Chapter 3 - Artificial intelligence in radiology and its application in liver disease

Abstract

There have been remarkable developments in artificial intelligence (AI) for the analysis of radiologic examinations. Two AI-based approaches have been actively investigated in the field of liver imaging: radiomics and deep learning. Radiomics is a method for radiologic phenotype assessment using a large number of handcrafted features derived from images. Machine learning methods are typically used for radiomics-based decision-making. Deep learning is based on representation learning, in which the algorithm learns the best way to solve the problem on its own using labeled training image data. The convolutional neural network is the most popular type of deep learning architecture in imaging analysis. Radiomics have potential for staging liver fibrosis and predicting the prognosis and treatment response of hepatic malignancy. Studies have demonstrated the feasibility of deep learning for various tasks for liver imaging, such as segmentation of the liver and liver tumors, staging of hepatic fibrosis, detection and classification of focal hepatic lesions, and improving the image quality of computed tomography and magnetic resonance imaging However, most previous studies on radiomics and deep learning were considered preliminary and mainly focused on technical feasibility. Further clinical validation is required for the application of radiomics and deep learning in real-world practice. Here, we introduce the investigations and basic technical aspects of radiomics and deep learning in liver imaging and discuss current limitations and future directions for the successful clinical application of these techniques.