A diagnosis system by U-net and deep neural network enabled with optimal feature selection for liver tumor detection using CT images

One of the crucial problems in medical field is liver cancer, which creates a huge impact on the mortality rate. Though, existing histopathological diagnostic approaches pose more trouble on the medical detection model due to the complexities in the workload. Identifying the tumor and liver regions from the clinical CT images through complete automatic diagnosis process is crucial for liver disease detection. Hence, by combining deep learning with CT images, this paper implements a new model to increase the efficiency of the liver tumor diagnosis. The benchmark and manually collected datasets are initially taken for pre-processing performed by histogram equalization and median filtering. Further, the segmentation of the liver is done by adaptive thresholding with level set segmentation. Once the liver segmentation is done, the enhanced deep learning method termed U-Net is adopted for segmenting the tumor using a new Grey Wolf-Class Topper Optimization (GW-CTO) algorithm. Further, a set of features are extracted and the length of features leads to complexity in network training, so the optimal feature selection is adopted based on a multi-objective function by the GW-CTO algorithm. These optimally selected features are subjected to the Hyper-parameter tuned Improved-Deep Neural Network (HI-DNN) enhanced by the same GW-CTO algorithm. From the performance analysis, the accuracy of the developed GW-CTO-HI-DNN is 4.3%, 2.4%, 5.2% and 4.3% progressed than PSO-HI-DNN, O-SHO-HI-DNN, CTO-HI-DNN and GWO-HI-DNN, respectively while considering the learning percentage as 85%. The experimental analysis confirms the efficiency of the developed model to get high classification accuracy over the other methods.