Automated Deep Learning Framework Enhances Pelvimetry Accuracy

A novel automated pelvimetry approach using deep learning shows promise in analyzing pelvic dimensions from MRI for total mesorectal excision (TME) difficulty assessment. The framework employs a 3D U-Net to precisely localize anatomical landmarks, achieving a mean localization error of 5.6 mm and demonstrating strong correlations (0.7 to 0.87) between automated and manual measurements. This advancement addresses the challenge of labor-intensive manual pelvimetry, offering a more efficient option for predictive assessments in surgical planning.