Combining deep learning with statistical shape modelling enables automated lower limb measurements with...

Area of Science:

• Orthopedics and Biomechanics
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Accurate lower limb alignment assessment is vital for diagnosing deformities and planning treatments.
• Traditional 2D methods lack 3D bone morphology insights and are position-dependent.
• Manual landmark identification on 3D weight-bearing CT (WBCT) is time-consuming and prone to variability.

Purpose of the Study:

• To develop and validate a deep learning (DL) and statistical shape modeling (SSM) hybrid approach for automated 3D lower limb alignment and morphology assessment.
• To enable observer-independent 3D evaluations under physiological loading conditions.
• To improve the efficiency and reliability of lower limb analysis using WBCT.

Main Methods:

• A hybrid DL-SSM model was developed to automatically segment key lower limb bones (femur, patella, tibia, talus, calcaneus, second metatarsal) from 30 full-leg WBCT scans.
• The model automatically identified 3D landmarks to derive 28 lower limb alignment and morphology measurements.
• The automated measurements were validated against manual measurements from three experienced raters.

Main Results:

• The DL segmentation model achieved high accuracy (Dice similarity coefficient > 0.96).
• Automated measurements showed good agreement with manual assessments, with differences comparable to interobserver reliability.
• Mean absolute errors for angular measurements ranged from 0.35° ± 0.39° to 5.53° ± 4.68°.

Conclusions:

• The hybrid DL-SSM methodology provides a reliable and observer-independent tool for 3D lower limb alignment and morphology assessment.
• This automated approach enhances the evaluation of lower limb alignment under weight-bearing conditions.
• The findings support the clinical utility of AI-driven tools in orthopedic assessment.