Joeky Senders

178 Chapter 10 that largely occurs outside the ‘black-box’. For example, the way patients are selected, input features preprocessed, complexities in the data accounted for, outcomes defined, hyperparameters optimized, and model performance evaluated are all specified manually based on clinical expertise and substantially determine the internal and external structure of the final model. Lastly, machine learning provides powerful methods for mapping numeric input to numeric output. However, not everything is reducible to numbers, especially not in healthcare. Primitive clinical characteristics, pictures, text, and images can all be expressed as 0’s or 1’s and thus easily be incorporated into a model, whereas human values pertinent to the patient remain irreducible to numbers. The model developed in Chapter 5 predicts personalized estimates of expected survival with high accuracy and precision; however, it cannot grasp the personal and clinical implications associated with these predictions. As such, clinical decision-making can still be very different in two patients, even if the predicted outcomes are exactly the same. Clinicians should therefore be trained in considering the appropriate machine learning tools on case-by- case basis and interpreting the clinical implications associated with their predictions. CONCLUSION The thin line between treatment effectiveness and patient harms underpins the importance of tailoring clinical management to the individual brain tumor patient. Machine learning algorithms have the potential to unlock unique insights from large, complex data sources and effectively personalize clinical decision-making to the needs of the individual brain tumor patient. However, the automated nature comes at the cost of its interpretability, which can limit their clinical implementation and acceptance. Machine learning algorithms should be considered as an extension to statistical approaches and exist along a continuum determined by how much is specified by humans and how much is learnt by the machine. The choice of algorithm should be guided by the nature and complexity of the input data, as well as the desired level of human guidance and model interpretability. Although machine learning algorithms can produce highly accurate predictions based on high-dimensional data, clinicians and researchers should interpret the clinical implications of these predictions on case- by-case basis.