Artificial Intelligence In The Diagnosis And Prediction Of Breast Cancer-Related Lymphedema: A Scoping Review.

INTRODUCTION

Lymphedema is a chronic complication of breast cancer treatments that can significantly impact the well-being of survivors. This scoping review aims to evaluate the role of artificial intelligence (AI) in enhancing diagnostic precision and supporting timely interventions.

METHODS

PubMed, Scopus, EMBASE, Web of Science, and the Cochrane Central Register of Controlled Trials were searched from database inception until September 2024. Studies were included if they examined AI-based techniques for detecting lymphedema, assessing severity, and/or facilitating early diagnosis. Inclusion criteria required studies to be published in English with full-text availability. Editorials, review papers, and inaccessible full-text studies were excluded.

RESULTS

From 300 studies identified from the database search, 13 studies met the inclusion criteria. The investigated AI-based models used input data such as Electronic Health Record (EHR) and clinical data (5 studies, 38.5%), patient-reported symptoms and demographics (4 studies, 30.8%), imaging (3 studies, 23.1%), and clinical factors like BMI and hypertension (2 studies, 15.4%) for outcome prediction. The most commonly used AI model was built using the Support Vector Machine (SVM) algorithm, which appeared in 8 studies (61.5%) and was often combined with other supervised learning techniques. Risk prediction models achieved accuracy rates of 81% to 93.75%, with sensitivity of 95.65%, specificity of 91.03%, and Area Under the Curve (AUC) of 0.751. Models classifying lymphedema severity demonstrated accuracy rates between 81 and 91%, with the best-performing models achieving a balanced accuracy of 99.4% and AUC ranging from 0.889 to 0.931.

CONCLUSIONS

AI demonstrates potential in improving the diagnosis and prediction of breast cancer-related lymphedema, offering enhanced diagnostic capabilities and personalized interventions. Further research is required to address data standardization, model validations, and the development of frameworks for clinical implementation.