Low-Cost Vision Screening for Anaemia and Breast-Cancer Risk

Problem

Cost, infrastructure, and geographic access can restrict use of conventional diagnostic methods. These projects explored whether accessible image modalities could support earlier, lower-cost screening research.

Approach

1. 1.Trained YOLOv8 to detect conjunctival images and paired detection with a supervised classifier for anaemic versus non-anaemic image classification.
2. 2.Evaluated pretrained VGG16, DenseNet, and ResNet architectures for breast-cancer risk prediction from infrared thermographic images.
3. 3.Framed both systems as research screening tools that require clinical validation and should complement, not replace, established diagnostic pathways.

Learnings

• Equitable clinical AI requires designing around real constraints in access, equipment, and infrastructure.
• Medical-image prototypes require rigorous external and clinical validation before any diagnostic use.