AUTOMATED SEGMENTATION OF REGIONS OF INTEREST ON HAND RADIOGRAPHS

Most radiologists do not use texture information contained in the trab ecular patterns of hand radiographs to diagnose erosive changes and de mineralization due to systemic inflammatory diseases that affect the s keletal system. However, high-resolution digitization achievable by a laser digitizer now makes it possible to access texture information th at may not be perceived visually. We are studying the feasibility of c omputer-assisted early detection of these processes with particular at tention to patients with hyperparathyroidism. In this paper the method s used to extract a region of interest (ROI) for texture analysis are discussed. The techniques include multiresolution sensing, automatic a daptive thresholding, detection of orientation angle, and projection t aken perpendicular to the line of least second moment. The methods wer e tested on a database of 50 pairs of hand radiographs. We segmented t he middle and the index fingers with an average success rate of 83% pe r hand. For the segmented finger strips, we located ROIs on both the m iddle and the proximal phalanges correctly over 84% of the times. Text ure information was collected in the form of a concurrence matrix with in the ROI. This study is a prelude to evaluating the correlation betw een classification based on texture analysis and diagnosis made by exp erienced radiologists.