National ICT Awards (NBQSA) - Silver Award

The Problem

β-thalassemia is one of the most common inherited blood disorders in Sri Lanka and South Asia. Identifying carriers early - before they have an affected child - relies on HPLC testing, which is accurate but relatively costly and not available at every primary care facility, especially in resource-limited settings. Most people never get tested until a family already has an affected child.

The Solution

• Hematological Screening - an MLP model estimates carrier probability directly from seven routine FBC indices (RBC, Hb, MCV, MCH, MCHC, RDW), values already produced by any hospital's automated haematology analyser.
• AI-Assisted Image Analysis - for cases the FBC model can't confidently clear, a CNN analyses a smartphone-captured peripheral blood smear image for the microscopic features of thalassemia trait, without needing a haematologist to manually review the slide.
• Screening Decision Support - the two models run sequentially: FBC first, smear image only when needed, and only patients flagged by either step are referred onward for confirmatory HPLC testing - cutting the number of smear reviews and HPLC referrals needed for the same coverage.

Technology

• Machine Learning - a Multi-Layer Perceptron (MLP) trained on red-cell indices for the hematological screening stage.
• Deep Learning - a Convolutional Neural Network (CNN) trained on smartphone-captured peripheral blood smear images for the image-analysis stage.
• Data & AI - trained and validated on a real clinical dataset with an independent, held-out test cohort to check generalisation.
• Platform - a lightweight web application (Flask) so the tool can run on ordinary laboratory hardware without specialised equipment.

Proof of Concept

Trained and evaluated on data from 150+ individuals and 5,000+ smear images, with performance verified on an independent test cohort. The RBC-indices model alone reached 88.2% sensitivity and 92.9% specificity; the image-analysis model reached 88.2% sensitivity and 64.3% specificity. Run sequentially, the combined pipeline reached 100% sensitivity on the independent test cohort, while cutting the need for blood smear preparation by 37.7% - meaning no true carriers were missed, at a substantially lower workload than smear review for every sample. As with any diagnostic-support tool, further prospective clinical validation across larger, more diverse cohorts is required before wider clinical deployment.

What Makes ThalScan Different?

• Dual-modal by design - most tools use either blood indices or images, not both in a single decision-support pipeline.
• Built for existing laboratory infrastructure - no new hardware; it works with the FBC analysers and light microscopes labs already have, plus an ordinary smartphone camera.
• Intelligent, sequential screening - the smear image step only runs when the cheaper FBC step can't confidently clear a sample, saving time and cost at scale.
• Designed for population screening - aimed at high-prevalence, resource-limited settings like Sri Lanka, where mass carrier screening is the most effective way to reduce new cases of thalassemia major.

Research & Scientific Recognition

• Published in BMC Medical Informatics and Decision Making (Springer Nature, Q1 journal, 2026).
• Presented at the National Health Research Symposium (NHRS) 2025, Ministry of Health, Sri Lanka.

Towards Real-World Validation

Early discussions are underway to pilot ThalScan in real clinical settings, including with the National Thalassaemia Centre, Teaching Hospital Kurunegala, and the Faculty of Medicine, University of Peradeniya - the next step toward the prospective clinical validation the tool needs before wider deployment.

Potential Impact

By making carrier screening cheaper and less dependent on specialist smear review, ThalScan could extend the reach of thalassemia screening programmes in Sri Lanka and similar high-prevalence settings - helping more couples learn their carrier status before starting a family, and reducing the number of children born with thalassemia major.

My Role

Researcher & Developer - I designed and built both the hematological (MLP) and image-analysis (CNN) models, the sequential screening pipeline, and the web application, over a development period from May 2024 to June 2026, under the supervision of Dr. C.R. Wijesinghe at the University of Colombo School of Computing.