Title: Classification of vertebral fractures in CT lumbar vertebrae
Authors: Adela Arpitha; Lalitha Rangarajan
Addresses: Department of Studies in Computer Science, University of Mysore, Mysuru – 570006, India ' Department of Studies in Computer Science, University of Mysore, Mysuru – 570006, India
Abstract: The existence of a vertebral fracture (VF) in particular compression fracture indicates osteoporosis and is a sole powerful predictor for the advancement of another osteoporotic fracture. With the number of imaging scans consistently expanding, identifying different cases and grades of osteoporotic fractures are missed by the over-burdened radiologist. The objective of this paper is to automatically segment and classify vertebral body fractures. Individual vertebral body is segmented by feeding preprocessed images to hybrid FCK-means algorithm. The shape features from the segmented output and texture features from the original input image are extracted and fed to an artificial neural network (ANN) which performs multi-class classification of vertebral body compression fractures and its associated fracture grades. Our method resulted in an overall classification accuracy of 93.14% based on Genant's scoring for VF. The result concludes that with this approach, the clinicians' task in diagnosing fractures is made simpler and also aids in suggesting for further treatment.
Keywords: artificial neural network; ANN; FCKmeans; vertebral body segmentation; vertebral body fracture classification; CT; shape features; texture features.
DOI: 10.1504/IJMEI.2021.115964
International Journal of Medical Engineering and Informatics, 2021 Vol.13 No.4, pp.279 - 288
Received: 30 Mar 2019
Accepted: 05 Oct 2019
Published online: 06 Jul 2021 *