AI applications to medical images: From machine learning to deep learning

Abstract

Artificial intelligence (AI) models are playing an increasing role in biomedical research and healthcare services. This review focuses on challenges points to be clarified about how to develop AI applications as clinical decision support systems in the real-world context. Methods: A narrative review has been performed including a critical assessment of articles published between 1989 and 2021 that guided challenging sections. Results: We first illustrate the architectural characteristics of machine learning (ML)/radiomics and deep learning (DL) approaches. For ML/radiomics, the phases of feature selection and of training, validation, and testing are described. DL models are presented as multi-layered artificial/convolutional neural networks, allowing us to directly process images. The data curation section includes technical steps such as image labelling, image annotation (with segmentation as a crucial step in radiomics), data harmonization (enabling compensation for differences in imaging protocols that typically generate noise in non-AI imaging studies) and federated learning. Thereafter, we dedicate specific sections to: sample size calculation, considering multiple testing in AI approaches; procedures for data augmentation to work with limited and unbalanced datasets; and the interpretability of AI models (the so-called black box issue). Pros and cons for choosing ML versus DL to implement AI applications to medical imaging are finally presented in a synoptic way. Conclusions: Biomedicine and healthcare systems are one of the most important fields for AI applications and medical imaging is probably the most suitable and promising domain. Clarification of specific challenging points facilitates the development of such systems and their translation to clinical practice.

Making Radiomics More Reproducible across Scanner and Imaging Protocol Variations: A Review of Harmonization Methods

“…This halts the training of initial lay because the gradient does not change the weights anymore. GANs are also prone to ge erate images with similar appearance as an effect of mode collapse [89] which occurs wh the generator produces only a limited or a single type of output to fool the discriminat Due to this the discriminator does not learn to come out of this trap resulting in a GA failure. Apart from this, GAN-based models can also add unrealistic artefacts in the i ages [87].…”

Deep Learning of Histopathological Features for the Prediction of Tumour Molecular Genetics

“…The requirement of large data sets for training deep learning models in digital pathology often results in WSIs collected across multiple research centres being integrated into a single data set. However, variations in WSI preparation may result in batch effects across images which must be mitigated to reduce bias and improve the generalisability of models [22]. For example, varying concentrations and volumes of stain used in slide preparation, as well as exposure to light during storage, may result in biases between WSIs.…”

“…Lack of interpretability remains a major obstacle to the widespread adoption of deep learning systems in healthcare [22]. Meaningful understanding of deep learning predictions is crucial in healthcare in order to instil trust from both a clinician and patent's perspective, and hence enable clinical translation.…”

Focus issue: Artificial intelligence in medical physics

“…When it comes to developing such applications, as reviewed by Castiglioni et al [8], each of the phases required for building them has its specific challenge. Researchers in AI need to collect a large set of high quality labelled and annotated data, as the accuracy of AI tools depends largely on the dataset used for training.…”