A review of image enhancement techniques for improved detection of spinal lesions in medical imaging

Accurate diagnosis of spinal lesions relies on clear and high-quality medical images obtained from X-ray, magnetic resonance imaging (MRI), and computed tomography (CT). In clinical practice, these images are often affected by noise, low contrast, imaging artifacts, and variations in acquisition settings. Such limitations can obscure spinal anatomical structures and reduce the accuracy of both clinical assessment and computer-aided image analysis. Consequently, image enhancement has become an essential preprocessing step for lesion detection, segmentation, and other automated image analysis tasks. This paper presents a comprehensive structured review of recent advances in spinal image enhancement by analyzing studies published between 2016 and 2025. The review is conducted by Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The reviewed studies encompass conventional image processing methods, machine learning (ML) techniques, deep learning (DL) approaches, including convolutional neural networks (CNNs), U-Net variants, generative adversarial networks (GANs), and Transformer models. For systematic comparison, the selected studies are categorized according to imaging modality, enhancement technique, and clinical application. The reviewed studies evaluate image enhancement using quantitative measures such as the Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR). However, relatively few studies investigate whether these improvements consistently translate into enhanced lesion detection, segmentation accuracy, or overall diagnostic performance. This finding emphasizes the importance of evaluating enhancement techniques using clinically relevant downstream tasks rather than relying solely on image quality metrics. Furthermore, this review summarizes the strengths and limitations of existing enhancement approaches and discusses current challenges, including limited dataset diversity, poor cross-dataset generalization, inconsistent evaluation protocols, insufficient model interpretability, reproducibility issues, and limited end-to-end clinical validation. Finally, future research work is presented to support the development of reliable, efficient, and clinically deployable AI-assisted spinal imaging systems.