YOLOv8 Medical Imaging

Project overview

Classification, detection and segmentation on medical scans in one Streamlit app: COVID and pneumonia from chest images, and blood cell counts from microscopy.

Classification, detection and segmentation are usually three separate projects with three separate codebases. YOLOv8 ships all three heads on one backbone, which means the same training loop, the same data layout and the same inference call cover all of them. This app is that idea applied to medical images: pick a task in the sidebar, upload a scan, get the annotated result back.

Each task has its own dataset and its own trained model. Chest images are classified into COVID-19, viral pneumonia or normal; blood cell microscopy is run through detection to find and count red cells, white cells and platelets; and breast ultrasound is segmented rather than boxed, because with a mass the shape is the diagnostic part. All three were trained from the nano configs for 100 epochs, and each run keeps its arguments, its per-epoch metrics and its checkpoints, so any result can be traced back to the run that produced it.