Automatic selection of region of interest for radiographic texture analysis

Li Lan; Maryellen L. Giger; Joel R. Wilkie; Tamara J. Vokes; Weijie Chen; Hui Li; Tracy Lyons; Michael R. Chinander; Ann Pham

doi:10.1117/12.711531

Translator Disclaimer

30 March 2007 Automatic selection of region of interest for radiographic texture analysis

Li Lan, Maryellen L. Giger, Joel R. Wilkie, Tamara J. Vokes, Weijie Chen, Hui Li, Tracy Lyons, Michael R. Chinander, Ann Pham

Author Affiliations +

Proceedings Volume 6514, Medical Imaging 2007: Computer-Aided Diagnosis; 651436 (2007) https://doi.org/10.1117/12.711531
Event: Medical Imaging, 2007, San Diego, CA, United States

Abstract

We have been developing radiographic texture analysis (RTA) for assessing osteoporosis and the related risk of fracture. Currently, analyses are performed on heel images obtained from a digital imaging device, the GE/Lunar PIXI, that yields both the bone mineral density (BMD) and digital images (0.2-mm pixels; 12-bit quantization). RTA is performed on the image data in a region-of-interest (ROI) placed just below the talus in order to include the trabecular structure in the analysis. We have found that variations occur from manually selecting this ROI for RTA. To reduce the variations, we present an automatic method involving an optimized Canny edge detection technique and parameterized bone segmentation, to define bone edges for the placement of an ROI within the predominantly calcaneus portion of the radiographic heel image. The technique was developed using 1158 heel images and then tested on an independent set of 176 heel images. Results from a subjective analysis noted that 87.5% of ROI placements were rated as "good". In addition, an objective overlap measure showed that 98.3% of images had successful ROI placements as compared to placement by an experienced observer at an overlap threshold of 0.4. In conclusion, our proposed method for automatic ROI selection on radiographic heel images yields promising results and the method has the potential to reduce intra- and inter-observer variations in selecting ROIs for radiographic texture analysis.

Citation Download Citation

Li Lan, Maryellen L. Giger, Joel R. Wilkie, Tamara J. Vokes, Weijie Chen, Hui Li, Tracy Lyons, Michael R. Chinander, and Ann Pham "Automatic selection of region of interest for radiographic texture analysis", Proc. SPIE 6514, Medical Imaging 2007: Computer-Aided Diagnosis, 651436 (30 March 2007); https://doi.org/10.1117/12.711531

ACCESS THE FULL ARTICLE