CT Scan Image Denoising and Exposure Optimization Using Cascaded U-Net with Sparse Constraints

Abstract

Accurate diagnoses in medical imaging heavily rely on high-quality 2D image slices from CT scans This study proposes a novel, data-driven pipeline to optimize these slices for improved diagnostic accuracy.  The pipeline integrates patient information with CT data acquisition, enabling personalized scan settings for each patient. An ARIMA time series model optimizes exposure time, balancing the need for high-quality images with minimizing radiation dose.Following data acquisition, the pipeline employs a cascaded network for pre-processing. This network meticulously removes noise and artifacts that can obscure anatomical details. Subsequently, a super-resolution model leveraging SRGAN and DENSE-Net enhances image resolution and sharpens intricate structures within the scanned area.The proposed methodology is rigorously evaluated on a dataset encompassing CT scans from 299 patients. This comprehensive analysis compares the quality of images generated by the pipeline against those produced by traditional methods. The study focuses on key metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) to assess improvements in image quality.This data-driven framework has the potential to significantly improve diagnostic accuracy in medical imaging. By providing clearer and more detailed images, healthcare professionals can make more informed decisions regarding treatment plans, ultimately leading to better patient outcomes.