Abstract
To propose a low-cost and effective image enhancement strategy based on multi-frame fusion for corneal confocal microscopy (CCM) that improves image quality without requiring additional hardware or changes to clinical workflows. The method involves aligning and integrating consecutive frames of the same region. Its performance was systematically evaluated across image alignment accuracy, noise reduction, morphological nerve feature extraction, and disease classification. Quantitative experiments showed that the proposed approach significantly enhances structural clarity and measurement reliability. Key parameters such as corneal nerve fiber length (CNFL), corneal nerve fiber density (CNFD), and corneal nerve branch density (CNBD) showed substantial improvements, especially in diabetic patients. Enhanced images consistently improved both traditional metrics-based discrimination and deep learning classification models across multiple architectures, demonstrating the method's generalizability and clinical value. The proposed multi-frame fusion strategy effectively enhances CCM images with minimal additional acquisition time and without burdening patients or operators, making it highly suitable for real-world clinical applications.