Smart Microscopy Camera Kit: Automatic Counting of Blood Cells in Peripheral Blood Smear Images Using RetinaNet on Raspberry Pi CM3+

Abstract

Microscopic examination of peripheral blood smear images for blood cell counting remains a critical yet labor-intensive task in clinical diagnostics. This research presents MicrosisDCN, an intelligent microscopy camera system designed to automate blood cell detection and counting, powered by a compact embedded platform based on the Raspberry Pi Compute Module 3+. The system incorporates a 5-megapixel image sensor and a versatile eyepiece fitting that is compatible with the most compound microscopes, providing a portable, cost-effective, and user-friendly solution. Calibration procedures ensure alignment with traditional high-power field (HPF) standards, allowing cell counts to be reported in standard mitotic count units. To detect red blood cells, white blood cells, and platelets in real-time, the system uses a special version of a deep learning model called RetinaNet, which has been improved with a technique called auto-anchor parameterization. MicrosisDCN achieves a mean Average Precision (mAP) of 86.81% in detecting a few types of blood cells with minimal errors: 1.06% for red blood cells, 0.06% for white blood cells, and 4.23% for platelets. The results indicate that MicrosisDCN, which combines traditional microscopy with advanced vision technologies, serves as an efficient, practical, and scalable solution for clinical and medical laboratory applications.