OCT Image Generation and Disease Classification using Deep Neural Networks
Eyesight is one of the most important senses in human body. It is essential to keep our eyes safe but defects in eye or in retinal area is a common cause and wide-spread due to technological improvements and other causes. Detection of these retinal pathologies in early stages will help in further diagnosis and helps in treating the patients. Deep learning helps in detection of these retinal diseases. But there is a huge lack of data in recent methods to proceed for detection and classification which might affect the overall accuracy of the model. So, we propose a Deep Convolutional Generative Adversarial Network (DCGAN) for generating the training data for the classification.OCT(Optical Coherence Tomography) is antechnique in Image processing to get cross sections of retinal areas of living patients in high resolution. OCT Imagesare generated through DCGAN and kept for further process. The existing methods include the classification of single type of diseases and doesn’t have enough training data. Our proposed method makes the model to instinctively concentrate on the required parts of the given images. The generated OCT Imaging data is further sent into our Convolutional Models (DENSE NET and RES NET) to classify into as normal and different retinal pathologies like DRUSEN, DME and CNV. The proposed method was evaluated and achieves an accuracy of 0.92(DENSENET) and 0.86(RESNET). This experimenting model of generating OCT datasets and detecting the pathologies outperforms the existing state-of-art models.
Keywords: Retina, OCT, DCGAN, DRUSEN, DME, CNV.