Saving lives with
AI-powered liver care

Hepato-Sentinel brings early jaundice detection to Ethiopia's primary care — using a smartphone and AI to cut mortality from 65% to under 30%.

65 % mortality (ICU)
47k health workers
2k patients in pilot
Support the project See how it works
4G 09:41
Sclera Scan
Bilirubin estimate
2.4 mg/dL
Risk
Moderate · Refer
Nearest hepatology unit
Normal Jaundice

A silent killer in primary care

In Ethiopia, 80% of primary facilities lack bilirubin tests. Visual inspection fails to detect jaundice until it's too late — especially on darker skin.

Detection Gap

No affordable point-of-care test. Jaundice is missed until bilirubin >5–7 mg/dL.

Integration Gap

Fragmented referral pathways cause fatal delays between community and specialist care.

Validation Gap

AI models trained on Asian/European data — not validated on Ethiopian skin tones (Fitzpatrick IV–VI).

Implementation Gap

No established framework to integrate digital tools into Ethiopia's Health Extension Program.

65% ICU mortality 28h earlier detection

Hepato-Sentinel: AI for all

A smartphone-based, offline AI system that detects jaundice at 2 mg/dL — validated for Ethiopian skin tones and integrated with the national health system.

Smartphone Colorimetry

Uses sclera images + colour calibration cards. Works in low-resource settings without lab.

CNN + EfficientNet

Lightweight, on-device AI (JaunENet architecture) with 98.9% accuracy, 0.128 GFLOPs per image.

Offline & Integrated

Works without internet. Syncs with MoH/eCHIS when available. Referral pathways built-in.

HEW Empowered

Designed for Health Extension Workers — 47,000+ across Ethiopia. Training & community engagement.

TensorFlow Lite Flutter DHIS2 / eCHIS End-to-end encryption Amharic + English

From lab to life-saving

30-month implementation science project — co-developed with MoH, hospitals, and communities.

Phase 1
Months 1–8
AI Validation — 2,000 patients across 5 hospitals (Addis Ababa, Bahir Dar, Debre Birhan, Hawassa). Serum bilirubin + smartphone images. Hybrid CNN + EfficientNet-B4.
Phase 2
Months 9–14
Platform & Pathways — Android app (Flutter, offline AI, DHIS2/eCHIS integration). Delphi process with 15 specialists for referral protocols.
Phase 3
Months 15–24
Pilot Implementation — Stepped-wedge cluster RCT in 15 health posts, 3 regions. 60 HEWs trained. Telemedicine consultation (optional).
Phase 4
Months 25–30
Evaluation & Scale-up — GLMM impact analysis, cost-effectiveness, national roadmap workshop with MoH. Patent filing + 4+ publications.

Measurable impact

Our pilot aims to demonstrate a 55% mortality reduction in liver failure and build a sustainable digital health model for Ethiopia.

55%
Mortality reduction (pilot sites)
200
Health Extension Workers trained
4+
Peer-reviewed publications
1
Patent (EIPA) · AI jaundice method
50
Health posts equipped
2k
Open-access annotated dataset (neonatal)

Who we are

Multidisciplinary team from AASTU, MoH, Yekatit 12 MC, DBU/Hakim Gizaw Hospital, and partners.

GN
Dr. Girma Neshir
PI · AASTU · ML & Data Science
LK
Dr. Lemlem Kassa
System Design · AASTU
HB
Dr. Hirut Bekele
Internal Medicine · Clinical Validation
EE
Dr. Ermiyas Endewunet
Hepatology Lead · DBU/Hakim Gizaw
DJ
Dr. Dureti Jemal
Hepatology (Neonates) · Yekatit 12 MC
WN
Mr. Wondossen Nigatu
MoH · Community Engagement & eCHIS
+ 2 MSc & 1 PhD student

Budget snapshot

Total request: ETB 1,559,000 (~USD 27,000). Co-funding from MoH, hospitals, and in-kind contributions.

Data Collection
ETB 600k
Model & Annotation
ETB 200k
App & Dashboard
ETB 254k
HEW Training
ETB 105k
Pilot Implementation
ETB 200k
M&E + Dissemination
ETB 200k
Total: ETB 1,559,000 (≈ USD 27,000)

Be part of the solution

Your support will train 200 health workers, equip 50 health posts, and save lives. Join us in building Ethiopia's first AI-powered hepatology network.

Donate now Download proposal

girma.neshir@aastu.edu.et  ·  +251-913-021313