X-ray image analysis using machine learning methods

Many diseases can be detected in the chest area using X-ray images. However, a correct diagnosis requires a lot of experience from the specialist who makes it. Nevertheless, even this does not guarantee a correct diagnosis, because the images may be, for example, of poor quality, which will make it di cult to make a correct diagnosis. In such cases and in general, Arti cial Intelligence (AI), namely a model for Computer Vision (CV), can help to optimize and automate the work. To train or ne-tune well-known models (such as ResNet and YOLO) for CV to distinguish an X-ray image of a healthy person from a sick one, as well as to train models how to distinguish certain diseases from each other. This study analyzed chest X-Ray images on NIH Chest X-ray Dataset that was comprised of 112,120 X-ray images with disease labels from 30,805 unique patients. Analysis included di erent subsets NIH Chest X-ray Dataset based on model and method of training, identifying Atelectasis, Consolidation, In ltration, Pneumothorax, Edema, Emphysema, Fibrosis, E usion, Pneumonia, Pleural_thickening, Cardiomegaly, Nodule Mass, Hernia and normal ndings. The ResNet152 model with frozen layers showed an F1-score of 0.8 on test data to distinguish the healthy patients from the sick in X-rays. ResNet50 model trained for multi-label classi cation showed F1-score in peak for Cardiomegaly up to 0.71. Trained models using AI can e ectively decrease workload on. However, accuracy and precision of such models still can be increased in future by improving di erent aspects of training.