Attention to Lesion: Lesion-Aware Convolutional Neural Network for Retinal Optical Coherence Tomography Image Classification

This paper presents an innovative approach to retinal Optical Coherence Tomography (OCT) image classification using Image Processing techniques combined with a Lesion-Aware Convolutional Neural Network (LACNN). The proposed method focuses on enhancing the classification accuracy by emphasizing lesion regions in retinal images, specifically targeting key retinal abnormalities such as drusen, choroidal neovascularization (CNV), diabetic macular edema (DME), and normal conditions. The LACNN architecture is designed to incorporate lesion-aware features, allowing the network to better focus on and classify regions of interest within the OCT images. Image Processing techniques, including preprocessing, feature extraction, and image enhancement, are employed to improve the quality and accuracy of the input data. The LACNN model is trained using a large dataset of retinal OCT images, and the classification results indicate high accuracy in distinguishing between the different conditions. By integrating lesion-awareness into the convolutional layers, the network achieves enhanced sensitivity to subtle abnormalities, which is crucial for early diagnosis and treatment planning in retinal diseases. The proposed method demonstrates significant potential in improving the diagnostic workflow for retinal diseases, providing a robust and reliable tool for clinicians to detect and monitor retinal conditions accurately.

Index Terms— Retinal Optical images Dataset, Deep Learning algorithm, Lesion-Aware Convolutional Neural Network.