E-Poster Presentation ESA-SRB-ANZBMS 2021

Identification of asymptomatic vertebral fracture: A novel shape-based algorithm (#754)

Huy G Nguyen 1 , Hoa T Nguyen 2 , Thach S Tran 3 , Steve H Ling 1 , Tuan V Nguyen 1 4
  1. School of Biomedical Engineering, The University of Technology Sydney, Sydney, NSW, Australia
  2. Can Tho University, Can Tho, Vietnam
  3. Biology of Bone, The Garvan Institute of medical research, Sydney, NSW, Australia
  4. School of Population Health, University of New South Wales, Sydney, NSW, Australia

Background: Most vertebral fractures are asymptomatic, with more than two-thirds of vertebral fractures are undiagnosed. We sought to develop an automated shape-based algorithm to identify asymptomatic vertebral fractures.

Aims: To quantify the accuracy of the algorithm in the identification of vertebral fractures in men and women.

Methods: The study involved 106 individuals (109 lateral spinal X-rays), among whom 53 patients were clinically diagnosed by a rheumatologist to have a fracture. A shape-based algorithm was designed to identify 4 vertices in the segmented vertebra according to its contour and centroid by maximising the area formed by the vertices; anterior and posterior heights were then calculated for classification based on Genant's classification system. The accuracy of the algorithm was assessed in terms of sensitivity and specificity using the clinician's diagnosis as standard.

Results: The mean (SD) age of the 106 individuals was 57.1 (10.6 years). Among whom, 48 were diagnosed to have a fracture, and the algorithm identified 46, a sensitivity of 96%. Among the 61 without a fracture, the algorithm identified 19 as having a fracture, making the false positive rate of 31%. Among the 663 vertebrae examined, 55 were diagnosed to have a fracture, and the algorithm identified 45 (sensitivity of 82%). Among the 608 vertebrae without a fracture, the algorithm identified 67 as having a fracture (false positive of 11%). Further analysis of concordance between the Genant's classification and the algorithm scores showed that the accuracy was good for non-fracture (concordance of 98%) and severe fracture (43%), but not for mild and moderate fracture (17%).

Conclusion: The automated shape-based algorithm has good sensitivity and specificity for identifying asymptomatic vertebral fractures. The algorithm can help clinicians to screen a large volume of lateral spinal X-rays in clinical settings.

61166872da417-Examples+of+diagnosis+on+6+films.png