Projet 10 – Deep learning-assisted automated segmentation of lungs and COVID-19 lesions from chest CT images Auteurs Habib Zaidi, Département Diagnostique, HUG Isaac Shiri, Département Diagnostique, HUG Hossein Arabi, Département Diagnostique, HUG Résumé du projet Chest CT imaging has emerged as a promising tool for early diagnosis and longitudinal follow up of COVID-19 patients. However, quantitative analysis of clinical CT images through lung and infectious lesions segmentation remains challenging. We developed an automated segmentation of whole lungs and infectious regions using deep learning algorithms to implement a fast, consistent, robust and human error immune framework for lung and pneumonia lesion detection and quantification. A mean accuracy in lesions and lung segmentation larger than 0.91 and 0.98, respectively, was achieved, with a low amount of false negative and false positive rates. Introduction A wide range of radiological examinations, including CT scanning, are being employed for diagnostic and prognostic purposes to tackle the COVID-19 pandemic. A challenging problem which arises in clinical routine quantitative analysis of CT images is lung and infectious lesion segmentation owing to large inter/intra-observer variability and timeconsuming annotation process. Deep learning algorithms have been widely utilized in various medical image analysis problems owing to the promising results achieved. An Artificial intelligent solution with robust and consistent performance would enable efficient management of COVID-19 patients. Innovation Novel deep learning techniques are revolutionizing clinical practice and are now offering unique capabilities to the clinical imaging community and biomedical researchers at large. We set out to develop an automated segmentation of 3D whole lung and infectious lesions in COVID-19 patients using deep learning algorithms to enable fast, consistent, robust and human error immune framework for lung and pneumonia lesion detection and delineation. Due to the complex nature of the problem and high variability in lesion manifestation (appearance, size, location, boundaries), a novel deep learning-assisted image segmentation technique was proposed and implemented to enrich specific COVID-19 pneumonia features identification from CT images. Moreover, a large multicenter and multi-scanner dataset was collected for the development of a deep learning model enabling efficient COVID-19 patients management. The developed artificial intelligence platform was evaluated using a wide range of COVID-19 patients at different stages of the disease and diverse populations from multiple centers around the world. Avantages A number of studies reported that CT is a fast and highly sensitive approach for COVID-19 detection, segmentation and management. The proposed model enables whole lung and infectious lesions segmentation in very short time (few seconds) compared to time26