Eye-redness grading system based on deep learning and smartphone images: Application and innovation of improved ConvNeXt model

Red eyes, usually referring to the dilation of conjunctival blood vessels, are a common eye symptom, and long-term or severe red eyes may indicate potential eye diseases. This study proposes an innovative red eyes grading system based on smartphone images and deep learning technology, aiming to provide a low-cost, convenient and accurate solution for eye health monitoring. This study used 4570 smartphone images, of which 1000 were used for palpebral fissure detection model training, and all 4570 images were used for red eyes grading. Based on these data, we built a system with two deep learning models: the first model is based on the yolov8 model for palpebral fissure site detection and key area cropping, which has a more scientific training process and more powerful feature extraction capabilities; the second model is used to grade red eyes in the cropped images. In order to further improve the performance of the red eyes grading model, we innovatively introduced two attention mechanisms, CBAM (Channel-Spatial Attention Module) and SimAM (Simple Attention Module), based on the traditional ConvNeXt architecture. This combination effectively enhances the model's attention to key visual features and improves the accuracy of red eyes grading. The palpebral fissure detection model achieved IoU, Recall and F1 scores of 0.947, 0.9798 and 0.9847, respectively. The improved red eye grading model achieved accuracy, specificity and F1 scores of 0.8971, 0.9675 and 0.8974, respectively. Compared with traditional slit lamp devices, this system not only achieves high efficiency and accuracy in red eye detection and grading, but also has low computational cost and wide application potential. Smartphones combined with deep learning solutions can provide convenient and low-cost eye health monitoring, which is particularly suitable for areas with limited resources. In the future, this system is expected to become a popular tool to provide strong support for early eye disease screening, alleviate the pressure on medical resources, and improve the level of global eye health management.