Classification of Tympanic Membrane Images based on VGG16 Model

Abstract

Otitis Media (OM) is a type of infectious disease caused by viruses and/or bacteria in the middle ear cavity. In the current study, it was aimed to detect the eardrum region in middle ear images and thus to diagnose OM disease by using artificial intelligence methods. The Convolution Neural Networks (CNN) model and the deep features of this model and the images obtained with the otoscope device were used. In order to separate these images as Normal and Anomalous, the end-to-end VGG16 model was directly used in the first stage of the experimental work. In the second stage of the experimental study, the activation maps of the fc6 and fc7 layers consisting of 4096 features and the Fc8 layer consisting of 1000 features of the VGG16 CNN model were obtained. Then, it was given as input to Support Vector Machines (SVM). Then, the deep features obtained from all activation maps were combined and a new feature set was obtained. In the last stage, this feature set is given as an input to SVM. Thus, the effect of the VGG16 model and the features obtained from the layers of this model on the success of distinguishing images of the eardrum was investigated. As a result of the experimental study, the best performance results were obtained with the fc6 layer. The results showed that OM disease could be accurately detected by using a deep CNN architecture. The proposed deep learning-based classification system promises highly accurate results for disease detection.