A Fundus-Based Color Mapping Method to Optimize RNFL Clock Hours for Glaucoma Screening

Glaucoma, a leading cause of irreversible blindness, necessitates early detection and monitoring. This study, utilizing the Chula-Thammasat Glaucoma dataset (GlauCUTU-DATA), builds upon our previous EMBC 2024 work by introducing a novel color mapping technique and substituting previously-used loss function to form a retinal nerve fiber layer (RNFL) clock hour representation. A new color mapping technique was developed to visualize RNFL thickness by assigning specific colors to RNFL thickness ranges, with warmer colors indicating thicker regions and cooler colors representing thinner areas. This method resulted in enhanced visualization quality in thicker regions; however, the depiction of thinner regions-which are crucial to glaucomatous screening-remained unclear. Thus, citing the method's effectiveness in thicker regions, detail-gained color map was employed. Furthermore, substituting Root Mean Square Error (RMSE) with Peak Signal-to-Noise Ratio (PSNR) as the loss function provided a more detailed evaluation of image quality. Results revealed performance variations across color classifications, with an overall micro F1-score of 70.51%. Notably, clock positions 2, 5, 7, and 12, which falls in or close to the inferior and superior region critical for glaucoma screening, exhibited micro-average F1-scores exceeding 80%. This study demonstrates the efficacy of our novel color mapping methodology and PSNR loss function in generating RNFL clock hours, offering a potential advancement in glaucoma diagnosis and management.