Automated selection of bone texture regions on hand radiographs: Data from the Osteoarthritis Initiative. Wolski, M., Englund, M., Stachowiak, G., & Podsiadlo, P. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine, 230(12):1117–1132, December, 2016.
Automated selection of bone texture regions on hand radiographs: Data from the Osteoarthritis Initiative. [link]Paper  doi  abstract   bibtex   
Manual selection of finger trabecular bone texture regions on hand X-ray images is time-consuming, tedious, and observer-dependent. Therefore, we developed an automated method for the region selection. The method selects square trabecular bone regions of interest above and below the second to fifth distal and proximal interphalangeal joints. Two regions are selected per joint (16 regions per hand). The method consists of four integral parts: (1) segmentation of a radiograph into hand and background, (2) identification of finger regions, (3) localization of center points of heads of distal phalanges and the distal interphalangeal, proximal interphalangeal, and metacarpophalangeal joints, and (4) placement of the regions of interest under and above the distal and proximal interphalangeal joints. A gold standard was constructed from regions selected by two observers on 40 hand X-ray images taken from Osteoarthritis Initiative cohort. Datasets of 520 images were generated from the 40 images to study the effects of hand and finger positioning. The accuracy in regions selection and the agreement in calculating five directional fractal parameters were evaluated against the gold standard. The accuracy, agreement, and effects of hand and finger positioning were measured using similarity index (0 for no overlap and 1 for entire overlap) and interclass correlation coefficient as appropriate. A high accuracy in selecting regions (similarity index ≥ 0.79) and a good agreement in fractal parameters (interclass correlation coefficient ≥ 0.58) were achieved. Hand and finger positioning did not affect considerably the region selection (similarity index ≥ 0.70). These results indicate that the method developed selects bone regions on hand X-ray images with accuracy sufficient for fractal analyses of bone texture.
@article{wolski_automated_2016,
	title = {Automated selection of bone texture regions on hand radiographs: {Data} from the {Osteoarthritis} {Initiative}.},
	volume = {230},
	issn = {0954-4119},
	shorttitle = {Automated selection of bone texture regions on hand radiographs},
	url = {https://doi.org/10.1177/0954411916676219},
	doi = {10.1177/0954411916676219},
	abstract = {Manual selection of finger trabecular bone texture regions on hand X-ray images is time-consuming, tedious, and observer-dependent. Therefore, we developed an automated method for the region selection. The method selects square trabecular bone regions of interest above and below the second to fifth distal and proximal interphalangeal joints. Two regions are selected per joint (16 regions per hand). The method consists of four integral parts: (1) segmentation of a radiograph into hand and background, (2) identification of finger regions, (3) localization of center points of heads of distal phalanges and the distal interphalangeal, proximal interphalangeal, and metacarpophalangeal joints, and (4) placement of the regions of interest under and above the distal and proximal interphalangeal joints. A gold standard was constructed from regions selected by two observers on 40 hand X-ray images taken from Osteoarthritis Initiative cohort. Datasets of 520 images were generated from the 40 images to study the effects of hand and finger positioning. The accuracy in regions selection and the agreement in calculating five directional fractal parameters were evaluated against the gold standard. The accuracy, agreement, and effects of hand and finger positioning were measured using similarity index (0 for no overlap and 1 for entire overlap) and interclass correlation coefficient as appropriate. A high accuracy in selecting regions (similarity index ≥ 0.79) and a good agreement in fractal parameters (interclass correlation coefficient ≥ 0.58) were achieved. Hand and finger positioning did not affect considerably the region selection (similarity index ≥ 0.70). These results indicate that the method developed selects bone regions on hand X-ray images with accuracy sufficient for fractal analyses of bone texture.},
	language = {en},
	number = {12},
	journal = {Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine},
	author = {Wolski, Marcin and Englund, Martin and Stachowiak, Gwidon and Podsiadlo, Pawel},
	month = dec,
	year = {2016},
	pages = {1117--1132},
}

Downloads: 0