Automated analysis of x-ray images using transparency masks

Abstract

The method for classification of x-ray images, which is implemented in this article, improves the differential diagnosis of socially considerable diseases. The novelty of this method is that the input digital x-ray image is supplemented by a transparency mask which is found by image segmentation, and the classified sign vector is formed by pixels which aren’t masked by the transparency mask. The original x-ray image is divided into segments to classify morphological structures with pathological formings. The classification of the selected segment is implemented by a modified convolutional neural network which uses pooling layers and layers of a fully-connected neural network. The proposed classifier allows dividing patients into two groups: patients without found pathology and patients with disorders of health condition. The found morphological pathological formings help clinicians to find the most promising way for prevention of the disease development at the patients, saving money and time.