ROP Classifier

Clinical retinal images vary a lot in quality, and a classifier trained on raw images does not hold up. The baseline needed to be beaten on the images that actually matter.

I built image preprocessing and segmentation pipelines ahead of the classifier so the model trained on cleaned, region-focused data, then measured against the baseline.

• Image preprocessing pipeline for variable-quality retinal images
• Segmentation pipeline isolating the relevant regions
• Classification accuracy improved over the baseline

Accuracy improved over the baseline through the preprocessing and segmentation stages.