RadioCovid: an image analysis tool to aid diagnosis

Date :
Changed on 07/07/2020
RadioCovid was born out of the circumstances of the epidemic. The project proposes to use image analysis, Machine Learning and Deep Learning tools to assist in the diagnosis and, above all, the prognosis of disease progression in the management of patients on the basis of lung imaging (CT scans) and associated biological and clinical data.
Olivier Saut from the Monc project-team, common within Inria, CNRS and Bordeaux INP, presents the main lines of this project.

What is the genesis of your project?

At the end of March, we began discussions with a radiologist from the Tenon Hospital of the APHP, F. Cornelis, and T. Colin from the Sophia Genetics company - with whom we collaborate in particular within the framework of the Pimiento project supported by MSDAvenir and the Inria Foundation - about the possibility of improving the management of Covid patients with the help of an AI evaluation of the severity of the disease. We had encouraging preliminary results fairly quickly but with too few patients to have sufficient statistical significance. Unfortunately, this collaboration did not go beyond the proof of concept stage for reasons beyond our control, but it did familiarize us with the radiological aspects of the disease.

Project holders : Olivier Saut (EPI MONC), F. Cornelis (Tenon APHP Hospital), T. Colin (Sophia Genetics) Partners : CHU de Bordeaux, Centre de recherche cardio-thoracique de Bordeaux (CRCTB), l'USMR  #MachineLearning #DeepLearning #IA

One week later, M. Zysman, P. Berger and F. Laurent proposed us to collaborate on their project of a hospital clinical research program (PHRC) with the Bordeaux University Hospital (radiology and respiratory diseases departments) and the Cardio-Thoracic Research Center of Bordeaux, the CRCTB, (INSERM U1045) which significantly increased the number of patients studied and added clinical and biological information to the radiological images for each one.

How is it developing today and what are its objectives?

In close collaboration with the clinical (CHU Bordeaux, CRCTB) and methodological (USMR) partners, the PHRC, finally accepted at the end of May, is developing a retrospective study by IA of the severity of the disease based on the admission data (scanner, biology, clinical) of several hundred patients from Bordeaux and several other CHU.

The inclusion of patients has begun and the start of the analysis of the first group of patients should not be long in coming. All our efforts on the disease are focused on this PHRC in order to be able to determine criteria and algorithms predictive of the degradation of the disease in order to be able, by anticipation, to better manage the patients most at risk.

How do you work with your partners?

The CHU was strongly involved and the legal aspects were quickly settled with the participation of the CPPI and the centre's lawyers. While waiting for the improvement of the health situation, we work mainly by email and video-conference. The number of Covid patients is now relatively low in the region; this will free up extra time for the project by reducing the clinical workload of the doctors.