A collaborative project combining the provision of health data and artificial intelligence, Deep.piste is one of the first pilot projects selected through a Health Data Hub (HDH) call for projects.

Deep.piste is led by Epiconcept in partnership with the Centre Régional de Coordination des Dépistages des Cancers de la Région Occitanie and Professor Fabien Reyal’s research team at the Institut Curie. Integrated with e-SIS, Epiconcept software for managing cancer screening campaigns – including breast cancer – the project offers an AI contribution to the analysis of mammograms in 2nd or 3rd reading.

With an estimated 11883 deaths in 2017, breast cancer is the biggest killer of women in France. Early detection can reduce its mortality by 21%. To assess the impact of an AI contribution in the organized breast cancer screening system, Deep.piste uses cross sources:
– The e-SIS database of anonymized breast cancer screening data from the Gard and Lozère departments (more than 250,000 images annotated in ACR classification);
– Data from the Système National des Données de Santé (National Health Data System), intended to identify important medical events outside the screening circuit (causes of death, reimbursement data, etc.) in order to reduce the number of false negatives.
Exploited in a scenario-based logic and in support of dense convolutional neural networks (deep learning), this data will participate in the screening process in a new way. The Deep.Piste project aims to develop an automatic analysis of mammograms and to refine the understanding of risk factors. This will make it possible to identify cases that could benefit from a lighter screening system (by reducing, for example, the frequency of mammograms to be performed for certain patients) or by strengthening the screening pathway for others.