AI for automatic quality assurance of medical image at a very large scale

Olivier Colliot and Ninon Burgos, PRAIRIE chairs, have published the first paper using medical imaging data from the clinical routine data warehouse of the Greater Paris Hospital (Entrepôt de données de santé, AP-HP). The data warehouse contains all medical images from patients hospitalized within the Greater Paris area hospitals of AP-HP.

In particular, it contains over 100,000 3D T1 brain MRIs. The paper describes an AI method that performs automatic quality control of medical images, which is a prerequisite for any subsequent analyses. This has a potentially very high impact. Indeed, clinical routine data warehouses, which contain data from millions of patients, have a huge potential for training AI models. Nevertheless, the quality of imaging data is heterogeneous and thus quality control is mandatory but is not feasible manually at such a large scale. The paper was published in Medical Image Analysis, the leading journal in AI for medical imaging: https://www.sciencedirect.com/science/article/abs/pii/S1361841521002644. It is also freely available on HAL: https://hal.inria.fr/hal-03154792v4.

This work was done as a collaboration between the ARAMIS project-team (CNRS, Inria, Inserm, Sorbonne University, Paris Brain Institute) and the EDS team of AP-HP. This work was also supported by the Abeona Foundation.