Predicting postpartum diastasis recti abdominis using early-pregnancy abdominal wall ultrasound radiomics

Diastasis recti abdominis (DRA) affects abdominal wall stability and is associated with low back pain and pelvic floor dysfunction. Early identification of high-risk individuals is important for prevention and individualized management.

To develop and validate a clinical-radiomics model based on early-pregnancy two-dimensional grayscale ultrasound images for predicting DRA at 42 days postpartum and establishing a risk stratification system.

This prospective study included 428 pregnant women at 5-8 gestational weeks, randomly divided into training (n = 300) and testing (n = 128) cohorts. Radiomic features were extracted from ultrasound images of the rectus abdominis and linea alba to construct Radscore1 and Radscore2. These scores were integrated with early-pregnancy BMI and exercise habits to develop a combined model. Model performance was evaluated using AUC, 10-fold cross-validation, Hosmer-Lemeshow test, calibration curves, and decision curve analysis. A risk scoring system was constructed from regression coefficients. Spearman correlation analysis assessed associations with postpartum inter-rectus distance (IRD).

Radscore1, Radscore2, early-pregnancy BMI, and regular exercise were independently associated with DRA, with regular exercise showing a protective association. The combined model achieved AUCs of 0.892 and 0.859 in the training and testing cohorts, respectively. In the scoring system, BMI ≥24 kg/m², absence of exercise habits, Radscore1 ≥ 2.73, and Radscore2 ≥ 1.85 were assigned 1, 1, 3, and 1 points, respectively. DRA incidence increased significantly with risk level (trend test, P < 0.001). Radscore1, Radscore2, early-pregnancy BMI, and rectus abdominis thickness were positively correlated with postpartum IRD.

The clinical-radiomics model and risk scoring system accurately predicted DRA at 42 days postpartum and enabled early risk stratification, providing a quantitative basis for individualized intervention and follow-up. This is a single-center study; the findings require validation in independent multicenter cohorts before clinical implementation.