A Machine Learning-Based Model for Breast Volume Prediction Using Preoperative Anthropometric Measurements

Background Accurate assessment of breast volume is helpful in preoperative planning and intraoperative judgment in both cosmetic and reconstructive breast surgery. In this prospective study, a formula was derived using machine learning algorithm (Gradient Boosted Model). Method A prospective study was performed on 39 female-to-male transgender patients. Bilateral mastectomy was done for all patients. Preoperative anthropometric measurements were performed on 78 breasts of these patients. Weight of breasts was calculated postoperatively with digital scale (weight), and then volume of breasts was calculated with the calibrated container (water displacement technique). Authors built a model based on Python CatBoostClassifier. Finally, an android application was built for ease of real-time utilization. Results Eight anthropometric measurements were collected preoperatively as independent variables. Breast vertical perimeter at lower half, upper pole, sternal notch to nipple and nipple to IMF had most correlation with volume and weight. Based on machine learning model, the following formula established: Breast volume = (breast width) × 24.69 + (nipple to IMF) × 49.03 − (sternal notch to nipple) × 1.34 + (anterior axillary line to medial border) × 6.57 − (upper pole) × 1.27 − (chest perimeter IMF) × 5.63 + (chest perimeter nipple) × 10.40 + (breast vertical perimeter at lower half) × 9.20 − 1133.74. The R 2 of the model is 0.93, and RMSE is 62.4. Conclusion Our formula is an accurate method for preoperative breast volume assessment. We built an android App (Breast Volume Predictor) for the real-time utilization of resulting formula. It is available at Google Play Store for free download. Level of Evidence IV This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .