Published 2024 · first author
AI-DxMH
Peer-reviewed diagnostic framework for clinics without GPUs.
- Role
- First author
- Scope
- Literature review · system design
- Code
- Repository →
1st
author on the publication
2024
peer-reviewed publication
5
authors on the system proposal
How it works
- 01 Multi-modal intake history · symptoms · labs
- 02 Evidence review prior methods + constraints
- 03 System design multimodal decision flow
- 04 Deployment plan resource-constrained clinics
- 05 Explained output suggestion + reasoning
Published scope: literature review and system proposal. Implementation metrics require separate validation.
Evidence boundary
Proposal verified. Clinical benchmark not claimed.
The defensible outcome is the peer-reviewed architecture and evaluation plan. Accuracy, latency, and compression results remain future implementation gates until a reproducible benchmark is attached.
The problem
Diagnostic AI is usually benchmarked on hardware the clinics that need it most will never own. A model that needs a datacentre to answer is not a diagnostic tool for a rural clinic; it is a demo of one.
The approach
A system design over patient history, presenting symptoms and lab results, scoped for constrained deployment. Explainability is treated as a first-class requirement so a physician can inspect why a suggestion was made.
Evidence constraint
The paper is a literature review and system proposal. Repository experiments are kept separate from the peer-reviewed evidence and are not presented here as clinical validation.
What is verified
Published as first author with four co-authors in 2024, establishing the deployment problem, system architecture, explainability requirement, and evaluation plan for diagnostic support in low-resource settings.