Super-resolution reconstruction of medical images based on deep residual attention network

Medical images are commonly used to determine the location, size, and shape of organs, as well as the scope and physical properties of lesions, which are important bases for intelligent medical diagnosis. Low-quality medical images have serious spots, noise, and weak boundaries between similar tissues, which might affect the clarity of human organs and lesions in the image. This problem seriously hinders doctors’ diagnoses and the accuracy of computer-aided detection. Therefore, enhancing the internal texture details of medical images, strengthening tissue boundary information, and suppressing noise are of great significance for experts to diagnose diseases. We propose a medical image super-resolution reconstruction method based on residual attention networks. The method combines channel attention and spatial attention modules to enhance weak boundaries of the tissues and suppress noise. In addition, we introduce the skip connection structure to prevent network feature extraction from causing the loss of shallow feature information. We built three medical image datasets (lung CT images, brain MR images, and transrectal ultrasound (TRUS) images) to evaluate the performance of the proposed method. The results reveal that the proposed method outperforms other methods of medical image reconstruction. Moreover, it accurately reconstructs the internal texture and edge information of medical images while effectively suppressing noise.