Aletheia: Offline AI Diagnostic Tool for Sub-Saharan Africa
Aletheia is a new clinical decision support system designed to address the shortage of specialist knowledge in sub-Saharan Africa, where the ratio of doctors to patients in rural areas can be as extreme as 1:25,000. Unlike existing AI diagnostic tools that require reliable internet and high-tech equipment, Aletheia is specifically built for offline use in resource-limited settings like district hospitals. It uses the Qwen2.5-3B-Instruct model, enhanced through Quantised Low-Rank Adaptation (QLoRA), and was trained on a collection of 27,000 clinical reasoning examples covering 50 common diseases in East Africa. Its evaluation shows an impressive Top-1 diagnostic accuracy of 80.0% and a perfect Top-3 accuracy of 100.0%, aiding healthcare workers in making differential diagnoses without needing constant internet access.
Key facts
- Aletheia is an offline-first clinical decision support system.
- It targets low-resource healthcare settings in sub-Saharan Africa.
- Physician-to-patient ratios can fall below 1:25,000 in rural areas.
- Built on Qwen2.5-3B-Instruct, fine-tuned with QLoRA.
- Trained on 27,000 clinical reasoning samples for 50 disease conditions.
- Diseases are those with elevated prevalence in East Africa.
- Top-1 diagnostic accuracy is 80.0%, Top-3 accuracy is 100.0%.
- BERTScore-F1 is 0.909, METEOR is 0.467.
Entities
Locations
- sub-Saharan Africa
- East Africa