Integrating DeblurGAN and CNN to improve the accuracy of motion blur X-ray image classification

X-rays are common tools used in clinical diagnoses. During the X-ray process, a patient is required to lie still or to hold a deep breath. However, when dealing with patients who may shake involuntarily or with restless children, the required conditions are not always achievable, and blurred images often occur. If these patients receive repeated X-rays, they are exposed to additional amounts of radiation. This research integrates the DeblurGAN model and a convolutional neural network (CNN) to increase the accuracy for classifying clinical X-ray motion blurred images, eliminating the need for repeated clinical X-rays. Results show that classification accuracy of the motion blur group was 87% while the deblurred group was 91%, indicating the process not only improves classification accuracy but also restores the deblurred group images to a 92% level of the original image group. This study verifies the feasibility of applying the DeblurGAN model to medical X-ray image deblurring, helping resolve the challenges presented by a lack of X-ray motion blur image data and allowing InceptionV3 to accurately identify the problem. This method can be further applied to other motion blur medical images (such as MRI and CT) to improve overall clinical diagnosis efficiency. (author)