AI model from Lund University indicates four out of ten breast cancer patients could avoid axillary surgery

The study in brief: interdisciplinary research // peer-reviewed publication // cohort study // number of patients in the study: 1,265 // statistical correlation // retrospective

A project at Lund University in Sweden has trained an AI model to identify breast cancer patients who could be spared from axillary surgery. The model analyses previously unutilised information in mammograms and pinpoints with high accuracy the individual risk of metastasis in the armpit. A newly completed study shows that the model indicates that just over 40 per cent of today’s axillary surgery procedures could be avoided.

In breast cancer cases, lymph nodes in the armpit (known as axillary lymph nodes) are examined to assess metastatic spread and thus the prognosis and choice of treatment. A minor operation is performed, in which the first lymph nodes are identified and surgically removed for analysis. The procedure is minor but may cause pain, swelling, numbness and sometimes fluid collection. The spread of cancer to the armpit affects approximately one in five breast cancer patients. The remaining 80 per cent or so have no trace of cancer in the lymph nodes. In this case the procedure is purely diagnostic, with no therapeutic effect.

“In this study our focus has been on predicting the risk of metastasis in the armpit,” states Lisa Rydén, professor of surgery at Lund University and senior consultant at Skåne University Hospital.

A considerable emphasis in research has been placed on finding non-invasive diagnostic methods that can outline lymph node status at an earlier stage than is possible today. Lisa Rydén is responsible for the NILS (Non-Invasive Lymph node Staging) research project, a multiple-year research collaboration between the Faculty of Medicine and Faculty of Science at Lund University.

“At present, a separate procedure is performed on the armpit, known as a sentinel lymph node biopsy, to determine the spread to the lymph nodes. Using NILS as the starting point and patient and tumour data as a basis, we would instead be able to make a more individually based risk assessment before the operation. If the risk is low, axillary surgery, after a dialogue with the patient, could be avoided. If the risk is high, we would plan for surgery. It would be a step towards more person-centred care in which each action has a clear benefit for that specific patient,” says Lisa Rydén.

In the study, the researchers created an AI model that was trained to analyse mammograms. The mammograms are taken routinely in connection with breast cancer diagnostic work up, and thus entail no extra measures or costs. Images from 1,265 women in Skåne, diagnosed with breast cancer between 2009 and 2017, were used. The common denominator for the study participants was that their breast cancer was at an early stage and that surgery was the first treatment. The results of the study have been published in the research journal NPJ Digital Medicine.

The AI model was trained to identify different types of information – from the whole mammogram, not just the part that showed the tumour. It is on the basis of this complex information that the model could then calculate the risk of metastasis.

“We developed our algorithm in three steps. Firstly, the AI model went through tens of  thousands of mammograms to learn their basic structure, such as edges, texture and shapes. The AI model was then trained to find specific clues for cancer, such as the boundaries of tumours. And finally it was given a “holistic mindset” by including other important patient information, like age and tumour type, in order to more accurately predict the risk of metastasis,” says Daqu Zhang, doctoral student at the Faculty of Science, Lund University.

The AI model was used to classify the lymph nodes as disease-free or not, and the researchers showed that sentinel lymph node biopsy could have been avoided in 41.7 per cent of the cases.

“This study indicates that by using the AI model instead of operating on all patients, we could identify some 40 per cent of the patients where it is possible to avoid the procedure,” explains Lisa Rydén.

Sweden stands out internationally for the frequent, public mammography screening it offers. Around 67 per cent of Swedish breast cancer cases are detected in mammography screening among women aged between 40 and 74. Examination invitations are sent out at 18 to 24-month intervals, depending on the age of the woman.

“Many countries do not have this kind of population-based screening. In order to claim that our results are universally applicable we need to have them validated, and thus confirmed, in an independent way. So we are now searching high and low for relevant independent patient material both in Sweden and abroad,” says Lisa Rydén. 

She hopes that in the future the AI model’s built-in calculation algorithms can be used at the mammography examination stage to assess the risk of lymph node metastasis, and that the treatment can then be promptly designed based on this risk.

“In a dream scenario it would be possible to obtain a lot more information on the tumour stage and prognosis from the diagnostic mammograms. Our article focuses on the spread to the lymph nodes, but in ongoing international studies the image pattern could also predict the prognosis,” says Lisa Rydén.