X-ray imaging is fundamental to medical diagnosis, but its effectiveness is often hampered by noise, poor contrast, and overlapping anatomical structures that can hide crucial signs of disease. Deep learning techniques have improved X-ray enhancement, yet many current approaches still struggle to retain fine details without introducing unwanted artifacts, which can limit their practical use in clinics. This work introduces Fractal-HyperNet, a novel deep learning framework designed to significantly advance X-ray image enhancement. Our approach uniquely combines two core innovations Dynamic Hyper-Dimensional Embedding (DHDE) module, which intelligently maps image features into an exceptionally high-dimensional space using an adaptive attention mechanism. This allows for more effective separation of subtle image signals from noise and interference. A Recursive Fractal-Scale Consistency (RFSC) architecture, complemented by a sophisticated multi-component loss function. This design enforces structural self-similarity and detail preservation across various image resolutions, reflecting the natural fractal characteristics of anatomical features. Conducted extensive evaluations on widely recognized public X-ray datasets, including ChestX-ray14, MIMIC-CXR, and a specially prepared simulated low-dose X-ray dataset. Compared to leading contemporary deep learning models, including advanced Transformer and diffusion-based architectures, Fractal-HyperNet demonstrated marked improvements. Specifically, our method achieved an average increase of 3.2 dB in Peak Signal-to-Noise Ratio (PSNR), an improvement of 0.04 in Structural Similarity Index (SSIM), a reduction of 0.025 in Learned Perceptual Image Patch Similarity (LPIPS), and a decrease of 7.5 points in Fréchet Inception Distance (FID), indicating superior image fidelity and perceptual quality. Furthermore, analysis of intensity histograms confirmed enhanced contrast restoration and a significant reduction in image artifacts. Crucially, in blinded evaluations, board-certified radiologists showed a strong preference for images enhanced by Fractal-HyperNet, with mean diagnostic quality scores improving by 0.9 points on a 5-point Likert scale compared to the next best method. These results underscore Fractal-HyperNet's capacity to retrieve more reliable diagnostic information from challenging X-ray images, offering a substantial step forward in the field.
Maaroof et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: