Abstract
Healthcare is one of the domains which has truly advanced to use the latest technologies like machine learning to help in diagnosis. Owing to its complexity of medical images, extracting features correctly makes the problem even tougher. The earlier image processing algorithms using descriptors were unable to detect the disease accurately, and also, using the correct form of descriptors based on the dataset was even a bigger challenge which further reduced the accuracy of the machine learning algorithms trained over the dataset. However the recent advancements in the field of deep learning are able to give better results for these classifications. Convolutional neural networks (CNNs) have proved to be a great algorithm choice in case of extracting spatial features, making it suitable for the medical diagnosis. However, as the number of layers increase in CNNs, the complexity of the network increases and the information passed from one later to another eventually decreases, thus causing information loss. In order to overcome this dense CNN can be considered. We have shown couple of case studies which were performed. One using the local binary patterns and other using dense CNN on two different types of medical images. The dense CNN work was also recognized by the IEEE Computer Society.
