YOLO-DRIVEN GLAUCOMA SCREENING MODEL FOR PORTABLE MEDICAL SCREENING DEVICES

Abstract

Glaucoma is a leading cause of irreversible blindness, and its early detection relies heavily on structural assessment of the optic nerve head, in particular the Cup-to-Disc Ratio (CDR) and Vertical Cup-to-Disc Ratio (VCDR). This work proposes a lightweight, YOLO-based computer-aided diagnosis (CAD) pipeline that automatically segments the optic disc and optic cup from colour fundus images, estimates CDR and VCDR. A dataset of 650 fundus images with disc–cup annotations was used to train a segmentation model with a 70/20/10 split for training, validation, and testing. On the validation set, the model achieved a precision of 98.98%, a recall of 99.23%, and a mean Average Precision (mAP@0.5) of 99.37%. For testing results, the model generated CDR and VCDR for each test image and mapped them to Normal, Suspect, or Glaucoma using simple rule-based thresholds, also hardware profiling showed an average end-to-end processing time of 173 ms per image and a total RAM footprint of ≈174 MB. The proposed Glaucoma Screening pipeline gives high segmentation accuracy, transparent CDR and VCDR staging, and a very small hardware requirement, which make it suitable for early glaucoma screening in portable screening devices.