Can Detecting 15 Landmarks on Full-length Lower-limb X-rays Reproduce Alignment Angles Automatically?

Angle-specific inter-observer ICCs for the automatic model were ≥0.98; repeat-measurement ICC was 1.00.

This single-institution study used 10,407 training, 500 validation and 305 test images. The figure shows the data split. Three-institution external validation was performed separately in the 2024 follow-up study.
KSSTA (2023). DOI: 10.1007/s00167-022-07124-x · Original explanatory chart based on reported data
Evidence & Original Source Image use terms

The research question

Participants and Methods

After separating the hip, knee, and ankle ROIs, the model detected 15 anatomical landmarks to calculate mLDFA, MPTA, JLCA, and HKAA. Inter- and intra-observer ICCs, mean absolute errors, and outliers were assessed against specialist reference values. The dataset contained 11,212 full-length lower-limb radiographs: 10,407 for training, 500 for validation, and 305 for testing. Right-leg images were used to reduce duplication in training.

Main Findings

The automated model’s inter-observer ICC was ≥0.98 for each angle, and the repeat-measurement ICC was 1.00. Excluding JLCA, mean absolute errors were below 0.52° and outlier rates were below 4.26%. Definitions and errors differed by angle. Numerical reproducibility and failures in landmark recognition should be examined together.

Implications for clinical and surgical practice

The study supports standardizing mLDFA, MPTA, and HKAA measurements used at the start of HTO and TKA planning. Automatic measurement does not replace standardized imaging, interpretation of intra-articular deformity, or the surgeon’s final plan.

Limitations

This was retrospective internal validation at a single center; performance at external institutions was not directly evaluated. The development data focused on right lower limbs, and image quality and positioning rotation must be considered. JLCA should not be grouped with the other angles to claim that all measurement errors were below 0.52°.

Let's put it briefly.

Which clinical question does this address?

Keywords

Angle measurement
Deep learning
Hip knee ankle angle (HKAA)
Joint line convergence angle (JLCA)
Landmark recognition
Lower extremity
Mechanical lateral distal femoral angle (mLDFA)
Medial proximal tibial angle (MPTA)

Abstract · Original text

Changwung Jo, Doohyun Hwang, Sunho Ko, Myung Ho Yang, Myung Chul Lee, Hyuk-Soo Han, Du Hyun Ro. Deep learning-based landmark recognition and angle measurement of full-leg plain radiographs can be adopted to assess lower extremity alignment. Knee surgery, sports traumatology, arthroscopy : official journal of the ESSKA. 2023;31(4):1388-1397. doi:10.1007/s00167-022-07124-x.

PURPOSE

Evaluating lower extremity alignment using full-leg plain radiographs is an essential step in diagnosis and treatment of patients with knee osteoarthritis. The study objective was to present a deep learning-based anatomical landmark recognition and angle measurement model, using full-leg radiographs, and validate its performance.

METHODS

A total of 11,212 full-leg plain radiographs were used to create the model. To train the data, 15 anatomical landmarks were marked by two orthopaedic surgeons. Mechanical lateral distal femoral angle (mLDFA), medial proximal tibial angle (MPTA), joint line convergence angle (JLCA), and hip-knee-ankle angle (HKAA) were then measured. For inter-observer reliability, the inter-observer intraclass correlation coefficient (ICC) was evaluated by comparing measurements from the model, surgeons, and students, to ground truth measurements annotated by an orthopaedic specialist with 14 years of experience. To evaluate test-retest reliability, all measurements were made twice by each measurer. Intra-observer ICCs were then derived. Performance evaluation metrics used in previous studies were also derived for direct comparison of the model's performance.

RESULTS

Inter-observer ICCs for all angles of the model were 0.98 or higher (p < 0.001). Intra-observer ICCs for all angles were 1.00, which was higher than that of the orthopaedic specialist (0.97-1.00). Measurements made by the model showed no significant systemic variation. Except for JLCA, angles were precisely measured with absolute error averages under 0.52 degrees and proportion of outliers under 4.26%.

CONCLUSIONS

The deep learning model is capable of evaluating lower extremity alignment with performance as accurate as an orthopaedic specialist with 14 years of experience.

LEVEL OF EVIDENCE

III, retrospective cohort study.

Read abstract on PubMed

The original paper.

Deep learning-based landmark recognition and angle measurement of full-leg plain radiographs can be adopted to assess lower extremity alignment

Jo C, Hwang D, Ko S, Yang MH, Lee MC, Han HS, Ro DH
KSSTA · 2023 · online 2022

DOI · 10.1007/s00167-022-07124-x PubMed Based data

Academic Discussion

Results explained with study numbers