Classification and Prediction of Lung Diseases According to Chest Radiography

Chest X-ray is a widely used and cost-effective medical imaging procedure, but its effectiveness is limited due to the shortage of qualified radiologists. Interest in deep learning methods in processing such medical modalities has increased with the advent of labeled x-ray image datasets and their growing availability. This paper proposes an approach for classifying and predicting pulmonary pathologies in frontal-thoracic radiography. Several architectures and different model depths have been considered for feature extraction from an image. To identify diseases on chest images, our approach utilizes image metadata (age, gender of the patient, X-Ray image projection). The effectiveness of the presented approach for various architectures was compared with other pathology classification methods using ROC statistics on a public test sample of the dataset containing more than 112 thousand labeled images. The model architecture adapted to work with patient data and based on the ResNet-34 convolutional network and gives the best average AUC results.