The Added Effect of Artificial Intelligence in CT Assessment of Abdominal Lymphadenopathy

Lymphadenopathy is associated with lymph node abnormal size or consistency due to many causes. We employed the deep convolutional neural network ResNet-34 to detect and classify CT images from patients with abdominal lymphadenopathy and healthy controls. We created a single database containing 1400 source CT images for patients with abdominal lymphadenopathy (n = 700) and healthy controls (n = 700). To train, test, and cross-validate the ResNet-34 classifier to detect specific lesions, we first resized and normalized all images. Then, we randomly divided the 1400 images into 88 groups of 16, and the classifier was trained to identify and label lesions using automatic volume delineation 3D reconstruction of target areas. The ResNet-34 had a diagnostic accuracy with receiver operating characteristic (ROC) curves of the true-positive rate versus the false-positive rate with the area under the curve (AUC) of 0.9957 and 1.00 for abdominal lymphadenopathy and healthy control CT images, respectively. This accuracy implied identical high sensitivity and specificity values of 99.57 % and 100% for the two groups. The added effect of ResNet-34 is a success rate of 99.57% and 100% for classifying random CT images of the two groups, with an overall accuracy of 99.79% in the testing subset for detecting and classifying lymph node lesions. Based on this high classification precision, we believe the output activation map of the final layer of the ResNet-34 is a powerful tool for the accurate diagnosis of lymph node lesions of abdominal lymphadenopathy from CT images.