Automated classification of increased uptake regions in bone single-photon emission computed tomography/computed tomography images using three-dimensional deep convolutional neural network.

作者: Seiichiro Ota , Hiroshi Toyama , Atsushi Teramoto , Yoshitaka Inui , Masakazu Tsujimoto

DOI: 10.1097/MNM.0000000000001409

关键词:

摘要: Objective This study proposes an automated classification of benign and malignant in highly integrated regions bone single-photon emission computed tomography/computed tomography (SPECT/CT) using a three-dimensional deep convolutional neural network (3D-DCNN). Methods We examined 100 35 patients with SPECT/CT classified as by other examinations follow-ups. First, SPECT CT images were extracted at the same coordinates cube, long side two times diameter high concentration images. Next, we inputted image to DCNN obtained probability benignity malignancy. Integrating output from each provided overall result. To validate efficacy proposed method, malignancy all was assessed leave-one-out cross-validation method; besides, accuracy evaluated. Furthermore, compared analysis results SPECT/CT, alone, whole-body planar scintigraphy region site. Results The volume interest 50 regions, respectively. alone 73% 68%, respectively, while that site 74%. When used, highest (80%), 82 78%, Conclusions suggests could be used for direct without extracting features accumulation patterns.

参考文章(0)