Deep Learning Assisted Image Reconstruction in Low-Dose Medical Imaging

Abstract

Low-dose medical imaging has emerged as a critical research direction in modern healthcare due to increasing concerns regarding radiation exposure associated with diagnostic imaging procedures. Although dose reduction strategies significantly improve patient safety, they often introduce image noise, artifacts, and degradation of anatomical details, thereby affecting diagnostic reliability. Recent advances in deep learning have provided transformative solutions for reconstructing high-quality medical images from low-dose acquisitions. Deep neural networks, including convolutional neural networks, residual learning architectures, generative adversarial networks, transformers, and hybrid reconstruction frameworks, have demonstrated remarkable capabilities in noise suppression, artifact removal, structural preservation, and enhancement of diagnostic features. These approaches enable the generation of clinically acceptable images while maintaining reduced radiation exposure levels. Furthermore, integration of data-driven learning with physics-based reconstruction models has improved reconstruction accuracy, robustness, and computational efficiency. This paper presents a comprehensive review of deep learning-assisted image reconstruction techniques for low-dose medical imaging, highlighting algorithmic developments, reconstruction methodologies, performance evaluation metrics, and clinical implications. The study also identifies current challenges and future research directions toward trustworthy, interpretable, and scalable intelligent imaging systems for next-generation healthcare applications.