Figure 1
From: Radiomics strategies for risk assessment of tumour failure in head-and-neck cancer
From radiomics analysis to treatment personalization. (a) Example of diagnostic FDG-PET and CT images of two head-and-neck cancer patients with tumour contours. The patient that did not respond well to treatment (right) has a more heterogeneous intratumoural intensity distribution in both FDG-PET and CT images than the patient that responded well to treatment (left). (b) The radiomics analysis strategy involves the extraction of features differentiating responders from non-responders to treatment. Features are extracted from the FDG-PET and CT tumour contours and quantify tumour shape, intensity, and texture. (c) Advanced machine learning combines radiomic features and patient clinical information via a random forest algorithm. The classifier is trained to differentiate between responders and non-responders to treatment (prediction model). (d) The output probability of the random forest classifier computed on new patients can be used to assess the risk of non-response to treatment via probabilities of occurrence of outcome events and time estimates. Eventually, accurate risk assessment of specific tumour outcomes via radiomics analysis could help to better personalize cancer treatments.
