reported results, but notebook contents were not sampled, so implementation details and precise call or data-transfer wiring cannot be verified; the diagram shows only the documented high-level workflow.
Main entry points: Dataset_Preprocessing.ipynb Dataset preprocessing · Experiments Accuracy and kappa · 3. CNN.ipynb Custom CNN · 2. InceptionV3.ipynb InceptionV3
Components 8
Data preparation
- Dataset preprocessing
Dataset_Preprocessing.ipynb
Model experiments
- Custom CNN
Experiments/3. CNN.ipynb - InceptionV3
Experiments/2. InceptionV3.ipynb - ResNet50
Experiments/4. ResNet50.ipynb - DenseNet121
Experiments/1. DenseNet121.ipynb - EfficientNetB6
Experiments/5. EfficientNet.ipynb
Evaluation
- Accuracy and kappa
Experiments
Other components
- Kaggle DR dataset
Connections 11
- Kaggle DR dataset → Dataset preprocessing: provides images
- Dataset preprocessing → Custom CNN: prepares data
- Dataset preprocessing → InceptionV3: prepares data
- Dataset preprocessing → ResNet50: prepares data
- Dataset preprocessing → DenseNet121: prepares data
- Dataset preprocessing → EfficientNetB6: prepares data
- Custom CNN → Accuracy and kappa: evaluates with
- InceptionV3 → Accuracy and kappa: evaluates with
- ResNet50 → Accuracy and kappa: evaluates with
- DenseNet121 → Accuracy and kappa: evaluates with
- EfficientNetB6 → Accuracy and kappa: evaluates with