An Introduction to Radiomics: Capturing Tumour Biology in Space and Time

Currently, there is a shift from visual interpretation of medical images, which is highly variable, to the extraction of high-dimensional meaningful data the so called Radiomic signatures that can be used in conjunction with machine learning algorithms to predict clinical outcomes. Quantification of imaging biomarkers can be used to make predictions on whether a specific treatment will work for a specific patient, and can aid in differential diagnosis problems or may offer prognostic capabilities related to disease recurrence or relapse. Modern algorithms based on machine learning techniques can be used to provide automatic or semi-automatic segmentation with minimal human interaction. Feature extraction is the calculation of texture and shape imaging features that can be used along with clinical biomarkers. Feature selection is important to avoid overfitting and exclude redundant features improving the quality of data, by reducing their dimensionality. Following to that, multiple machine learning algorithms can be recruited in order to find the optimal that can provide with the best performance. Algorithms like Bays, linear regression, Support Vector Machines, Random Forests and others are currently used.