© nerthuz – stock.adobe.com
News • Machine learning
AI model to detect breast cancer metastasis in lymph nodes
The study also concludes that a group of primary tumour cells could constitute a candidate for a new biomarker and, in the future, a possible therapeutic target.
Researchers of different Catalan research institutions and hospitals have developed a new AI model to detect whether breast cancer metastasizes in areas near the chest, specifically the armpit lymph nodes, essential immune system filters.
The researchers published their findings in the journal Breast Cancer Research.
While previous studies had fundamentally considered the clinicopathological data of the tumour (such as size and characteristics), the new AI system has been developed within the framework of the first study to combine these data with those of patients’ immune response.
The research confirms that the immune response is a crucial factor in determining whether or not luminal breast cancer, one of the most common types, metastasizes. However, current methods for diagnosis and technologies do not routinely assess it in clinical practice. Thus, the new machine learning model can serve to reinforce current diagnostic methods.
It is a step towards better understanding why some tumours spread to the lymph nodes and others do not, and what role the immune response may play
Carlos López Pablo
The AI has been developed technologically by UPF researchers Alba Fischer Carles, Gemma Piella and Jérôme Noailly, of the BCN MedTech Unit at the Department of Engineering. The clinical part of the study was led by Carlos López Pablo, linked to Verge de la Cinta University Hospital in Tortosa, the Biomedical Research Institute of Southern Catalonia (IRB CatSud), and Rovira i Virgili University. Other researchers from these same institutions and from Joan XXIII University Hospital in Tarragona also participated.
This research reflects the progress made in the knowledge of the immune mechanisms involved in the spread of breast cancer to the lymph nodes. According to the latest data published by the World Health Organization (WHO) for 2024, that year over 2.4 million women worldwide were diagnosed with breast cancer of whom 694,000 died. The mortality rate from this disease increases significantly when tumour cells spread to the lymph nodes of the armpit.
“In this context, we have developed a machine learning model that integrates immune response data with clinical and pathological data with the aim of exploring its potential to determine the presence of metastases in the axillary lymph nodes”, UPF researcher Alba Fischer Carles explains. Seven of the nine immune variables of the final model are taken from the study of tissues extracted from the axillary lymph nodes (without metastasis) of this group of patients.
Before designing the final model, a further three models were tested. Two of them were trained with clinical and immunological data from the same group of 83 women diagnosed with the same type of luminal breast cancer between 1995 and 2008, and a third with a larger series of 344 patients (including these 83), solely with clinical data. With the final model, which includes immune data, false positive diagnoses of luminal breast cancer metastasis are reduced with respect to the AI system trained only with clinical data. The final AI model does not have a predictive value, that is, it does not enable predicting whether or not a patient will develop metastasis, but it does determine which clinical and especially immune variables -present in the primary tumour and in the lymph nodes themselves- are associated with having metastases in the axilla.
The research has also identified a potential new biomarker of axillary breast cancer metastasis (intratumoral CD21+). Biomarkers are biological features that enable objectively measuring pathological alterations. In this case, a group of cells located within the primary tumour (follicular dendritic cells) could be decisive in detecting the extent of the tumour.
Carlos López Pablo explains that, so far, the lymph node has been analysed mainly to determine whether or not it contains tumour cells. “This work tells us that nodes without metastasis also provide information: the way in which the immune response is organized, together with that of the primary tumour, is related to the behaviour of the disease”, he assures. “It is a step towards better understanding why some tumours spread to the lymph nodes and others do not, and what role the immune response may play. However, before reaching clinical practice, this will need to be confirmed in broader series and in patients from other centres”, he concludes.
Thus, the efficacy of CD21+ as a biomarker should continue to be evaluated in future studies and this research paves the way to exploring it as a possible therapeutic target.
The study was funded through the European project BosomShield (MSCA Doctoral Networks, HORIZON-MSCA-2021-DN-01-01, GA 101073222), of the SCARLET project (TED2021-130081B-C22, MCIN and NextGenerationEU).
Source: Universitat Pompeu Fabra Barcelona
26.10.2026
