aptos2019-blindness-detection

APTOS 2019 Blindness Detection

Prerequisites: Ensure you have installed mlebench following instructions.

Given a dataset of retinal images, predict the severity level of diabetic retinopathy on a scale of 0 to 4. This is a multi-class classification task with medical significance.

Dataset Overview

The APTOS 2019 Blindness Detection dataset contains high-resolution retinal images, classified into five severity levels:

• 0: No diabetic retinopathy
• 1: Mild diabetic retinopathy
• 2: Moderate diabetic retinopathy
• 3: Severe diabetic retinopathy
• 4: Proliferative diabetic retinopathy

This challenge addresses a critical healthcare issue as millions of people suffer from diabetic retinopathy, the leading cause of blindness among working-aged adults. Early detection can prevent blindness.

Download Dataset

mlebench prepare -c aptos2019-blindness-detection

Run Curie

• Update the configuration: Open curie/configs/mle-aptos-config.json and verify the paths to the dataset and starter code.
• Execute Curie:

cd Curie/
python3 -m curie.main -f benchmark/mle_bench/aptos2019-blindness-detection/aptos2019-blindness-detection-question.txt --task_config curie/configs/mle_config.json --dataset_dir /home/amberljc/.cache/mle-bench/data/aptos2019-blindness-detection/prepared/public 

• Change --dataset_dir to the absolute path to your dataset.

Curie Results

After asking Curie to solve this question, the following output files are generated:

• Report: Auto-generated report with experiment design and findings
• Experiment results: All detailed results for all conducted experiments
• Curie logs: Execution log file
• Curie workspace: Generated code, complete script to reproduce and raw results (we didn't upload the model checkpoint).

Curie Performance Summary

Data Understanding and Preprocessing

We used the APTOS 2019 Kaggle dataset, which consists of 3,662 high-resolution retinal images (3,295 for training and 367 for testing). The dataset exhibits significant class imbalance:

• Class 0 (No DR): 1,628 images (49.41%)
• Class 1 (Mild): 340 images (10.32%)
• Class 2 (Moderate): 896 images (27.19%)
• Class 3 (Severe): 176 images (5.34%)
• Class 4 (Proliferative): 255 images (7.74%)

Multiple preprocessing techniques were implemented and evaluated:

1. Basic Normalization: Standard image normalization using ImageNet mean and standard deviation values
2. CLAHE Enhancement: Contrast Limited Adaptive Histogram Equalization to improve visibility of retinal features
3. Circular Crop: Removing irrelevant black borders around the retinal image

All images were resized to the input dimensions required by each model architecture (224×224 pixels for most models).

Model Performance

• Overall Performance

The EfficientNetB4 model with CLAHE preprocessing achieved the best performance:

• Validation Quadratic Weighted Kappa: 0.9096
• Validation Accuracy: 0.8376
• Best performance at epoch 8 (of 18 total epochs)
• Total training time: 113 minutes

• 3.2 Per-class Performance

The model demonstrated varying accuracy across different DR severity grades:

• Class 0 (No DR): 98.78% accuracy
• Class 1 (Mild): 66.15% accuracy
• Class 2 (Moderate): 76.88% accuracy
• Class 3 (Severe): 46.88% accuracy
• Class 4 (Proliferative): 56.25% accuracy

For complete details on methodology, experiments, and analysis, refer to the generated report