A deep learning model for preoperative differentiation of uterine adenomyoma and uterine leiomyoma on transvaginal ultrasound

Introduction To develop and validate a deep learning model for the preoperative differentiation of uterine adenomyoma and uterine leiomyoma on transvaginal ultrasound. Methods This retrospective study included 207 transvaginal ultrasound images from 98 patients between January 2020 and January 2025, split at the patient level into a training set (68 patients, 140 images) and an independent test set (30 patients, 67 images). A comparative framework was established: (1) radiomics models; (2) standard deep learning models; and (3) the proposed ResNet-50 model enhanced with the channel-aware feature modulation module. Three gynecologic ultrasonographers independently reviewed the test images for comparison with model performance. Model performance was evaluated using accuracy, precision, sensitivity, specificity, F1-score, and the area under the receiver operating characteristic curve (AUC). Exploratory explainability analyses used SHAP for the best-performing radiomics model and Grad-CAM for the proposed model. Results On the patient-level test set, the proposed model showed favorable numerical performance among the evaluated models, with an accuracy of 0.833, F1-score of 0.800, and AUC of 0.855 (95% CI: 0.693–0.982). The best-performing radiomics model, Logistic Regression, achieved a patient-level accuracy of 0.667 and AUC of 0.814 (95% CI: 0.630–0.962). The proposed model showed numerically higher accuracy than the three readers. SHAP summarized the contributions of radiomic features, whereas Grad-CAM visualized image regions associated with model predictions. Conclusion The proposed model showed promising internal test performance for differentiating uterine adenomyoma from uterine leiomyoma on transvaginal ultrasound and may serve as a potential decision-support approach, warranting external and prospective validation.

Authors

Institutions

Publication Details

Journal
Frontiers in Medicine
Published
2026-11-12
DOI
https://doi.org/10.3389/fmed.2026.1918669
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A deep learning model for preoperative differentiation of uterine adenomyoma and uterine leiomyoma on transvaginal ultrasound

Xiaobo Lai, Jiajun Wang, Qian Lü, Liming Yang et al.
Frontiers in Medicine
Radiomics and Machine Learning in Medical Imaging
article

A deep learning model for preoperative differentiation of uterine adenomyoma and uterine leiomyoma on transvaginal ultrasound

Xiaobo Lai, Jiajun Wang, Qian Lü, Liming Yang, Zicheng Zeng, Peilin Sheng, Jianmin Wang
article en

Abstract

Introduction To develop and validate a deep learning model for the preoperative differentiation of uterine adenomyoma and uterine leiomyoma on transvaginal ultrasound. Methods This retrospective study included 207 transvaginal ultrasound images from 98 patients between January 2020 and January 2025, split at the patient level into a training set (68 patients, 140 images) and an independent test set (30 patients, 67 images). A comparative framework was established: (1) radiomics models; (2) standard deep learning models; and (3) the proposed ResNet-50 model enhanced with the channel-aware feature modulation module. Three gynecologic ultrasonographers independently reviewed the test images for comparison with model performance. Model performance was evaluated using accuracy, precision, sensitivity, specificity, F1-score, and the area under the receiver operating characteristic curve (AUC). Exploratory explainability analyses used SHAP for the best-performing radiomics model and Grad-CAM for the proposed model. Results On the patient-level test set, the proposed model showed favorable numerical performance among the evaluated models, with an accuracy of 0.833, F1-score of 0.800, and AUC of 0.855 (95% CI: 0.693–0.982). The best-performing radiomics model, Logistic Regression, achieved a patient-level accuracy of 0.667 and AUC of 0.814 (95% CI: 0.630–0.962). The proposed model showed numerically higher accuracy than the three readers. SHAP summarized the contributions of radiomic features, whereas Grad-CAM visualized image regions associated with model predictions. Conclusion The proposed model showed promising internal test performance for differentiating uterine adenomyoma from uterine leiomyoma on transvaginal ultrasound and may serve as a potential decision-support approach, warranting external and prospective validation.

Frontiers in MedicineVol. 13
Zhejiang Chinese Medical University (CN), Sir Run Run Shaw Hospital (CN), Zhejiang University (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
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