BloodBox: A Diagnostic Tool For Blood-Borne Diseases

The Project

BloodBox is a diagnostic tool for blood borne diseases which takes a magnified image of blood cells using a raspberry pi and camera. The image is then processed by a Tensorflow Lite machine learning model on the pi. We use image classification models to detect malaria and leukaemia in individual cells. We also use an object detection model based on the YOLO Framework to return a Complete Blood Cell Count report. We 3D printed parts for this device ourselves.

View this project on YouTube

Team Comments

We were fascinated by the development of a paper microscope called the foldscope, but still realised that this device had limitations, namely that a doctor was required. To eliminate this flaw, we decided to use machine learning to fully automate the process of diagnosis.

We found it challenging to obtain a clear image with our microscopic lens. This occured because our camera was very light sensitive, and we were using low quality cow blood. To solve this, we conducted many iterations of our design, used a better camera, and we were able to obtain a clear image.

All the machine learning models work individually and the classification models work on the pi. If we had more time we would have liked to have tested the Object detection model on the pi as well.