Abstract
The purpose of the study is to apply a neural network to detect changes of trabecular patterns on skeletal radiographs due to osteoporosis. Regions of interest (ROIs)(64×64 pixels, 6.4×6.4mm2) near the center of the vertebral body were extracted from digitized bone radiographs. After background trend corrections, texture measures were calculated in terms of standard deviations (total, horizontal, and vertical directions) and pixel values (maximum and minimum), all of which were used as input to the input layer of the neural network. The network was trained with a back-propagation technique to discriminate textural differences between the normal and the diseased cases. A data base of 46 bone radiographs was used in this study. Half of the cases was used as the training set, and the other half was used as the testing set. The neural network correctly classified about 90% of the cases, using the diagnosis made by an experienced physician as a standard for comparison. Our results suggest that a neural network analysis is useful to discriminate textural variations of trabecular patterns and may be effective for computer-aided diagnosis of osteoporosis.
