OCT Retinal Scan Classification Using Machine Learning

This study explores the use of machine learning to classify retinal OCT scans into four categories (Normal, Drusen, CNV, and DME) to support more accurate and consistent eye disease diagnosis. By comparing Logistic Regression, MLP, and CNN models, the study found that a CNN using the VGG16 architecture achieved the highest accuracy at 98%, demonstrating the potential of deep learning and transfer learning in retinal disease detection.

Optical Coherence Tomography (OCT) scans are non-invasive images that use light waves to take detailed pictures of the retina. These images can detect a variety of optical diseases that can affect the vision and function of the eye. Our project aims to compare different machine learning models in their ability to classify retinal images. We want to provide support to ophthalmologists by finding an algorithm that can produce consistent accurate diagnosis. In this project, we classified the OCT scans into four classes of diseases: Normal, Drusen, Chloroidal Neovascularization (CNV), and Diabetic-Related Macular Edema (DME). We used a Logistic Regression, MLP Classifier, convolutional neural network (CNN) with a VGG16 architecture, and compared each on their testing accuracy score. We found the Logistic Regression gave us an accuracy of 52%, the MLP Classifier gave us 73%, and the highest, CNN, with 98% accuracy. These results show that convolutional neural networks can efficiently classify OCT retinal scans accurately. The successful results from using a pre-trained architecture like VGG16 suggests that transfer learning is sufficient in separating the four classes in this dataset. However, before implementing in clinical settings, the model would require validation using scans from different patient populations and imaging devices.

Explore More!

Research Paper