Liver Tumor Detection Using Deep Learning Techniques

Segmentation and categorization of Liver CT scan images is very crucial phase, many algorithms are used to detect the tumor from CT scan images.. We designed a model for liver tumor detection using U-net architecture. U-net is a convolutional neural network (CNN) which is generally used for image segmentation method, such as medical image analysis. The U-Net model is very much efficient for a multiple image segmentation assignments, and has been employed in a various types of applications, including medical image segmentation and biological image segmentation. It uses many convolutional layers to extract the features automatically. We have used 3DIRCADB dataset and we have got the accuracy around 98.61 % and dice coefficient is 75.97%. The accuracy is showing the efficiency of U-net architecture, but it requires big size dataset and high processing head. So we have used Thepades sorted block truncation coding (TSBTC) algorithm on the 3dircadb dataset and extracted the features of the dataset till 8ary. Then we have used different classifier families like function(Logistic), Bayes(BayesNet), Lazy(IB1, Kstar), Tree(RandomForest). We got highest accuracy of97.82% by using BayesNet classifier with TSBTC “Septanary” version. Still work is going on to improve our results using the ensemble based classifier with TSBTC.