Classification of Lesion in 3D CT Liver Image Using 3D-CNN Based on GLRLM Feature Extraction

Abstract

CT imaging is considered as most reliable techniques for non-invasive diagnosis. In this paper we used 3D CT images for the classification of cancerous tumor in the liver. The dimension reduction is applied in 3D CT image using PCA(Principal Component Analysis) . Noise is reduced by 3D median filter and Data Augmentation techniques are implemented, Random Forest Algorithm gives the segmentation of the abnormal part of the liver. GLRLM features are extracted using Pyradiomics of Python and fed to 3D CNN classifier for Haemangioma and Cancer Tumor. The result is compared with other models and found that our model give better performance. The performance metrics is evaluated for our model that gives an accuracy of 98.6% for train dataset and 96.66% for test dataset.