Predicting Satisfaction with the Aesthetic Outcome after Breast Cancer Reconstruction Using Machine Learning: Prelimirary Results
Topuzov E.E. Skvortsov V.A. Orlova R.V. Talyshinskii A.E.
3 July 2025Autonomous non-profit scientific and medical organization
Voprosy Onkologii
2025#71Issue 3516 - 522 pp.
Aim. To develop and obtain preliminary performance met rics for a machine learning-based model that predicts aesthetic satisfaction of female patients after breast reconstruction, using clinical and anamnesis data. Materials and Methods. In the period from 2015 to 2024, information was retrospectively collected on 333 patients who had previously undergone complex treatment for breast can cer at the St. Petersburg State Clinical Oncologic Dispensary and received one-stage or delayed breast reconstruction during treatment. The inputs comprised quantitative and qualitative clinical and anamnesis data. Five different machine learning algorithms were compared: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR), XG Boost and Decision Trees (DT). Results. Logistic regression demonstrated the best perfor mance on all key metrics, including sensitivity (0.84) and ac curacy (0.73). Among patients who had undergone neoadjuvant chemotherapy (NACT) and/or radiation therapy (RT), the fac tor of final weight before surgery was the most prognostically significant, confirming the positive effect of this metric. The opposite results were obtained for initial weight, indicating that being overweight has an inherently negative effect on patient satisfaction after reconstruction. The surgeon’s experience, co morbidities, postoperative LT and preoperative disease stage were also important factors. The final ROC-AUC value was 0.7, which is acceptable for diagnostic systems under develop ment at an intermediate stage. Conclusion. The performance metrics obtained from the second opinion system for predicting satisfaction with the aes thetic outcome of breast cancer reconstruction are promising. This is despite the obvious limitations and approaches to level ling them so that other inputs can be included in the prognostic model, and so that the accuracy metrics can be reproduced in external testing.
AI , breast cancer , reconstruction , satisfaction
Text of the article Перейти на текст статьи
City Clinical Oncological Dispensary, St. Petersburg, Russian Federation
North-Western State Medical University named after I.I. Mechnikov, St. Petersburg, Russian Federation
St. Petersburg State University, St. Petersburg, Russian Federation
NpJSC Astana Medical University, Astana, Kazakhstan
LLC ‘Med-Ray’, Moscow, Russian Federation
City Clinical Oncological Dispensary
North-Western State Medical University named after I.I. Mechnikov
St. Petersburg State University
NpJSC Astana Medical University
LLC ‘Med-Ray’
10 лет помогаем публиковать статьи Международный издатель
Книга Публикация научной статьи Волощук 2026 Book Publication of a scientific article 2026