A Robust Machine learning based method to classify normal and abnormal CT scan images of mastoid air cells

The abnormality of mastoid air cells represents various types of ear maladies. The current traditional manual analyzing of huge amount of collected images from ear cavity is time consuming and the low accuracy of diagnosing these abnormalities by humans is inevitable, thus the subsequent consequences could threaten the patient's health. This study presents an automated machine learning based method to classify normal and abnormal mastoid air cells using CT images procured in the clinical center. Methods This paper introduces the first robust method based on convolutional layers and deep neural network to classify normal and abnormal mastoid air cells. The used dataset is comprised of total of 24,800 (right and left mastoid) CT slides of 152 patients who have been referred to the Tabriz Golgasht Imaging Center(TGIC) at the request of the ENT specialist which include the mastoid air cells from most upper to the lowest part of the ear cavity. Results The proposed fully automatic classification and diagnosing method provides a promising result compared to the manual classification by ENT specialists. In our classification algorithm the accuracy, f 1_score, Precision, Recall, were 98.10%, 98.05%, 98.32%, 97.89% respectively(over the five-fold cross-validation on validation dataset) and the accuracy of this method on test data was 97.56% (the average of 5 times running of five-fold cross-validation). The robustness and efficiency of the proposed method are demonstrated by comparison with some of most common deep learning architectures ResNet50 and AlexNet. Conclusions The proposed machine learning method directly learned from B-scan labels, requiring no manual detailed annotations at image. Medically, the image investigation of ear CT scan images mainly remains at the doctor’s manual diagnosis stage, but manual examination and diagnosis could be labor intensive and time consuming. In this paper, a deep convolutional neural network (ConvNet) is used to achieve automatic classification of mastoid air cells using CT images by analyzing the characteristics of the patient’s CT images of the ear cavity.