EP4.11 Accurate CT Segmentation and Reliable 3D Morphometric Measurements to Support Clinical Interpretation in FAI

Abstract

Abstract Introduction Accurate and reliable hip morphometry measurements can be used for clinical decision-making. This study assessed the accuracy and reliability of the segmentation process and the reliability of resulting 3D morphometric measurements used for creating a HipMap FAI Analysis. Methods Segmentation accuracy was assessed by comparing binary DICOM masks and 3D surface reconstructions from two internal raters against a reference segmentation generated by external experts (3D Medical Systems). Dice and Jaccard scores were calculated to quantify how closely internal segmentations matched the expert reference, while sensitivity assessed segmentation completeness and precision assessed segmentation correctness. Segmentation reliability was evaluated between the two internal raters and across two iterations of one rater. Inter-rater and intra-rater reliability of HipMap FAI analyses were additionally assessed using intra class correlation coefficients. The study included 10 patients (5 male, 5 female; age range 17–53 years) with femoroacetabular impingement (3 cam, 3 pincer, and 4 combined cam–pincer deformities based on radiographic measurements). For measurement analyses, bilateral hip segmentations were required, although clinical measurements were ultimately calculated only for the pathological hip. Results Segmentation accuracy showed near identical agreement between internal and expert segmentations. Average mean surface distances between internal and external segmentations were <1 mm, with absolute mean distances of 0.15mm for the pelvis and 0.17mm for the femur. The distribution of surface differences remained minimal, with 5th percentile values of −0.22mm (pelvis) and −0.12mm (femur), and 95th percentile values of 0.35mm (pelvis) and 0.44mm (femur); all within 1 mm. Shape agreement metrics (Dice and Jaccard scores), as well as measures of segmentation completeness (sensitivity) and segmentation correctness (precision), were >0.97 for the pelvis and >0.98 for the femur, indicating a near voxel perfect match with the expert reference. Segmentation reliability was excellent across internal raters and repeated segmentations. Absolute mean surface distances were 0.05mm for the pelvis and 0.04mm for the femur between raters, and 0.05mm (pelvis) and 0.06mm (femur) across two iterations of one rater. Dice, Jaccard, sensitivity, and precision values were 1.0 for both the pelvis and femur in inter rater and intra rater comparisons, indicating voxel perfect reproducibility. HipMap FAI measurements demonstrated excellent reliability. Intra rater reliability ranged from 0.97 to 1.0, and inter rater reliability ranged from 0.89 to 1.0. The average standard deviation between two iterations and between two raters was 0.53° and 2.03° for the alpha angle, 0.28° and 0.43° for femoral torsion, 0.14° and 0.29° for neck–shaft angle, 0.14° and 0.60° for center edge angle, 0.42° and 0.80° for Tönnis angle, and 0.28% and 0.45% for acetabular coverage, respectively. Conclusion Our results demonstrate that HipMap FAI analysis provides highly accurate segmentation and includes reliable 3D morphometric measurements that can be used for clinical interpretation.