Skip to main content
Fig. 2 | Radiation Oncology

Fig. 2

From: X-change symposium: status and future of modern radiation oncology—from technology to biology

Fig. 2

Machine learning in prognostic modelling using omics data. a Omics data consisting of N (number of patients) times G (number of molecular features) measurements in combination with time-to-event data are the basis in prognostic machine learning. Features can be measurements from any molecular layer such as the genome, transcriptome, proteome, post-transcriptome or epigenome. b Machine learning selects the molecular features to be added to a Cox pH model in such a way that the time-to-event data are optimally explained, while overfitting of data must be prevented. Here, exemplarily iterative forward-selection is shown. c The best model is used to calculate risk scores for all patients that allow, in combination with a cutoff, assignment of patients to high- or low-risk groups with regard to the chosen time-to-event endpoint (e.g. overall survival). The model coefficients for each molecular feature of the model, in combination with the defined risk score cutoff, allow the typing of new patients assigned to risk groups and thus the prediction of the individual risk for the endpoint of interest

Back to article page