A New Approach to Improve TMJ Morphological Information from Plain Film Radiographs

Abstract

The hard tissues of the temporomandibular joint (TMJ) can be assessed through radiographic imaging to provide information to assist in diagnosis and treatment. However, the quality of information gathered from such imaging is often less than desired due to the small size of the TMJ, the widely varying fossa and condylar morphology, and the surrounding dense osseous structures. These make a clear and undistorted radiographic image of the hard tissue of the joint technically difficult. It is postulated that, if the degree of inaccuracy of a given radiograph is known quantitatively and taken into account, the clinician will be able to make a better informed qualitative assessment of TMJ morphology. The aim of this study is: 1. to improve the qualitative information that can be acquired from routine clinical plain film radiographs of the TMJ; 2. to use quantitative data to test the novel Neural Network (NN) model; and 3. to statistically show the discrepancies between routine radiographic images and the actual joint. Linear measurements of excised TMJs and their radiographic images were used to train a multilayer perceptron (MP) type NN model to predict joint dimensions more accurately. Such a neural network, developed using a statistical software package such as SPSS (SPSS, Inc. Chicago, IL), functions as a computer software program and predicts joint dimensions with increased accuracy when radiographic measurements are entered into the program. The NN model described here predicts the actual linear distances of the TMJ more closely than radiographic measurements and is more reliable in assessing the TMJ morphology.