AFRICAI-RI

The challenge

Diagnosis can be one of the most challenging areas of medicine, including for infectious diseases. Many infections share similar symptoms, and health facilities may lack the analytical tools needed to distinguish between them. These challenges are compounded by a shortage of specialist clinicians in sub-Saharan Africa. As a result, many conditions in the region are being under-diagnosed.

There are hopes that AI-based systems could overcome these bottlenecks. For example, they could be used to support automated analysis of medical images, using devices suitable for health workers without extensive specialist training. However, such AI tools need to be trained on large quantities of well-annotated images, and sub-Saharan Africa currently lacks large-scale medical image repositories.

The project

The AFRICAI-RI project is establishing a regional platform for the collation and analysis of medical images from across sub-Saharan Africa, to underpin the development of AI-based image-analysis tools for disease diagnosis. With an initial focus on lung conditions, it will provide a model that could be adapted in other areas of medicine.

The project is building on several recent developments, including the creation of an African Network for AI in Biomedical Imaging and a proposal to create a pan-African medical image repository. A secure data repository, Health Data Research West Africa, has been established in The Gambia under the auspices of the EDCTP-funded West Africa Network of Excellence for Clinical Trials in TB, AIDS and Malaria (WANETAM) network. This hub has established a robust data governance framework and data-sharing protocols.

The AFRICAI-RI project is using this hub as the basis for a larger secure image repository. It will apply the approach of federated data analysis, through which data remain within a secure local data centre but can be analysed remotely from a central hub, enabling pooled data analyses. Through the federated approach, the benefits of data sharing can be gained without data needing to be transferred to a central site, which can create data-security and data-governance challenges, particularly for international projects.

The project is focusing on regionally important lung conditions, particularly pneumonia in children and TB in children and adults, and two types of imaging data – chest X-rays and ultrasound. The latter is a relatively new technology with potential to be used at lower-level health facilities. The initial focus will be on four diagnostic use cases:

• TB detection using chest X-rays in adults with additional health conditions, including HIV infections.
• TB detection using chest X-rays in children.
• TB detection in adults using ultrasound.
• Pneumonia detection in children using ultrasound.

During the project, implementation studies will be organised for two of these use cases (paediatric chest X-ray and adult chest ultrasound). Using data already part of electronic health records, ‘in silico’ trials will enable the results of AI tools to be compared with the conclusions reached by specialist clinicians looking at the same source material.

Images will be shared by ten institutions across participating countries. Computing power will be strengthened at each site to enable them to participate, and data standards and a submission portal are being established to facilitate sharing of data in a consistent format. The project will also issue an open call that will enable three other institutions from sub-Saharan Africa to join the consortium.

The project has a strong focus on ethical frameworks and policy issues. An extensive multidisciplinary co-creation process will be organised to establish appropriate governance and operational guidelines, including management of intellectual property and ways to ensure academic recognition for centres whose images are included in third-party data analyses. 

Validation of the repository will be tested through six-month ‘data challenges’, addressing specific clinical diagnostic issues in areas of interest, which will be open to up to 15 teams for each of two challenges. In addition, a ‘hacking challenge’ will be organised to test security. 

Impact

The AFRICAI-RI project will establish a large-scale international medical image repository in sub-Saharan Africa. It will:

• Create a repository of around 75,000 annotated images related to lung conditions during the project.
• Support work to develop AI-based tools for diagnosis of important lung conditions in adults and children.
• Validate the performance of new diagnostic tools using existing health data.
• Demonstrate the workability of region-wide data-governance and data-sharing mechanisms.

The AFRICAI-RI project will establish a robust regional infrastructure for the storage and secure access of medical images, creating a resource to support diagnostic tool development for key lung conditions.  It will provide a platform to support innovation in the medical imaging and diagnostics field, while also demonstrating proof of principle for an approach that could eb extended to other disease areas.