1. To understand what radiomics is.
2. To appreciate the clinical potential of radiomics in radiology.
3. To address the professional challenges of radiomics.
The term "Radiomics" has emerged a few years ago and is attracting a lot of attention as a tool to extract quantitative information out of the images. Several softwares are on the market that provide a large amount of quantitative features from medical images using data-characterisation algorithms.These features, termed radiomic features, have the potential to uncover disease characteristics that fail to be appreciated by the naked eye. The goal of the New Horizon session is to review the concepts behind the name and appreciate the potential clinical applications in radiology.
1. To learn about new methods of image analysis derived from '-omics' methods.
2. To understand processing of big data derived from images.
3. To become familiar with new vocabulary such as radiomics, radiogenomics, clusters, heat map, etc.
Radiomics is a new 'data-driven' approach for extracting large sets of complex descriptors from routine (or not) clinical images, based on the assumption that there is a relationship between the imaging features of tumours and their underlying gene expression patterns and biology. The radiomics process aims to establish links between the imaging phenotype and genotypic and phenotypic characteristics of a tumour governed by its molecular substratum. Advanced methods of image processing are applied to images to extract a large number of descriptors, such as texture analysis from histograms, co-occurrence matrices, and fractal analysis. This large set of data can be analysed using bioinformatics and biostatitics methods into clusters defining metadata sets describing combinations of imaging features, or imaging 'profiles'. Finally, these data can be correlated to gene expression profiles, often called radiogenomics, or to outcomes, such as treatment response or survival.
1. To learn about the robustness and reproducibility of radiomics features.
2. To understand the differences and similarities between radiomics software.
3. To understand that quality of big data is the key.
Radiomics is increasingly more important in cancer research; the high-throughput mining of quantitative image features from standard-of-care medical imaging that enables data to be extracted and applied within clinical decision support systems to improve diagnostic, prognostic, and predictive accuracy. The field of radiomics is emerging rapidly. However, this field lacks standardized evaluation of both the scientific integrity and the clinical significance of the numerous published radiomics investigations resulting from this growth. With the prospect of multicentre clinical applications, it has become clear that variation in, for example, software implementations, feature nomenclature, mathematical definitions and methodology, makes reproducibility and validation of studies in radiomics a major challenge. For radiomics to mature as a discipline, there is a clear need for rigorous evaluation criteria, reporting guidelines and tools to facilitate standardization, interoperability and advancement of the field.
1. To understand how radiomics features relate to the underlying biology.
2. To learn what information radiomics can give on tumour heterogeneity.
3. To become familiar with potential added value of radiomics in predicting treatment response and outcome.
Tumour heterogeneity in cancers has been observed at the histologic and genetic levels, and increased levels of intra-tumour genetic heterogeneity have been reported to be associated with adverse clinical outcomes. In oncologic imaging, phenotypic heterogeneity between and within tumours of a given patient is readily apparent and various imaging features are routinely described subjectively in radiology reports. However, recently, imaging research has focused increasingly on the newly emergent field of radiomics, which involves a high-throughput process in which a large number of shape, edge and texture metrics are extracted and quantified objectively and in a reproducible form. These quantitative metrics can provide important insights into tumour phenotype as well as the interaction of the tumour with its microenvironment, referred to as “habitat imaging”. In the effort to delineate the biological and clinical implications of these new quantitative metrics, radiomic metrics obtained from MRI including diffusion-weighted and dynamic-contrast-enhanced MRI sequences, computed tomography and FDG-PET/CT have been further correlated with genomics data, a process intrinsic to the field known as radiogenomics. Radiogenomics and outcome data can be meaningfully mined with the goal of developing robust biomarkers that may potentially aid cancer diagnosis, improve assessment of treatment response and better predict patient outcome.
1. To understand the potential role in oncological and non-oncological clinical settings.
2. To define prerequisites for the integration into clinical workflows, including: patient acceptance, legal challenges and quality control.
3. To appreciate the future roadmap and its impact on training of young radiologists.
Radiomics, either as feature classifiers or in the form of machine intelligence, is poised to play an increasing role in clinical imaging. It offers ample potential to extend the current practice of oncological and non-oncological imaging: currently around the corner are automatic anatomy annotation, next-generation CAD systems, lesion tracking and tumour heterogeneity analysis. More complex systems with an integrated understanding of a wider range of pathologies seem markedly further down the road. Before radiomics can be implemented and used to enrich our well-tested and robustly working workflows, a number of prerequisites need to be met: it is important to understand who wants and accepts semiautomated and/or fully automated decision-making systems in radiology. Patients, insurance companies and health-care provides are important stakeholders and might or might not be drivers of the implementation of radiomics. Additionally, current laws governing regulation of medical devices are drafted with monolithic, invariable products in mind. This could be impracticable for continuously self-improving machine learning systems. Lastly, quality control measures will certainly play an important role in tackling these issues, including the liability in case of "algorithmic malpractice". Is seems reasonable to assume, that future generations of radiologists, instead of being replaced, will have a wider range of quantitative information at their disposal. Knowledge in data science to better understand, maintain and extend the systems will be a prime skill, just as technological understanding of imaging is part of every radiological curriculum.