Automatic Computer Propped Diagnosis Framework of Liver Cancer Detection using CNN LSTM

Abstract

Liver cancer detection using the computer vision methods and machine learning already received significant attention of researchers for authentic diagnosis and on-time medical attentions. The Computer Aided Diagnosis (CAD) preferred for cancer detection all over the world which is based on image processing service. Earlier CAD tools were designed using conventional machine learning techniue using semi-automatic approach. The modern growth of deep learning for automatic detection and classification leads to significant improvement in accuracy. This paper the automatic CAD framework for liver cancer detection using Convolutional Neural Network (CNN) including Long Short Term Memory (LSTM). The input Computed Tomography (CT) scan images early pre-processed for quality enhancement. After that we applied the lightweight and accuracy field of Interest (ROI) extraction technique using dynamic binary segmentation. From ROI images, we extracted automated CNN-based appearance and hand-craft features. The consolidation of both features formed unique feature set for classification purpose. The LSTM block is then achieve the classification either into normal or diseased CT image. The CNN-LSTM model is designed in this paper to complement the accuracy of liver cancer detection compared to other deep learning solutions. The experimental results of proposed design using CNN-based features and hybrid hand craft features outperformed the recent state-of-art methods.