Deep learning-based prediction of mastectomy and reconstruction weights using three-dimensional CT and MRI: a retrospective proof-of-concept study

Background To evaluate the feasibility of deep learning-based prediction of mastectomy specimen and final reconstruction weights from three-dimensional CT and MRI, including multimodal and clinical-feature fusion. Methods This retrospective single-center study included 200 patients who underwent total mastectomy and TRAM (transverse rectus abdominis myocutaneous) flap breast reconstruction. Preoperative imaging was used for mastectomy-weight prediction and postoperative imaging for reconstruction-weight prediction. Three-dimensional deep learning models and CT-MRI feature fusion were evaluated with patient-level five-fold cross-validation, and clinical-feature fusion was additionally explored. Results The paired cohorts comprised 184 patients for mastectomy-weight prediction and 146 for reconstruction-weight prediction. MRI with ResNet3D-18 showed the lowest error for mastectomy weight (mean absolute error [MAE], 134.3 g; 95% confidence interval [CI], 119.2–151.4; weighted absolute percentage error, 28.3%), whereas CT-MRI fusion showed the lowest error for reconstruction weight (MAE, 123.0 g; 95% CI, 103.3–144.0; weighted absolute percentage error, 23.2%). Adding clinical features minimally affected mastectomy-weight prediction but numerically improved reconstruction-weight prediction (MAE, 123.0 to 108.2 g), although the 95% CI for the improvement crossed zero. Predictions tended toward the cohort mean at extreme weights. Conclusions Three-dimensional CT and MRI enabled direct deep learning-based prediction of measured surgical weights without explicit volumetric measurement, demonstrating the feasibility of direct image-to-weight prediction in breast reconstruction.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-20
DOI
https://doi.org/10.21203/rs.3.rs-11239746/v1
Primary Topic
Breast Implant and Reconstruction
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Deep learning-based prediction of mastectomy and reconstruction weights using three-dimensional CT and MRI: a retrospective proof-of-concept study

Taesu Cheong, D.‐H. Shin, Sumin Jung, Deok-Woo Kim et al.
1 links
Research Square
Breast Implant and Reconstruction
preprint

Deep learning-based prediction of mastectomy and reconstruction weights using three-dimensional CT and MRI: a retrospective proof-of-concept study

Taesu Cheong, D.‐H. Shin, Sumin Jung, Deok-Woo Kim, Tae-Yul Lee, Hi-Jin You, Han-Sang Song
preprint en

Abstract

Background To evaluate the feasibility of deep learning-based prediction of mastectomy specimen and final reconstruction weights from three-dimensional CT and MRI, including multimodal and clinical-feature fusion. Methods This retrospective single-center study included 200 patients who underwent total mastectomy and TRAM (transverse rectus abdominis myocutaneous) flap breast reconstruction. Preoperative imaging was used for mastectomy-weight prediction and postoperative imaging for reconstruction-weight prediction. Three-dimensional deep learning models and CT-MRI feature fusion were evaluated with patient-level five-fold cross-validation, and clinical-feature fusion was additionally explored. Results The paired cohorts comprised 184 patients for mastectomy-weight prediction and 146 for reconstruction-weight prediction. MRI with ResNet3D-18 showed the lowest error for mastectomy weight (mean absolute error [MAE], 134.3 g; 95% confidence interval [CI], 119.2–151.4; weighted absolute percentage error, 28.3%), whereas CT-MRI fusion showed the lowest error for reconstruction weight (MAE, 123.0 g; 95% CI, 103.3–144.0; weighted absolute percentage error, 23.2%). Adding clinical features minimally affected mastectomy-weight prediction but numerically improved reconstruction-weight prediction (MAE, 123.0 to 108.2 g), although the 95% CI for the improvement crossed zero. Predictions tended toward the cohort mean at extreme weights. Conclusions Three-dimensional CT and MRI enabled direct deep learning-based prediction of measured surgical weights without explicit volumetric measurement, demonstrating the feasibility of direct image-to-weight prediction in breast reconstruction.

Research Square
Korea University (KR), Korea University Ansan Hospital (KR)
Korea University
Good health and well-being
Breast Implant and Reconstruction
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon