Convolution neural network model for predicting various lesion-based diseases in diabetic macula edema in optical coherence tomography images

Diabetic Macular Edema (DME), a rare eye disease primarily found in diabetic patients, is due to the formation of fluid in the extra-cellular space of the macular area in the retina. In the earlier days, it was detected through fundus images that provided less accuracy and were difficult for early detection. Optical Coherence Tomography (OCT) is widely adopted to overcome such issues. It is an advanced imaging modality that provides a better view of the retinal structure. However, the medical professionals manually carried out the detection of DME from the OCT images. Advancement in machine learning algorithms has enabled easy processing of OCT images for DME detection. However, the machine learning algorithm provided less accuracy as it was limited to 2-dimensional datasets and their parameters. The lesions in the images were detected using the 3-dimensional structure of the OCT images. The deep learning algorithms consisted of various layers that increased the algorithm's efficiency and provided the necessary features and parameters that helped in the earlier detection of DME. It is also important to detect the types of DME like hemorrhages, Microaneurysms, and exudate. In this paper, a novel lesion-based CNN algorithm is proposed for the efficient detection of the lesions that help in better prediction. The proposed model is compared with other deep learning models, and the results show that the LCNN model provides improved accuracy than normal deep learning models like ResNet, VGG16, and Inception. The accuracy of deep learning models like AlexNet and Inception is increased to 96%.