SP32. Predicting The “Crunch”: Capsular Contraction Prediction in Two-stage Breast Reconstruction Using Neural Network

Abstract

PURPOSE: Two-stage breast reconstruction is a common technique used to restore pre-operative appearance in patients undergoing mastectomy. However, capsular contracture may develop and lead to significant discomfort. The objective of this study is to build a machine learning model that can evaluate the risk of capsular contracture after two-stage breast reconstruction. METHODS: 209 women (406 samples) were included in the cohort. Patient characteristics that remained statistically significant predictors of capsular contracture were included as input data in the machine learning model. Supervised learning models were evaluated using k-fold cross validation (k=3). A neural network model was also evaluated with a 0.8/0.1/0.1 train/validate/test split on the dataset. RESULTS: Among the subjects, 144/406 (35.47%) developed capsular contracture. Older age (OR 1.0), smaller nipple-inframammary fold distance (OR 0.90), slower tissue expansion rate/longer time delay (OR 0.998), and the use of post-operative radiation (OR 3.46) increased odds of capsular contracture (p<0.05). Neural network achieved the best performance metrics among the models tested, with test accuracy of 0.82 and ROC AUC 0.79. CONCLUSION: To our knowledge, this is the first study that uses neural network model to predict the development of capsular contraction after two-stage implant-based breast reconstruction. The application of machine learning will allow for better counseling of patients prior to breast reconstruction. High risk patients may be guided to autologous reconstruction or advised of the potential for additional surgeries. It may also provide better instruction on who should use acellular dermal matrix or other adjuvant techniques at the time of initial reconstruction.