Improving MVCT image quality for quantitative analysis of inter-fractional organ motion in prostate cancer radiotherapy

Image-guided radiation therapy is crucial for mitigating daily uncertainties in patient setup during radiation dose delivery to the target volume and adjacent organs at risk. The analysis of mega-voltage computed tomography (MVCT) images acquired for intensity-modulated radiation therapy in helical tomotherapy enables the evaluation of inter-fractional motion, which corresponds to organ movement and changes between radiation treatment sessions. However, poor MVCT image quality can hinder organ delineation and lead to inter-observer variability, resulting in inaccurate tracking of changes over time. This study proposes a deep learning-based approach to enhance the quality of MVCT images for quantitative analysis of inter-fractional organ motion in prostate cancer patients. We collected datasets comprising planning kilo-voltage computed tomography and MVCT images from prostate cancer patients who underwent helical tomotherapy. Three deep-learning techniques (U-Net, pix2pix, cycleGAN) were employed to reduce noise and enhance contrast, thereby improving the quality of the MVCT images. We conducted a quantitative assessment of the effectiveness of specific improvements using different metrics such as structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), mean absolute error (MAE), and root mean square error (RMSE). From the results, the cycleGAN method showed better overall performance in terms of PSNR, MAE, RMSE, and SSIM across various cases. Specifically, in the case of images containing the prostate, the cycleGAN-enhanced images demonstrated a 2.46% increase in PSNR, a 24.52% reduction in MAE, and an 11.54% reduction in RMSE compared to the original MVCT images. The improved MVCT images using the cycleGAN method delineated critical organ contours relevant to prostate cancer treatment, including the bladder, prostate, and rectal balloon. Consequently, this study confirmed the feasibility of conducting quantitative analysis on the changes in prostate position in relation to rectal balloon placement and bladder volume variations during prostate cancer treatment using the enhanced MVCT images.