Medical image classification with YOLO to speed up diagnosis

The challenge

Image volume grew faster than the team's capacity to review it in the right order: urgent cases waited in the queue alongside routine ones.

The objective

Triage and pre-classify images automatically so the team looks first at what matters, without taking the professional out of the decision.

The solution

Training of a YOLO model on data labelled by the institution itself, an inference pipeline integrated into the existing workflow, marking of regions of interest on the image and a queue prioritized by classification. Precision and recall tracked per class; every output reviewed by a professional.

The impact

• Exam queue prioritized by automatic classification, with regions of interest highlighted
• Model trained on the institution's own data with per-class metrics tracked
• Integrated into the existing workflow, without replacing the systems in use
• Human-in-the-loop: no report is issued without professional review