Optimized spatial automatic color enhancement technique: A novel approach for color restoration in retinopathy of prematurity (ROP) retinal images

Infant retinal images are crucial for clinicians in diagnosing pediatric retinal diseases such as Retinopathy of Prematurity (ROP). These images are highly inclined towards distortion due to various factors such as errors in imaging instruments, transmission channels, variable atmospheric and environmental conditions leading to degradation of image quality. Such distortions can manifest in different ways such as noise, backscattering, low saturation, poor contrast, low illumination, and blurring, compromising the effectiveness of retinal images, potentially limits the accurate ROP diagnosis and treatment. To address these challenges, we present an Optimized Spatial Automatic Color Enhancement (OS-ACE) technique which employs a locally adaptive enhancement algorithm that operates on individual color channels of the RGB retina image. By utilizing the convolutional based enhancement coupled with a power-law transformation, the technique selectively amplifies local contrast while preserving overall image balance. This local enhancement is further complemented by a normalization procedure to ensure the retention of true color perception and mitigate the introduction of artifacts. The study integrates the OS-ACE algorithm into the established framework of conformal mapping to develop a comprehensive processing pipeline which significantly enhances the quality of retinal images ensuring optimal visualization and better clinical decision for retinal disease treatment. The performance of proposed framework was evaluated quantitatively and qualitatively on 1205 ROP retinal image datasets. The method achieved accuracy 0.9846, F1 score 0.8362, Jaccard score 0.7186, recall 0.8091, and precision 0.8652 respectively. The image quality assessment (IQA) models achieved results of PSNR 28.8371, SSIM 0.8705, FSIM 0.7024, BRISQUE 35.8471 respectively. Compared to existing methods, such as Alimanov, A. et al.’s super resolution technique using deep learning, Shen Z., et al.’s COFE-Net, Bataineh, B. et al.’s multi-stage enhancement method, the proposed OS-ACE technique achieves better results in PSNR, SSIM, F1 score, recall, precision and accuracy highlighting the robustness and effectiveness in clinical applications.