Multiparameterized Inception-V3 Convolution Neutral Network for Liver Lesion Classification and lesion staging using CT images

Liver Cancer is highest cause of cancer death all over the world due to its high morbidity and mortality. Risk factor associated with liver cancer includes, cirrhosis fibrosis and hepatitis B and C. Liver cancer is highly challenging to diagnosis in an early stage due to less symptoms tumor located deeply into body. Machine learning architecture is high implemented for diagnosing disease in earlier stage and monitoring the progress of the disease. However those architecture are time consuming and error prone due to complex features on Couinaud segment annotation. Therefore, new methodology for early identification of liver cancer is required to compute the liver condition and cure disease before the liver deterioration. In order to handle those challenges, a new deep learning mechanism entitled as multiparameterized Inception V3 Convolution Neural Network is proposed for liver lesion classification and lesion staging. Initially image preprocessing using median filter has been implemented to improve the results of the image segmentation through noise removal. Next, global thresholding based segmentation has been employed to pre-processed images to segment the region of the interest and lesion boundaries effectively. Those segmented image has been employed to fast kernel discriminant analysis which acts feature extraction technique to extract the multiple lesion feature of the liver lesion region. Extracted feature has been employed to the model of the generate the learning model on the employing multiparameterized Inception V3 Convolution Neural Network Classifier for liver lesion classification and staging of disease on basis of feature values. multiparameterized Inception V3 Convolution Neural Network Classifier contains the convolution layer for feature map generation, max pooling layer to obatin the high level features and fully connected layer uses softmax layer to discriminate the feature map into liver lesion classes such as hepatocellular carcinoma, hemangioma and liver metastasis. Further proposed model reduces the network complexity and enhances the computing efficiency using loss layer. Experimental results of the current model is verified in the MATLAB software on using LiTs dataset which contains 1500 CT images. Performance analysis of the proposed architecture generates the disease classes such as cirrhosis, fibrosis, basal hepatocellular carcinoma, hemangioma and liver metastasis with 97.75% accuracy, 96.46 specificity and 98.4% sensitivity respectively on comparing against conventional classifiers