Training and Diagnosis of Retinal Oct Images With Auxiliary Data Using Triplegan When Imbalanced Class Occurs
by Justin Joshua Park
Abstract – As a result of the recent COVID-19 outbreak, global interest in health-care issues are developing. Due to the pandemic, people are meeting via the internet and online more and more, and because of that, we are entering the era of hyper-connection between industries and IT at a fast rate. When a doctor treats a patient, there are two possible issues. Those concern the possibility of misdiagnosis and physical limitations. However, because of the virtual barrier nowadays, we can utilize deep learning and remote medical treatment that can be performed. This is more accurate and decision making can also be easier. However, Deep learning requires a lot of data to learn and train, and medical data is personal information, meaning that it is difficult to obtain because of security factors. We created additional data using our proposed model, which is Triple GAN to solve the problem of healthcare deep learning that cannot be learned. The data Imbalance problem contributes to the decrease in accuracy, and we solved this problem to produce a noticeable improvement in our accuracy. Afterwards, this method can be applied to various industries where it is difficult to obtain data.