The predictive value of multiple artificial intelligence models in axillary lymph node metastasis of breast cancer detected by ultrasound - a network meta-analysis.
Authors
Affiliations (1)
Affiliations (1)
- Department of Ultrasound Medicine, the First Affiliated Hospital of Shihezi University, Shihezi, Xinjiang, China.
Abstract
To compare the diagnostic performance of different AI models in predicting axillary lymph node metastasis in breast cancer detected by ultrasound. A systematic search strategy was conducted in PubMed, Embase, Web of Science, Cochrane Library, CNKI, and Wanfang Data from July 2015 to July 2025 for diagnostic studies on the prediction of axillary lymph node metastasis in breast cancer. Relevant diagnostic accuracy data were extracted and synthesized. A network meta-analysis was performed to compare the performance of different AI models. This study finally included 9 studies involving 2,813 patients through strict inclusion and exclusion criteria. The predictive value of 11 artificial intelligence models for axillary lymph node metastasis was assessed. The methodological quality of the included studies was good, and there was no obvious publication bias. The pooled diagnostic odds ratio of all diagnostic methods was 10.92, showing statistically significant heterogeneity (I² = 92%, <i>P</i> < 0.01). Pooled diagnostic performance showed that Boosting-based algorithm models achieved the highest sensitivity (83.0%) and negative predictive value (86.4%), while the Temporal Interlace Network model achieved the highest specificity (90.0%) and positive predictive value (80.0%). Pairwise comparisons revealed that the sensitivity of Boosting-based algorithm models was significantly higher than that of Support Vector Machines (0.17, 95% CI: 0.01-0.33), and the negative predictive value was significantly higher than that of Support Vector Machines (0.09, 95% CI: 0.01-0.17). No statistically significant differences were observed in the predictive performance of the remaining models. Boosting-based algorithm models showed a possible signal of favorable and balanced predictive value in ultrasound-based axillary lymph node metastasis, though substantial heterogeneity and inconsistency limit definitive conclusions. https://www.crd.york.ac.uk/prospero/, identifier CRD420251119253.