Ocu-Hb Vision: Non-Invasive AI-Based Anaemia Screening at IIT Madras

Ocu-HB vision : A non-invasive AI-Based Tool for Anaemia Screening

Objective

• To develop, validate, and deploy a smartphone-based application capable of estimating haemoglobin levels from eye images, focused on screening and triage rather than diagnosing the cause of anaemia, To collect and analyse approximately 8,000 labeled eye images and short video clips from hospital and community settings over three years, covering a haemoglobin range of 7 to 18 g/dL, To develop an AI model capable of estimating haemoglobin with a mean absolute error below 1 g/dL under real-world conditions, To ensure that the application works offline, allowing use in areas with limited internet connectivity, To enable community health workers to perform quick anaemia screening in less than ten seconds per individual

Description

• The primary objective is to develop an AI model capable of estimating haemoglobin with a mean absolute error below 1 g/dL under real-world conditions. A secondary objective is to ensure that the application works offline, allowing use in areas with limited internet connectivity. Ultimately, the goal is to enable community health workers to perform quick anaemia screening in less than ten seconds per individual.

Impact

• A validated smartphone application for non-invasive anaemia screening, ready for integration into community health screening programs, A curated dataset of approximately 8,000 labeled eye images to support further research, alongside documented clinical validation results submitted for peer-reviewed publication, Significantly strengthened early anaemia detection, with a non-invasive method reducing hesitation and increasing participation in screening programs and allowing earlier nutritional intervention and medical referral, A triage mechanism that reduces unnecessary laboratory load while ensuring high-risk individuals are not missed, contributing to national anaemia control initiatives, Democratized access to basic screening in communities where laboratory infrastructure is limited, with the same platform adaptable to other blood-related screening applications, Approximately 15,000 to 20,000 individuals screened directly during the three-year project period, with potential annual reach extending to several lakh individuals across multiple districts