HCSP-Net: A Novel Model of Age-Related Macular Degeneration Classification Based on Color Fundus Photography

• DOI: 10.32604/cmc.2024.048307
• Publication Date: 2024-03-28T06:52:47Z
• Authors: Cheng Wan; Jiani Zhao; Xiangqian Hong; Weihua Yang; Shaochong Zhang
• Keywords: Fundus Photography; Fundus (uterus)

Age-related macular degeneration (AMD) ranks third among the most common causes of blindness.As conventional and direct method for identifying AMD, color fundus photography has become prominent owing to its consistency, ease use, good quality in extensive clinical practice.In this study, a convolutional neural network (CSPDarknet53) was combined with transformer construct new hybrid model, HCSP-Net.This model employed tri-classify into normal macula (NM), dry (DMD), wet (WMD) based on classification manifestations, thus resolving AMD as early possible photography.To further enhance performance grouped convolution introduced study without significantly increasing number parameters.HCSP-Net validated using an independent test set.The average precision HCSP-Net diagnosis 99.2%, recall rate 98.2%, F1-Score 98.7%, PPV (positive predictive value) NPV (negative 99.6%.Moreover, knowledge distillation approach also adopted develop lightweight student (SCSP-Net).The experimental results revealed noteworthy enhancement accuracy SCSP-Net, rising from 94% 97%, while remarkably reducing parameter count quarter attribute positions SCSP-Net highly suitable candidate deployment resource-constrained devices, which may provide ophthalmologists efficient tool diagnosing AMD.