Using artificial intelligence in the analysis of CT scans of the axillary nodes in breast cancer: A systematic review
Julie Cox, Amir Bhatti, Amir Atapour–Abarghouei · European Journal of Radiology Artificial Intelligence · 2025
Background Breast cancer is the most diagnosed cancer worldwide, with early diagnosis key to improved outcomes following treatment. Accurate assessment of whether breast cancer has spread to axillary lymph nodes is of vital importance in deciding on treatment paths for patients and for providing prognostic information. We therefore aimed to evaluate and synthesise findings from existing research regarding the effectiveness of using artificial intelligence (AI) to diagnose axillary lymph node metastasis on CT scans in patients who have a diagnosis of breast cancer and have had axillary surgery to establish the histological status of the axillary nodes. Methods The protocol was prospectively registered with PROSPERO (CRD42024598174). Four electronic databases (Embase, Medline, the Cochrane Library, and Google Scholar) were searched for relevant studies relating to different AI methods, axillary lymph nodes and breast cancer. The quality of included articles was independently appraised using the Quality Assessment Tool for Diagnostic Accuracy Studies (QUADAS-2) tool. Data were narratively synthesised. Findings Three studies were eligible for inclusion in this review. All were single centre retrospective studies of breast cancer patients. Two studies used convolutional neural networks and one used machine learning. Results indicated that AI models were able to accurately diagnose axillary lymph node metastasis in breast cancer CT scans. Interpretations AI-based analyses of CT images show substantial potential to accurately localise axillary lymph nodes and predict metastasis. This method can offer an increasingly precise non-invasive diagnostic strategy for breast cancer staging. There are, however, some outstanding limitations, including limited samples and a small number of retrospective studies on this topic, which need to be overcome in future research. Funding No additional funding was received for this research