Deep Learning Based Automated Detection of Arcus Senilis and Its Clinical Risks in Ocular Health

Abstract

Arcus Senilis is a clinical indicator of lipid deposition in the cornea, commonly observed in aging individuals. This study aims to develop an automated deep learning-based pipeline for detecting Arcus Senilis and estimating cholesterol levels from ocular images. We implemented an image-based classification system using EfficientNetB0, a state-of-the-art convolutional neural network (CNN). The dataset was pre-processed using Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance contrast. The model was trained using transfer learning, incorporating global average pooling and fully connected layers to classify Arcus Senilis presence and estimate cholesterol levels. Additionally, patient metadata, including age and lipid levels, was integrated to enhance prediction accuracy. The model was trained on a labelled dataset, with a multi-task learning approach handling both classification (Arcus Senilis detection) and regression (cholesterol level estimation). Performance was evaluated using Mean Absolute Error (MAE), R² Score, Accuracy, and Confusion Matrices. The proposed model achieved an accuracy of 92.5% for Arcus Senilis detection and a Mean Absolute Error (MAE) of 8.4 mg/dL for cholesterol level estimation. The system effectively distinguished Arcus Senilis from normal eyes and provided clinically relevant cholesterol estimations. Evaluation metrics, including precision, recall, and F1-score, demonstrated its reliability compared to traditional machine learning approaches such as SVM + HOG Features, ResNet50, and VGG16. The proposed deep learning pipeline provides a non-invasive, accurate, and automated solution for Arcus Senilis detection and cholesterol level estimation. The findings suggest potential applications in ophthalmic diagnostics and lipid metabolism assessment.