Automatic segmentation of thoracic aorta segments in low-dose chest CT

Abstract

Morphological analysis and identification of pathologies in the aorta are important for cardiovascular diagnosis and risk assessment in patients. Manual annotation is time-consuming and cumbersome in CT scans acquired without contrast enhancement and with low radiation dose. Hence, we propose an automatic method to segment the ascending aorta, the aortic arch and the thoracic descending aorta in low-dose chest CT without contrast enhancement. Segmentation was performed using a dilated convolutional neural network (CNN), with a receptive field of 131 × 131 voxels, that classified voxels in axial, coronal and sagittal image slices. To obtain a final segmentation, the obtained probabilities of the three planes were averaged per class, and voxels were subsequently assigned to the class with the highest class probability. Two-fold cross-validation experiments were performed where ten scans were used to train the network and another ten to evaluate the performance. Dice coefficients of 0.83 ± 0.07, 0.86 ± 0.06 and 0.88 ± 0.05, and Average Symmetrical Surface Distances (ASSDs) of 2.44 ± 1.28, 1.56 ± 0.68 and 1.87 ± 1.30 mm were obtained for the ascending aorta, the aortic arch and the descending aorta, respectively. The results indicate that the proposed method could be used in large-scale studies analyzing the anatomical location of pathology and morphology of the thoracic aorta.

3D Automatic Segmentation of Aortic Computed Tomography Angiography Combining Multi-View 2D Convolutional Neural Networks

“…Moreover, the CTAs in our dataset comprise annotations of thoracic, abdominal, and iliac segments, including aortic arch branch vessels (innominate artery, left common carotid artery, and left subclavian artery) and abdominal aortic branch vessels (celiac trunk, superior mesenteric artery, left and right renal arteries), while most of the studies available in literature are focused on a single area. [7][8][9]14 In this way, the proposed approach allows the analysis of the whole aortic morphology.…”

“…According to Noothout et al, a separate deep learning model is trained for each orthogonal view using the same U-Net architecture. 14 The three networks are trained using the same CTA scans, but taken from different perspectives. (c) Multi-view Aggregation.…”

“…According to Noothout et al, the single-view prediction maps are averaged to provide a final prediction map. 14 The final prediction for each voxel x in the per-view prediction maps is computed as follows:…”

“…Since 3D CNNs usually require larger dataset and are memory-demanding, 5 some studies focused on the integration of orthogonal 2D-CNNs for medical image segmentation. 10,11,17,20 To the best of our knowledge, only one publication deals with 3D integration of 2D CNNs in the endovascular field, 14 where the authors average together axial, sagittal, and coronal predictions to provide a smoother segmentation of the aortic lumen from CT scans. Following these state-of-the-art approaches, we have adopted a combination of 2D U-Nets trained on orthogonal planes to provide a final 3D segmentation.…”

Deep learning from dual‐energy information for whole‐heart segmentation in dual‐energy and single‐energy non‐contrast‐enhanced cardiac CT

“…Therefore, several works using reference segmentations obtained in NCCT have only focused on the aorta or the full heart, and not on cardiac chambers. [15][16][17] Studies aiming to segment cardiac chambers in NCCT either rely on ambiguous reference segmentations obtained in NCCT or challenging and error-prone inter-modality registrations to transfer manual reference segmentations from other modalities in which cardiac structures are clearly distinguishable. Kaderka et al 18 directly obtained manual segmentations on NCCT images to build an atlas for the automatic segmentation of the cardiac chambers.…”