OBJECTIVE: The aim of this study was to investigate the diagnostic ability of a completely automated computer-assisted diagnosis (CAD) system to detect metastases in bone scans by two patterns: one was per region, and the other was per patient. MATERIALS AND METHODS: This study included 406 patients with suspected metastatic bone tumors who underwent whole-body bone scans that were analyzed by the automated CAD system. The patients were divided into four groups: a group with prostatic cancer (N = 71), breast cancer (N = 109), males with other cancers (N = 153), and females with other cancers (N = 73). We investigated the bone scan index and artificial neural network (ANN), which are parameters that can be used to classify bone scans to determine whether there are metastases. The sensitivities, specificities, positive predictive value (PPV), negative predictive value (NPV), and accuracies for the four groups were compared. Receiver operating characteristic (ROC) analyses of region-based ANN were performed to compare the diagnostic performance of the automated CAD system. RESULTS: There were no significant differences in the sensitivity, specificity, or NPV between the four groups. The PPVs of the group with prostatic cancer (51.0 %) were significantly higher than those of the other groups (P < 0.01). The accuracy of the group with prostatic cancer (81.5 %) was significantly higher than that of the group with breast cancer (68.6 %) and the females with other cancers (65.9 %) (P < 0.01). For the evaluation of the ROC analysis of region-based ANN, the highest Az values for the groups with prostatic cancer, breast cancer, males with other cancers, and females with other cancers were 0.82 (ANN = 0.4, 0.5, 0.6, 0.7, and 0.8), 0.83 (ANN = 0.7), 0.81 (ANN = 0.5), and 0.81 (ANN = 0.6), respectively. CONCLUSION: The special CAD system "BONENAVI" trained with a Japanese database appears to have significant potential in assisting physicians in their clinical routine. However, an improved CAD system depending on the primary lesion of the cancer is required to decrease the proportion of false-positive findings.
OBJECTIVE: The aim of this study was to investigate the diagnostic ability of a completely automated computer-assisted diagnosis (CAD) system to detect metastases in bone scans by two patterns: one was per region, and the other was per patient. MATERIALS AND METHODS: This study included 406 patients with suspected metastatic bone tumors who underwent whole-body bone scans that were analyzed by the automated CAD system. The patients were divided into four groups: a group with prostatic cancer (N = 71), breast cancer (N = 109), males with other cancers (N = 153), and females with other cancers (N = 73). We investigated the bone scan index and artificial neural network (ANN), which are parameters that can be used to classify bone scans to determine whether there are metastases. The sensitivities, specificities, positive predictive value (PPV), negative predictive value (NPV), and accuracies for the four groups were compared. Receiver operating characteristic (ROC) analyses of region-based ANN were performed to compare the diagnostic performance of the automated CAD system. RESULTS: There were no significant differences in the sensitivity, specificity, or NPV between the four groups. The PPVs of the group with prostatic cancer (51.0 %) were significantly higher than those of the other groups (P < 0.01). The accuracy of the group with prostatic cancer (81.5 %) was significantly higher than that of the group with breast cancer (68.6 %) and the females with other cancers (65.9 %) (P < 0.01). For the evaluation of the ROC analysis of region-based ANN, the highest Az values for the groups with prostatic cancer, breast cancer, males with other cancers, and females with other cancers were 0.82 (ANN = 0.4, 0.5, 0.6, 0.7, and 0.8), 0.83 (ANN = 0.7), 0.81 (ANN = 0.5), and 0.81 (ANN = 0.6), respectively. CONCLUSION: The special CAD system "BONENAVI" trained with a Japanese database appears to have significant potential in assisting physicians in their clinical routine. However, an improved CAD system depending on the primary lesion of the cancer is required to decrease the proportion of false-positive findings.
Authors: Simin Liu; Ming Feng; Tingting Qiao; Haidong Cai; Kele Xu; Xiaqing Yu; Wen Jiang; Zhongwei Lv; Yin Wang; Dan Li Journal: Cancer Manag Res Date: 2022-01-03 Impact factor: 3.989