Enhancing Chest X-ray Image Super-Resolution through Residual Group and Channel Attention

다국어 초록 (Multilingual Abstract)

Medical imaging plays a crucial role in medical diagnosis, with chest X-ray imaging being a widely employed method for screening and diagnosing pulmonary diseases due to its cost-effectiveness. However, low-resolution images generated by expensive equipment and suboptimal imaging techniques often lead to a loss of critical features and acceptable texture. The acquisition of high-quality medical images is paramount for accurate disease diagnosis. This study introduces an innovative approach for reconstructing super-resolution medical images using deep learning techniques, explicitly targeting chest X-ray images. The proposed method, the Pathologically Invariant Remaining Enhanced Channel Attention Block (RECAB), incorporates the Exponential Linear Unit (ELU) activation function. The primary objective is to accurately recover high-resolution (HR) chest X-ray images from their low-resolution (LR) counterparts, leveraging a channel attention mechanism and convolution layer with the ELU activation function―the evaluation of the proposed method involved two datasets, X-Ray 2017 and X-Ray 2014. A comprehensive comparison was conducted with several state-of-the-art techniques, including GAN-based super-resolution, deep learning-based super-resolution, and interpolation-based super-resolution. The quality of the preprocessed images was assessed using the Structural Similarity Index (SSIM) and the Multi-Scale Structural Similarity Index (MSIM). The results prove the higher performance of the proposed method, which outperformed the average of 8.6% and 11.6% in SSIM and MSIM values by 11% and 12.14% on the two datasets, respectively. This research signifies a significant advancement in enhancing the resolution and quality of chest X-ray images, holding substantial potential for improving diagnostic accuracy and aiding in medical decision-making.