top of page

Topic:

Limb Length Discrepancy

Issue:

Leg

Category:

Research

Title:

Fully Automated Analysis of the Anatomic and Mechanical Axes From Pediatric Standing Lower Limb Radiographs Using Convolutional Neural Networks

Author:

Murad, Yousif MD; Chhina, Harpreet PhD; Cooper, Anthony FRCSC

Journal:

Journal of Pediatric Orthopaedics

Date:

April 2024

Reference:

44(4):p 244-253, DOI: 10.1097/BPO.0000000000002611

Level Of Evidence:

​

# of Patients:

Not specified

Study Type:

Machine learning development and validation study

Location:

Single-center study (exact location not mentioned)

Summary:

This study evaluated the use of convolutional neural networks (CNNs) for fully automated analysis of pediatric weight-bearing lower limb radiographs. The CNN-based approach measured parameters like mechanical axis deviation, lateral distal femoral angle, and medial proximal tibial angle and compared them to manual measurements by orthopaedic fellows.

Methods:

CNNs were employed to segment radiographs and extract anatomic landmarks. Custom Matlab code analyzed these landmarks to compute limb alignment parameters. Automated measurements were compared with manual measurements by orthopaedic fellows. Mean deviations and calculation times were recorded.

Exclusions:

Not specified

Results:

Mechanical axis deviation: Mean deviation of 2.02 mm compared to manual measurements. Lateral distal femoral angle: Mean deviation of 1.73°. Medial proximal tibial angle: Mean deviation of 2.90°. Average calculation speed: ~2 seconds per radiograph.

Conclusions:

The CNN-based approach produced results comparable to manual measurements by fellows and did so much faster. With further validation and refinement, CNNs could become a valuable tool for automating repetitive radiographic tasks, particularly in clinical research and large-scale studies.

Relevance:

Limitations:

Perspective:

bottom of page