We introduce MvBody, a multi‑view hybrid Transformer that predicts cesarean‑section risk from optical 3D body scans acquired between weeks 31 and 38 of gestation together with self‑reported medical data. The network projects the raw point cloud onto several views, feeds each view into a shared Transformer encoder to capture global shape relations, and augments the standard cross‑entropy loss with a metric‑learning term $$L_{metric}=\frac{1}{N}\sum_{i}^{N}\|f_i - \mu_{y_i}\|_2^2$$ to improve generalization under limited data. Integrated Gradients are applied post‑hoc to obtain theoretically grounded explanations. On an independent test set the model reaches 84.62% accuracy and an AUC of 0.724, outperforming conventional machine‑learning baselines and recent 3D analysis approaches. Importance analysis highlights pre‑pregnancy weight, maternal age, obstetric history, prior cesarean, and body shape around the head‑shoulder region as key predictors. Review: This study demonstrates the feasibility of inexpensive 3D scanning for prenatal risk assessment and provides an interpretable decision‑support tool suitable for resource‑constrained settings.