Normal and Acute Tympanic Membrane Diagnosis based on Gray Level Co-Occurrence Matrix and Artificial Neural Networks

Abstract

Otitis Media (OM) is the general name of middle ear inflammation. In order to diagnose this disease, it is important to examine the middle ear tympanic membrane (TM) by a standard otoscopy device. In recent years, biomedical image processing and machine learning algorithms have become quite effective in diagnostic applications. To this aim, we propose a combination of gray-level co-occurrence matrix (GLCM) and artificial neural network (ANN) to distinguish acute tympanic membrane otoscope images from normal images. For the experiment, totally 223 middle ear otoscope images were collected from the volunteer patients admitted to Van Ozel Akdamar Hospital. In the experimental study, the texture features are obtained separately from R, G, and B channels and then consolidated. In addition to the texture features provided by GLCM, the average values of each channels of the otoscope images are taken into account so as to determine whether the otoscope image belongs to acute or normal class. Lastly, this feature set is applied as the input to ANN. By experimental studies, we achieved 76.14% accuracy and with this model, we achieved promising results in diagnosing normal and acute OM disease. Consequently, the texture features were found as useful to classify normal and acute OM disease.