Early Detection of Liver Fibrosis Using Ultrasound Imaging and Artificial Intelligence Ref.No.SSTCRC2637
1. Introduction
From WHO report in 2021 liver disease accounts for over two million deaths annually (cirrhosis, viral hepatitis, and liver cancer) and accounts for 4% of all deaths worldwide (1 out of every 25 deaths); 1 out of 3 liver-related deaths occur among females. Within this estimate, liver cancer accounts for 600,000 to 900,000 deaths. Currently liver disease is the eleventh-leading cause of death, but liver deaths may be underestimated. Cirrhosis is currently the tenth-leading cause of death in Africa (thirteenth-leading cause in 2015); ninth-leading cause in South East Asia and Europe; and the fifth-leading cause of death in the Eastern Mediterranean.
Following the Lancet Global Health The burden of chronic liver disease in sub-Saharan Africa is high, with cirrhosis and other chronic liver diseases accounting for 2·5% of deaths and 1·3% of total disability-adjusted life-years in 2019.
The diagnosis of liver fibrosis resulting from hepatitis B poses a significant public health challenge, often being detected at a late stage. Until now, the identification of liver fibrosis relied on invasive procedures such as histology, specifically biopsy, where a needle is inserted into the liver to collect a tissue sample for analysis by pathologists.
In recent years, non-invasive markers for detecting liver fibrosis or cirrhosis have emerged as alternatives to biopsy. Examples include the Fibrotest, a blood marker, and the Fibroscan, which measures liver hardness or elasticity using shear waves. In the Gastroenterology Department, patients with Hepatitis B virus undergo periodic abdominal ultrasound examinations to monitor for cirrhosis. While ultrasound is relatively accessible, its resolution does not enable early detection of cirrhosis development, which is crucial for timely intervention.
People living in poverty face two challenges regarding hepatitis B:
Affordability: Biopsy costs in West Africa are often prohibitive, and even when individuals can afford it, obtaining results can take up to a month due to tests being sent to France.
Limited effectiveness of ultrasound: Although some individuals have access to ultrasound examinations, the resolution is insufficient for early detection, a vital factor in surviving this disease While fibrotest and fibroscan impose financial challenges on individuals, they also place a substantial financial burden on hospitals and present significant maintenance issues.
We aim to propose cost-effective AI-based solutions for early detection of liver fibrosis, regardless of its severity. We seize this opportunity to develop a comprehensive AI-based healthcare solution framework that can address a wide range of diseases and adapted to our African context.
2. Research Progress
In 2024 we laid the technical and clinical foundations of the project. Working with Le Dantec Hospital, we assembled an initial dataset of 125 liver-ultrasound cases under ethics approval and consent. On this basis, we trained first-pass models—both a convolutional neural network (CNN) and a Vision Transformer (ViT)—to screen for ≥F2 liver fibrosis. The work produced two peer-reviewed outputs (a Springer book chapter and a MICCAI proceedings paper) and we set up a dedicated validation and clinical-liaison team.
From January to August 2025 with a grant from Gates Foundation of 100000 $ we shifted to scale and productization. We launched a new collection drive targeting 500 additional ultrasound exams with matched clinical labels and, by the end of August, had captured roughly 80% of that goal. In parallel, we began formal validation, identified and documented sources of bias, and implemented a mitigation plan. We upgraded the models with interpretability features to support clinician trust. On the product side, the mobile AI application reached about 80% functional completeness, a companion app for structured data capture was built, and the nurse-oriented UI/UX was finalized.
From September to December 2025 our focus is deployment and clinical proof. Using Qualcomm AI Hub, we are optimizing a multimodal, on-device model designed to run offline with sub-300-ms latency per frame.
In next steps we want to integrate a handheld ultrasound probe to deliver a bundled “probe + app” experience and start pilot validation with health professionals at two sites, tracking operational metrics such as decision time (<30 s) and diagnostic targets (sensitivity/specificity). In parallel, we are preparing an intellectual-property filing as a utility model that covers the configured on-device apparatus
In short, we have progressed from concept to published models, a near-complete clinical dataset, and a working mobile application ready for pilot use. The next phase will demonstrate that an offline, smartphone-based screener can flag likely ≥F2 fibrosis quickly and at zero cost to patients, creating a viable path to scale in low-resource settings.
3. Cooperation Required
-Co-funding to finish data collection, model training, and hardware integration over 12–18 months, linked to clinical and commercial milestones.
-A technical co-builder (AI + hardware): joint training on Asian/African cohorts, fairness/bias audits and external validation; plus a handheld probe OEM/ODM to provide SDKs, certify compatibility, and co-brand the “probe + app”.
-Eventually a go-to-market partner for distribution in Africa (hospitals, NGOs, Ministries of Health) and Asia eventually, with field training, service support, and procurement channels.
4. Benefits
Healthcare Improvement
-Early Detection and Diagnosis: The project enables non-invasive, cost-effective, and accurate diagnosis of liver fibrosis, particularly in its early stages. This significantly improves patient outcomes by facilitating timely intervention and management.
-Reduced Dependency on Invasive Methods: By replacing biopsies with AI-enhanced ultrasound imaging, the project minimizes patient discomfort, risks of complications, and the financial burden associated with traditional diagnostics.
-Accessibility in Resource-Limited Settings: The scalable and cost-effective solution ensures healthcare accessibility for underserved populations, particularly in low-resource regions like sub-Saharan Africa.
Economic Benefits
-Cost-Effective Diagnostics: Lower reliance on expensive procedures like biopsies or advanced imaging techniques reduces healthcare costs for both patients and medical institutions.
-Optimized Resource Allocation: Hospitals and clinics can allocate resources more efficiently by focusing on AI-based screening methods.
Societal Benefits
-Improved Public Health: Early detection and timely treatment of liver diseases reduce overall disease burden, leading to healthier populations.