There is much more to diagnostic imaging than what meets the eye. While quantitative approaches are not new in the field of oncologic imaging, the multi-step process known as radiomics has turned medical images into mineable data and many believe it will pave the way towards precision medicine. The assumption behind radiomics is that image heterogeneity at a pixel level might reflect tumoral heterogeneity at a biological level. Radiogenomics, in particular, investigates possible correlations between tumor gene expression and imaging features. A virtually unlimited number of quantitative parameters can be extracted through radiomics from all imaging modalities routinely employed in clinical practice, with many possible biomarkers to be discovered. These high-dimensional datasets can be efficiently and effectively managed using artificial intelligence. Indeed, machine learning and deep learning can be used to build radiomics-powered predictive models and decision support tools.
Unfortunately, the translation of radiomics into clinical practice is proving rather difficult to achieve. Radiomics pipelines need to be sufficiently robust and generalizability is a major concern. Indeed, high dimensionality and the risk of overfitting undermine the findings of many preliminary investigations. Issues related to each step of radiomics workflow, from segmentation (e.g. feature stability) to model validation (e.g. the need for external datasets) must be addressed. Nevertheless, the promises of radiomics in the field of genitourinary oncology are definitively worthy of being explored. Radiomics models might increase the diagnostic accuracy of imaging modalities in the detection of tumors, aid radiologists in the characterization of imaging findings, and even predict treatment response or patient prognosis. With this Research Topic, we aim to contribute in moving radiomics of genitourinary tumors to the next step. It is time to build solid grounds and support the claims that radiomics might truly redefine the role of imaging in genitourinary oncology. More specifically, we wish to highlight the added value of artificial intelligence and deep learning in this field.
The topics of interest include, (but are not limited to):
• Development of radiomics-powered deep learning tools (e.g. CAD systems)
• Reproducibility of radiomics features in genitourinary tumors
• Development of diagnostic/prognostic models with state-of-the-art radiomics pipelines and artificial intelligence techniques
• Validation of previously published radiomics signatures
• Evaluation of the clinical applicability of radiomics models
• Investigating correlations between radiomics features and genitourinary tumors gene expression
• Artificial intelligence applications to radiomics in genitourinary oncology, specifically original articles, and reviews (narrative and systematic)
There is much more to diagnostic imaging than what meets the eye. While quantitative approaches are not new in the field of oncologic imaging, the multi-step process known as radiomics has turned medical images into mineable data and many believe it will pave the way towards precision medicine. The assumption behind radiomics is that image heterogeneity at a pixel level might reflect tumoral heterogeneity at a biological level. Radiogenomics, in particular, investigates possible correlations between tumor gene expression and imaging features. A virtually unlimited number of quantitative parameters can be extracted through radiomics from all imaging modalities routinely employed in clinical practice, with many possible biomarkers to be discovered. These high-dimensional datasets can be efficiently and effectively managed using artificial intelligence. Indeed, machine learning and deep learning can be used to build radiomics-powered predictive models and decision support tools.
Unfortunately, the translation of radiomics into clinical practice is proving rather difficult to achieve. Radiomics pipelines need to be sufficiently robust and generalizability is a major concern. Indeed, high dimensionality and the risk of overfitting undermine the findings of many preliminary investigations. Issues related to each step of radiomics workflow, from segmentation (e.g. feature stability) to model validation (e.g. the need for external datasets) must be addressed. Nevertheless, the promises of radiomics in the field of genitourinary oncology are definitively worthy of being explored. Radiomics models might increase the diagnostic accuracy of imaging modalities in the detection of tumors, aid radiologists in the characterization of imaging findings, and even predict treatment response or patient prognosis. With this Research Topic, we aim to contribute in moving radiomics of genitourinary tumors to the next step. It is time to build solid grounds and support the claims that radiomics might truly redefine the role of imaging in genitourinary oncology. More specifically, we wish to highlight the added value of artificial intelligence and deep learning in this field.
The topics of interest include, (but are not limited to):
• Development of radiomics-powered deep learning tools (e.g. CAD systems)
• Reproducibility of radiomics features in genitourinary tumors
• Development of diagnostic/prognostic models with state-of-the-art radiomics pipelines and artificial intelligence techniques
• Validation of previously published radiomics signatures
• Evaluation of the clinical applicability of radiomics models
• Investigating correlations between radiomics features and genitourinary tumors gene expression
• Artificial intelligence applications to radiomics in genitourinary oncology, specifically original articles, and reviews (narrative and systematic)