Keywords:
Performed at one institution, Diagnostic or prognostic study, Retrospective, Osteoporosis, Metabolic disorders, Demineralisation-Bone, Experimental investigations, Computer Applications-Detection, diagnosis, CAD, Digital radiography, Absorptiometry / Bone densitometry, Musculoskeletal bone, Computer applications, Artificial Intelligence, Information Technology
Authors:
A. Creeden; Coventry/UK
DOI:
10.26044/ecr2020/C-02138
Conclusion
A limitation of this study is that the radiographs were acquired using the x-ray systems of a single manufacturer. The test dataset also intentionally comprised equal proportions of each DXA classification, making it unrepresentative of any specific population. Use of a wider variety of x-ray systems and a more representative patient population would provide greater external validity.
The J48 algorithm demonstrated a modest improvement over baseline accuracy. The textural analysis and machine learning approach described shows some potential for the prediction of osteoporosis using routine pelvic radiographs but the accuracy obtained in this study would be insufficient for use in clinical practice.
Further research to investigate the effects of variations in radiographic exposure parameters (kV, mAs, detector resolution, magnification etc.) and patient positioning (hip rotation, abduction, flexion etc.) on textural measurements might allow prediction accuracy to be optimised. Repeating the study with a significantly larger training dataset may also improve the accuracy achieved.