Simple net: Convolutional neural network to perform differential diagnosis of ampullary tumors
Jae Duk Seo, Dong Wan Seo, Javad Alirezaie · 2018
Diagnosing different stages of cancer has only been performed by doctors due to the complexity of the task. However recent advancements made in the field of deep learning has pushed the capabilities of what an algorithm can achieve. In this study, we have trained a convolutional neural network to perform differential diagnosis of Ampullary tumors. Our proposed network is only made out of seven layers. However, when compared with other state of the art classification networks such as VGG 16, VGG 19, Res Net, and Dense Net our model not only had the best performance but also shortest training time. All of the networks were trained for 150 epochs with step wise learning rate with Adam optimizer to converge as quick as possible. Our model was able to reach average of 78.14 percent accuracy with average training time of 50.60 seconds on Asus Zephyrus, with Nvidia 1080 GPU and Max Q technology.