Institution: | 1. Philips Research France, 33, rue de Verdun, 92150 Suresnes, France;2. LTCI, Télécom ParisTech, Université Paris-Saclay, 75013 Paris, France;3. CREATIS, CNRS UMR 5220, Inserm U1206, INSA-Lyon, Claude Bernard Lyon 1 University, 69100 Villeurbanne, France;4. Department of Radiology, Hospices Civils de Lyon, 69002 Lyon, France;5. Department of Radiology, Hôpital Tenon, AP–HP, GRC-UPMC n°5 Oncotype-URO, Sorbonne universités, 75020 Paris, France;6. Department of Radiology, CHRU de Besançon, 25000 Besançon, France;g. Department of Radiology, Hôpitaux Universitaires Paris Sud, 94270 Le Kremlin Bicêtre, France;h. Department of Radiology, CHU de Strasbourg, Nouvel Hôpital Civil, 67000 Strasbourg, France;i. Department of Uroradiology, Hospices Civils de Lyon, Faculté de Médecine Lyon Est, 69002 Lyon, France;j. Department of Radiology, Hôpital Pitié Salpétrière, AP–HP, 75013 Paris, France;k. Department of Radiology, Gustave Roussy, IR4M, UMR8081, CNRS, Université Paris-Sud, Université Paris-Saclay, 94805 Villejuif, France |
Abstract: | PurposeThis work presents our contribution to one of the data challenges organized by the French Radiology Society during the Journées Francophones de Radiologie. This challenge consisted in segmenting the kidney cortex from coronal computed tomography (CT) images, cropped around the cortex.Materials and methodsWe chose to train an ensemble of fully-convolutional networks and to aggregate their prediction at test time to perform the segmentation. An image database was made available in 3 batches. A first training batch of 250 images with segmentation masks was provided by the challenge organizers one month before the conference. An additional training batch of 247 pairs was shared when the conference began. Participants were ranked using a Dice score.ResultsThe segmentation results of our algorithm match the renal cortex with a good precision. Our strategy yielded a Dice score of 0.867, ranking us first in the data challenge.ConclusionThe proposed solution provides robust and accurate automatic segmentations of the renal cortex in CT images although the precision of the provided reference segmentations seemed to set a low upper bound on the numerical performance. However, this process should be applied in 3D to quantify the renal cortex volume, which would require a marked labelling effort to train the networks. |