Open Access

ARTICLE

Novel Statistical Shape Relation and Prediction of Personalized Female Pelvis, Pelvic Floor, and Perineal Muscle Shapes

Tan-Nhu Nguyen^1,2, Trong-Pham Nguyen-Huu^1,2, Tien-Tuan Dao^3,*

1 School of Biomedical Engineering, International University, Ho Chi Minh City, Vietnam
2 Vietnam National University, Ho Chi Minh City, Vietnam
3 Univ. Lille, CNRS, Centrale Lille, UMR 9013-LaMcube-Laboratoire de Mécanique, Multiphysique, Multiéchelle, Lille, France

* Corresponding Author: Tien-Tuan Dao. Email:

Computer Modeling in Engineering & Sciences 2026, 146(2), 1 https://doi.org/10.32604/cmes.2026.075386

Received 30 October 2025; Accepted 16 January 2026; Issue published 26 February 2026

Abstract

Vaginal delivery is a fascinating physiological process, but also a high-risk process. Up to 85%–90% of vaginal deliveries lead to perineal trauma, with nearly 11% of severe perineal tearing. It is a common occurrence, especially for first-time mothers. Computational childbirth plays an essential role in the prediction and prevention of these traumas, but fast personalization of the pelvis and floor muscles is challenging due to their anatomical complexity. This study introduces a novel shape-prediction-based personalization of the pelvis and floor muscles for perineal tearing management and childbirth simulation. 300 subjects were selected from public Computed Tomography (CT) databases. The pelvic bone nmjmeshes were generated using a coarse-to-fine non-rigid mesh alignment procedure. The floor muscle meshes were personalized using the bone mesh deformation information. A feature-to-pelvic structure reconstruction pipeline was proposed, incorporating various strategies. Ten-fold cross-validation helped determine the optimal reconstruction strategy, regression method, and feature sizes. The mesh-to-mesh distance metric was employed for evaluating. The statistical shape relation-based strategy, coupled with multi-output ridge regression, was the optimal approach for pelvic structure reconstruction. With a feature set ranging from 3 to 38, the mean errors were 2.672 to 1.613 mm, and 3.237 to 1.415 mm in muscle attachment regions. The best- and worst-case predictions had errors of 1.227 ± 0.959 mm and 2.900 ± 2.309 mm, respectively. This study provides a novel approach to achieving fast personalized childbirth modeling and simulation for perineal tearing management.

---

Keywords

---