TY - JOUR T1 - Application of dual-stream three-dimensional convolutional neural network based on 18F-FDG PET/CT in distinguishing benign and invasive adenocarcinoma in ground-glass lung nodules JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1043 LP - 1043 VL - 62 IS - supplement 1 AU - Xiaonan Shao AU - Rong Niu AU - Xiaoliang Shao AU - Yuetao Wang Y1 - 2021/05/01 UR - http://jnm.snmjournals.org/content/62/supplement_1/1043.abstract N2 - 1043Objectives: The purpose of this work is to train, verify and test a dual-stream three-dimensional convolutional neural network (3D-CNN) based on Fluorine 18 (18F)-fluorodeoxyglucose (FDG) PET/CT to distinguish benign lesions and invasive adenocarcinoma (IAC) in ground-glass nodules (GGNs). Methods: We retrospectively analyzed patients with suspicious GGNs who underwent 18F-FDG PET/CT examinations in our hospital from November 2011 to November 2020 and screened out benign lesions and IAC. According to the ratio of 7:3, the data set was randomly divided into training data and testing data. Partial image feature extraction software was used to segment PET and CT images, and the training data after using the data augmentation was used for the training and validation (5-fold cross-validation) of the three CNNs (PET, CT, and PET/CT networks). According to the average accuracy, recall, and precision, the classification performance of different CNNs on the validation set and testing set was evaluated. Results: Finally, 23 benign nodules and 92 IAC nodules from 106 patients were included. In the validation set, the performance indicators of the PET network (accuracy of 0.92 ± 0.02; recall and precision of 0.97 ± 0.03 and 0.94 ± 0.04) were better than CT network (accuracy of 0.84 ± 0.03; recall and precision of 0.90 ± 0.07 and 0.90 ± 0.04); in the testing set, the performance indicators of CT and PET networks had decreased, but when the precision was close, the accuracy and recall of the PET network (0.79 ± 0.05 and 0.90 ± 0.05) were still higher than the CT network (0.72 ± 0.05 and 0.79 ± 0.08). In the validation set, the performance indicators of the PET/CT network were almost the same as those of the PET network; in the testing set, the performance indicators of the PET/CT network had declined (accuracy was 0.81 ± 0.03; recall and precision were 0.92 ± 0.03 and 0.86 ± 0.00), but still higher than both CT and PET networks, and the accuracy was higher than two nuclear medicine physicians [Physician 1 (3-year experience): 0.70 and Physician 2 (10-year experience): 0.73]. Conclusions: It is feasible to distinguish benign lesions and IAC in GGNs using 3D-CNN based on 18F-FDG PET/CT, and higher classification performance can be achieved when CT and PET images are used at the same time. View this table:Three 3D-CNN performance indicators ER -