Artificial intelligence: Deep learning in oncological radiomics and challenges of interpretability and data harmonization

Abstract

Over the last decade there has been an extensive evolution in the Artificial Intelligence (AI) field. Modern radiation oncology is based on the exploitation of advanced computational methods aiming to personalization and high diagnostic and therapeutic precision. The quantity of the available imaging data and the increased developments of Machine Learning (ML), particularly Deep Learning (DL), triggered the research on uncovering "hidden" biomarkers and quantitative features from anatomical and functional medical images. Deep Neural Networks (DNN) have achieved outstanding performance and broad implementation in image processing tasks. Lately, DNNs have been considered for radiomics and their potentials for explainable AI (XAI) may help classification and prediction in clinical practice. However, most of them are using limited datasets and lack generalized applicability. In this study we review the basics of radiomics feature extraction, DNNs in image analysis, and major interpretability methods that help enable explainable AI. Furthermore, we discuss the crucial requirement of multicenter recruitment of large datasets, increasing the biomarkers variability, so as to establish the potential clinical value of radiomics and the development of robust explainable AI models.

Prostate Cancer Radiogenomics—From Imaging to Molecular Characterization

“…Radiomic flow is a complex process, and every aspect of the image acquisition, such as defining and contouring the regions of interests, and choosing the best features to be extracted and the proper statistics to be applied, remains challenging. Lately, explainable AI (XAI), using DNNs (deep neural networks), may help radiomics in classification and prediction in the clinical setting [ 174 , 175 ], and a controllable and explainable probabilistic radiomics framework was proposed, through which a 3D CNN feature is extracted upon the lesion region only and is used to approximate the ambiguity distribution over human experts [ 176 ]. These new features will have to be further validated.…”

Strategies to develop radiomics and machine learning models for lung cancer stage and histology prediction using small data samples

“…Despite the encouraging results that we obtained, the ML-based decision system we have developed is not yet ready to be applied in clinical workflows. Before application-specific ML algorithms can be successfully translated into clinics, a number of challenges must indeed be overcome, including the harmonization of different data samples [7,59], the reliability and reproducibility of the results [7,10,60] the interpretability of the models and the results [59,61], and the compliance with current regulations [62].…”

Machine Learning and Radiomics Applications in Esophageal Cancers Using Non-Invasive Imaging Methods—A Critical Review of Literature

“…One key challenge of applying deep networks in clinical decision making is that deep networks are black box models with multilayer nonlinear operations, thus the reasoning behind the results from deep networks are very difficult to interpret clinically. Explainable AI is an emerging field of active research in trying to address this challenge [ 90 , 91 ].…”