Deep learning to predict muscle, nerve and pain problems after breast cancer surgery

Lymphedema can occur at any stage of a patient's life following breast cancer. Patients with breast cancer-related lymphedema require lifelong treatment, and as the stage of lymphedema progresses, response to therapy decreases. Advanced stages of lymphedema negatively affect functional status, and patients experience difficulties in performing activities of daily living.

Axillary web syndrome (AWS) is characterized by a taut cord extending from the axilla to the volar surface of the wrist, typically appearing within the first 8 weeks postoperatively. AWS can complicate the administration of radiotherapy. Shoulder dysfunction may occur independently or in association with AWS. In particular, scapular dyskinesis developing after mastectomy can lead to secondary shoulder conditions such as rotator cuff syndrome or adhesive capsulitis, which are commonly observed in these patients.

Peripheral neuropathy is frequently seen in patients receiving chemotherapy, adversely affecting daily life and sometimes preventing continuation of treatment. Other complications related to chemotherapy and radiotherapy include cardiotoxicity, pulmonary toxicity, fatigue, osteoporosis, and cognitive impairment.

There are also specific painful syndromes that may occur after breast cancer, including post-mastectomy pain syndrome, phantom breast pain, and musculoskeletal symptoms associated with aromatase inhibitors. All these conditions can significantly impair daily functioning and even hinder continuation of cancer treatment. Therefore, predicting these complications and implementing or developing preventive interventions is crucial.

If it is possible to predict the early development of lymphedema, axillary web syndrome, peripheral neuropathy, and painful syndromes after breast cancer, early intervention may prevent progression. This study is designed to develop and validate a predictive model using deep learning methods to determine the risk of these complications in patients undergoing breast cancer surgery. Among deep learning architectures, ResNet50, AlexNet, GoogleNet, and UNet, which have been widely used in recent studies, are planned to be implemented.

Additionally, based on the results of this study, a risk calculation program will be developed, allowing clinicians to input baseline patient data and calculate the individual patient's risk for each complication prior to treatment. No specific risk is expected in the study.