Automated Malaria Detection | Case Study | Elgen Mar Arinasa

A robust computer vision pipeline designed for clinical deployment to assist pathologists in rural areas in detecting malaria parasites from thin blood smear images.

01. The Problem

Manual inspection of blood smears is highly time-consuming, prone to human error, and suffers from a lack of trained pathologists in endemic areas.

02. Architecture & Solution

We developed a dual-stage architecture. First, a YOLOv11 model isolates and crops individual Red Blood Cells. Second, a stacked Convolutional Neural Network (CNN) classifies each cell as Infected or Uninfected. To build medical trust, we integrated Grad-CAM, producing heatmaps that highlight the exact morphological features the CNN used to make its decision.

Interactive Deep-Dive

Ensemble CNN Pipeline

01. RAW INPUTHigh-res microscopic thin smear (Giemsa stained)

03. Impact

Achieved high accuracy on test datasets while retaining high interpretability for medical professionals.