Bio-Markers Presence Detection Using Transfer and Ensemble Learning on Optical Coherence Tomography of Retinal Imagery

This research explores the potential of transfer and ensemble learning techniques applied to retinal OCT scans to classify six biomarkers. This work evaluates a variety of model combinations and preprocessing techniques. It prioritizes simplicity and efficiency over complicated architectures. Various image processing techniques, including FFT-based noise reduction, median filtering, binarization, and contrast and brightness adjustments, has been used at the same time their performance was evaluated. Our work identifies the most effective denoising technique for OCT scans, which helps to optimize the learning process. Excellent outcomes are achieved by combining DenseNet169 and GoogleNet and making specific brightness and contrast adjustments. A training accuracy of 93.628% and an F1 score of 0.87 have been achieved. It determines the most effective denoising technique for OCT scans, contributing to the optimization of the learning process. This study underscores the significance of model diversity and preprocessing strategies in optimizing machine learning outcomes, offering valuable insights for future research and practical applications.