Developing an automated skeletal phenotyping pipeline to leverage biobank-level medical imaging databases

Area of Science:

• Medical imaging analysis
• Bioinformatics
• Anthropometry

Background:

• Collecting skeletal measurements from large medical imaging databases is time-consuming.
• Limited research utility of biobank-level data due to manual measurement challenges.

Purpose of the Study:

• To develop and validate an automated phenotyping pipeline for skeletal measurements from DXA scans.
• To compare automated measurements with manual data for accuracy and reliability.

Main Methods:

• Developed an automated pipeline using the Advanced Normalization Tools (ANTs) framework.
• Applied the pipeline to 341 UK Biobank South Asian DXA scans for 10 skeletal measurements.
• Validated performance using percent error and concordance correlation coefficients (CCC) on 20 additional scans.

Main Results:

• Automated measurements showed high accuracy for pelvic aperture breadth, bi-iliac breadth, femoral length, and tibia length (high agreement).
• Variable accuracy was observed for trunk length, upper thoracic breadth, and innominate height.
• Poor accuracy was noted for sacral and acetabular breadths.
• Stature regressions using automated measurements aligned with published data for South Asian populations.

Conclusions:

• A subset of skeletal measurements can be reliably extracted from DXA scans using the automated pipeline.
• This enhances the research potential of large biobank datasets for biological anthropologists and medical researchers.
• The pipeline offers a scalable solution for skeletal phenotyping.