Machine learning models can define clinically relevant bone density subgroups based on patient-specific calibrated
Daniel Ritter1, Patrick J Denard2, Patric Raiss3
1Department of Orthopedic Research, Arthrex, Munich, Germany; Department of Orthopaedics and Trauma Surgery, Musculoskeletal University Center Munich (MUM), University Hospital, LMU, Munich, Germany.
Journal of Shoulder and Elbow Surgery
|August 18, 2024
Summary
Accurate bone density assessment before reverse shoulder arthroplasty (RSA) is crucial. This study introduces a patient-specific calibration method using computed tomography (CT) scans and machine learning to objectively quantify humeral bone quality, aiding surgical planning.
Area of Science:
- Orthopedic Surgery
- Radiology
- Biomedical Engineering
Background:
- Reduced bone density is a known risk factor for complications in reverse shoulder arthroplasty (RSA).
- Preoperative computed tomography (CT) imaging aids in implant selection, but lacks reproducible methods for bone density quantification.
- Objective assessment of humeral bone quality is needed for improved surgical outcomes.
Purpose of the Study:
- To develop and validate a patient-specific calibration method for bone density analysis in RSA patients using preoperative CT scans.
- To hypothesize that CT-derived bone density measures can objectively quantify humeral bone quality.
- To apply machine learning models for improved bone density classification.
Main Methods:
- A 3-part study involving cadaveric CT analysis, retrospective clinical RSA cohort analysis (n=345), and machine learning (ML) models.
- Patient-specific calibration using CT scans was compared to standard Hounsfield units.
- Hierarchical clustering and support vector machine models were employed for bone density classification.
Main Results:
- Patient-specific calibration demonstrated high accuracy for cancellous bone density (ICC >0.75).
- ML clustering identified distinct high-density (96 patients) and low-density (146 patients) subgroups.
- Support vector machine achieved high prediction accuracy (91.2% training, 90.5% testing) for bone densities.
Conclusions:
- Preoperative CT scans with patient-specific calibration and ML models can accurately quantify proximal humeral bone quality in RSA patients.
- This objective bone quality assessment provides valuable preoperative information for surgeons.
- The developed method enhances the prediction of potentially poor bone quality, optimizing surgical planning.
