Evaluation of dose reduction potential in scatter-corrected bedside chest radiography using U-net.

Bedside radiography has increasingly attracted attention because it allows for immediate image diagnosis after X-ray imaging. Currently, wireless flat-panel detectors (FPDs) are used for digital radiography. However, adjustment of the X-ray tube and FPD alignment are extremely difficult tasks. Furthermore, to prevent a poor image quality caused by scattered X-rays, scatter removal grids are commonly used. In this study, we proposed a scatter-correction processing method to reduce the radiation dose when compared with that required by the X-ray grid for the segmentation of a mass region using deep learning during bedside chest radiography. A chest phantom and an acrylic cylinder simulating the mass were utilized to verify the image quality of the scatter-corrected chest X-rays with a low radiation dose. In addition, we used the peak signal-to-noise ratio and structural similarity to quantitatively assess the quality of the low radiation dose images compared with normal grid images. Furthermore, U-net was used to segment the mass region during the scatter-corrected chest X-ray with a low radiation dose. Our results showed that when scatter correction is used, an image with a quality equivalent to that obtained by grid radiography is produced, even when the imaging dose is reduced by approximately 20%. In addition, image contrast was improved using scatter radiation correction as opposed to using scatter removal grids. Our results can be utilized to further develop bedside chest radiography systems with reduced radiation doses.