TLDR: AmarDoctor is a groundbreaking AI-driven, multilingual, voice-interactive digital health application designed to bridge the healthcare gap for Bengali speakers. It offers a dual-interface for patients and providers, featuring an AI assistant for symptom assessment and clinical decision support. Evaluated against human physicians, AmarDoctor demonstrated significantly higher accuracy in provisional diagnoses (81.08% vs. 50.27%) and specialty recommendations (91.35% vs. 62.6%), showcasing its potential to enhance primary care access and patient management in underserved communities.
In an increasingly digitized world, access to advanced healthcare solutions remains uneven, particularly for non-European populations. A significant gap exists for the 270 million Bengali speakers across Bangladesh, India, and the Gulf region, who often lack adequate digital health resources. Bangladesh, for instance, faces a severe shortage of primary care doctors, with only 9.9 physicians per 10,000 patients, a situation exacerbated in rural areas. This digital health divide is further widened by low digital literacy, a lack of support for non-European languages in existing platforms, and AI systems often trained on data predominantly from white European patients, leading to reduced accuracy for diverse ethnolinguistic backgrounds.
Addressing this critical need, a new digital health application called AmarDoctor has emerged. This AI-driven, multilingual, voice-interactive platform is designed to revolutionize primary care triage and patient management specifically for Bengali speakers. It aims to bridge the existing digital health divide by offering accessible and inclusive healthcare solutions.
How AmarDoctor Works
AmarDoctor features a dual-interface system catering to both patients and healthcare providers. It supports three major Bengali dialects (Standard, Sylheti, and Chittagonian) alongside English, ensuring broad linguistic accessibility. At its core, the patient module utilizes an adaptive questioning algorithm to assess symptoms and guide users to the appropriate medical specialist. To overcome digital literacy barriers, it integrates a voice-interactive AI assistant named “Aisha,” which helps users navigate the app’s services seamlessly.
For healthcare providers, the clinician-facing interface offers AI-powered decision support. This enhances workflow efficiency by generating structured provisional diagnoses and treatment recommendations, which are presented in a comprehensive Subjective, Objective, Assessment, and Plan (SOAP) note format. These outputs facilitate key services such as e-prescriptions, video consultations, and medical record management.
The foundation of AmarDoctor is a meticulously constructed health knowledge graph. This graph models the intricate relationships among diseases, symptoms, drugs, and diagnostic tests. It was developed by integrating insights from 1.4 million anonymized patient clinic visit records and proprietary patient data from pilot studies in Bangladesh. Crucially, this knowledge graph was refined by a panel of expert clinicians to address potential biases and accurately reflect symptom prevalence and manifestation in the South Asian patient population.
The app’s multilingual symptom checker module allows patients to report medical concerns using various input modalities, including voice, free text, curated symptom lists, and algorithm-suggested symptoms. This multi-modal approach is vital for individuals with low digital literacy. The system recognizes over 3,237 Bengali variations, including regional dialects, ensuring precise communication of health concerns.
Also Read:
- Training Proactive Language Models from Real-World Conversations
- BhashaBench V1: Evaluating Language Models for India’s Diverse Knowledge Systems
Impressive Performance and Future Outlook
The clinical accuracy of AmarDoctor was rigorously evaluated against a gold-standard set of 185 clinical vignettes developed by experienced physicians. Its performance was also compared with five independent physicians using the same vignettes. The results were striking: AmarDoctor achieved a top-1 diagnostic precision of 81.08%, significantly outperforming the physicians’ average of 50.27%. For specialty recommendations, the app achieved 91.35% precision, compared to the physicians’ average of 62.6%.
While these results are promising, the research acknowledges limitations and outlines future improvements. These include expanding language coverage, enriching the underlying data to incorporate more comprehensive patient demographics and medical histories, and exploring advanced AI techniques for complex disease manifestations. The team also plans to conduct large-scale randomized controlled trials and develop specialized modules for complex disease areas like cardiology and oncology.
AmarDoctor represents a significant step towards more inclusive and accessible AI-driven healthcare. By functioning as a machine-in-the-loop tool, it empowers both patients and medical professionals with timely, accurate, and personalized clinical information, ultimately aiming to democratize healthcare access in resource-limited settings. You can read the full research paper here.
