Automated segmentation of regions of interest on hand radiographs

Most radiologists do not use texture information contained in the trabecular patterns of hand radiographs to diagnose erosive changes and demineralization due to systemic inflammatory diseases that affect the skeletal system. However, high-resolution digitization achievable by a laser digitizer now makes it possible to access texture information that may not be perceived visually. We are studying the feasibility of computer-assisted early detection of these processes with particular attention to patients with hyperparathyroidism. In this paper the methods used to extract a region of interest (ROI) for texture analysis are discussed. The techniques include multiresolution sensing, automatic adaptive thresholding, detection of orientation angle, and projection taken perpendicular to the line of least second moment. The methods were tested on a database of 50 pairs of hand radiographs. We segmented the middle and the index fingers with an average success rate of 83% per hand. For the segmented finger strips, we located ROIs on both the middle and the proximal phalanges correctly over 84% of the times. Texture information was collected in the form of a concurrence matrix within the ROI. This study is a prelude to evaluating the correlation between classification based on texture analysis and diagnosis made by experienced radiologists.