Localization and Classification of Adrenal Masses in Multiphase Computed Tomography: Retrospective Study

• DOI: 10.2196/65937
• Publication Date: 2025-04-24T20:30:47Z
• Authors: Liuyang Yang; Xinzhang Zhang; Zhenhui Li; Jian Wang; Yiwen Zhang; Liyu Shan; Xin Shi; Yapeng Si; Shuailong Wang; Lin Li; Ping Wu; Ning Xu; Lizhu Liu; Junfeng Yang; Jinjun Leng; Maolin Yang; Zhuorui Zhang; Junfeng Wang; Xingxiang Dong; Guangjun Yang; Ruiying Yan; Wei Li; Zhimin Liu; Wenliang Li
• Keywords: Computer applications to medicine. Medical informatics; R858-859.7; Public aspects of medicine; RA1-1270

Background The incidence of adrenal incidentalomas is increasing annually, and most types masses require surgical intervention. Accurate classification common based on tumor computed tomography (CT) images by radiologists or clinicians requires extensive experience often challenging, which increases the workload leads to unnecessary surgeries. There an urgent need for a fully automated, noninvasive, precise approach identification accurate masses. Objective This study aims enhance diagnostic efficiency transform current clinical practice preoperative diagnosis Methods retrospective analysis that includes patients with who underwent adrenalectomy from January 1, 2021, May 31, 2023, at Center 1 (internal dataset), 2016, 2 (external dataset). include unenhanced, arterial, venous phases, 21,649 used training set, 2406 validation 12,857 external test set. We invited 3 experienced precisely annotate images, these annotations served as references. developed deep learning–based mass detection model, Multi-Attention YOLO (MA-YOLO), can automatically localize classify 6 In order scientifically evaluate model performance, we variety evaluation metrics, in addition, compared improvement efficacy doctors after incorporating assistance. Results A total 516 were included. MA-YOLO achieved intersection over union 0.838, 0.885, 0.890 localization phase CT respectively. corresponding mean average precision was 0.913, 0.915, Additionally, assistance this performance improved. Except cysts, least physician significantly improved other 5 tumors. Notably, categories adenoma (for senior clinician: P=.04, junior radiologist: P=.01, P=.01) cortical carcinoma (junior P=.02, intermediate P=.001), half physicians showed significant improvements using Conclusions demonstrates ability achieve efficient, accurate, noninvasive examinations, showing promising potential future applications.