Abstract
The advent of radiomics has revolutionized medical image analysis, affording the extraction of high dimensional quantitative data for the detailed examination of normal and abnormal tissues. Artificial intelligence (AI) can be used for the enhancement of a series of steps in the radiomics pipeline, from image acquisition and preprocessing, to segmentation, feature extraction, feature selection, and model development. The aim of this review is to present the most used AI methods for radiomics analysis, explaining the advantages and limitations of the methods. Some of the most prominent AI architectures mentioned in this review include Boruta, random forests, gradient boosting, generative adversarial networks, convolutional neural networks, and transformers. Employing these models in the process of radiomics analysis can significantly enhance the quality and effectiveness of the analysis, while addressing several limitations that can reduce the quality of predictions. Addressing these limitations can enable high quality clinical decisions and wider clinical adoption. Importantly, this review will aim to highlight how AI can assist radiomics in overcoming major bottlenecks in clinical implementation, ultimately improving the translation potential of the method.