False-negative and false-positive outcomes of computer-aided detection on brain metastasis: Secondary analysis of a multicenter, multireader study

• DOI: 10.1093/neuonc/noac192
• Publication Date: 2022-08-09T14:04:10Z
• Authors: Xiao Luo; Yadi Yang; Shaohan Yin; Hui Li; Weijing Zhang; Guixiao Xu; Weixiong Fan; Dechun Zheng; Jianpeng Li; Dinggang Shen; Yaozong Gao; Ying Shao; Xiaohua Ban; Jing Li; Shanshan Lian; Cheng Zhang; Lidi Ma; Cuiping Lin; Yingwei Luo; Fan Zhou; Shiyuan Wang; Ying Sun; Rong Zhang; Chuanmiao Xie
• Keywords: 03 medical and health sciences; 0302 clinical medicine; ROC Curve; Brain Neoplasms; Computers; Clinical Investigations; Humans; Prospective Studies; Magnetic Resonance Imaging; Sensitivity and Specificity

Abstract Background Errors have seldom been evaluated in computer-aided detection on brain metastases. This study aimed to analyze false negatives (FNs) and positives (FPs) generated by a metastasis system (BMDS) readers. Methods A deep learning-based BMDS was developed prospectively validated multicenter, multireader study. Ad hoc secondary analysis restricted the prospective participants (148 with 1,066 metastases 152 normal controls). Three trainees 3 experienced radiologists read MRI images without BMDS. The number of FNs FPs per patient, jackknife alternative free-response receiver operating characteristic figure merit (FOM), lesion features associated were analyzed for readers using binary logistic regression. Results FNs, FPs, FOM stand-alone 0.49, 0.38, 0.97, respectively. Compared independent reading, BMDS-assisted reading 79% fewer (1.98 vs 0.42, P < .001); 41% more (0.17 0.24, .001) but 125% (P higher (0.87 0.98, .001). Lesions small size, greater number, irregular shape, lower signal intensity, located nonbrain surface Small, irregular, necrotic lesions frequently found mainly resulted from blood vessels Conclusions Despite improvement performance, attention should be paid enhancement radiologists, especially less-experienced radiologists.