Automatic cervical lymphadenopathy segmentation from CT data using deep learning.

摘要: Abstract Purpose The purpose of this study was to develop a fast and automatic algorithm detect segment lymphadenopathy from head neck computed tomography (CT) examination. Materials methods An ensemble three convolutional neural networks (CNNs) based on U-Net architecture were trained the lymphadenopathies in fully supervised framework. resulting predictions assessed using Dice similarity coefficient (DSC) examinations presenting one or more adenopathies. On without adenopathies, score given by formula M/(M + A) where M mean adenopathy volume per patient A segmented algorithm. 117 annotated CT acquisitions. Results test set included 150 additional acquisitions unseen during training. performance yielded 0.63. Conclusion Despite limited available data partial annotations, our CNN approach achieved promising results task cervical segmentation. It has potential bring precise quantification clinical workflow assist clinician detection task.