Thales AI for Healthcare: Pulmonary Embolism Diagnostics

During the last month, our AI team participated in?the data challenge organized by Société Fran?aise de Radiologie et d'imagerie médicale (SFR) .

The ambition of this challenge was to develop a fully automatic approach for Pulmonary Embolism detection and severity evaluation, by the prediction of two gravity scores: the RV/LV ratio and the Qanadli score. While embolism detection is a problem with known solutions, automatic scoring remains an open challenge.

The Qanadli score is an index that aim to quantify the embolic load by taking into account the location of the clots and the degree of obstruction. 泰雷兹 team explored?two approaches to obtain the best accuracy. The first required a segmentation of the pulmonary tree in which the clots are positioned, while the second one used a 3D neural network to directly estimate the score. This last approach performed best and was selected. Concerning the estimation of the RV/LV ratio, a pre-trained segmentation model specifically dedicated to the heart was used. This model was finetuned with databases external to the data challenge.?

Pulmonary arterial tree and Qanadli areas of greater importance for embolic load (modified from Sketchfab)?

Within a very short time, our team accomplished this challenge and built a complete pipeline with multiple neural networks. 泰雷兹 ranked top 3, along with 飞利浦 and GE医疗 . Pulmonary Embolism detection and accurate scores (RV/LV ratio and Qanadli score) prediction seem to be now possible.?

The results were shared with radiology specialists and major industrial players of medical imaging during the annual JFR.plus event in Paris between 7-10 October.

Thanks to my AI team, I'm so proud of you : Elodie Callea , Liv Bastos , Léa Tsangarakis , Yang Ji , Geoffrey Portelli, Ph.D. , Florent DUPONT and our special guests, Gladis Valenzuela & Nicolas BILLET

Thanks for their support : Marion Boulay , Hadrien Gascuel

Thanks to our partners : Thales Microwave & Imaging Sub-Systems , Trixell & 英伟达 ( Sabrina Onasch )

Very warm thanks to Société Fran?aise de Radiologie et d'imagerie médicale (SFR) : Nathalie LASSAU , Fran?ois Cotton , Mathieu LEDERLIN, Philippe Soyer , Younes Belkouchi