Using of neural networks to search for errors of patient's positioning on chest X-rays

The paper presents the results of the application of transfer learning of deep convolutional neural networks for the task of searching for chest X-rays with errors of patient styling and positioning. Evaluated neural network architectures: InceptionV3, Xception, ResNet152V2, InceptionResnetV2, DenseNet201, VGG16, VGG19, MobileNetV2, NASNetLarge. For training and testing we used chest X-rays from open datasets and the unified radiological information service of the city of Moscow. All the models obtained had diagnostic accuracy metrics above 95., while models based on the ResNet152V2, DenseNet201, VGG16, MobileNetV2 architectures had statistically significantly better metrics than other models. The best absolute values of metrics were shown by the ResNet152V2 model (AUC =0.999 , sensitivity=0.987, specificity=0.988, accuracy=0.988, F1 score = 0.988). The MobileNetV2 model showed the best processing speed of one study ($67.8 pm5.0$ ms). The widespread use of the algorithms we have obtained can facilitate the creation of large databases of high-quality medical images, as well as optimize quality control when performing chest X-ray examinations.

neural networks, deep learning, quality control, chest X-ray