Project 01 / 06

CDIB

Leukemia detection from blood-cell images

A CNN that detects leukemia in blood-cell images. Quantization cut the model size by 83% without losing accuracy, which makes it realistic for edge devices. Experiments are tracked in MLflow and the model is served through FastAPI in Docker.

Live demo View code

Results

What it achieved

test accuracy: 98.53%
myeloblast recall: 100%
model size reduction: 83%

Built with

PyTorch
CNN
Quantization
MLflow
FastAPI
Docker