Statistical Exploration of Relationships Between Routine and Agnostic\n Features Towards Interpretable Risk Characterization

Abstract

As is typical in other fields of application of high throughput systems,\nradiology is faced with the challenge of interpreting increasingly\nsophisticated predictive models such as those derived from radiomics analyses.\nInterpretation may be guided by the learning output from machine learning\nmodels, which may however vary greatly with each technique. Whatever this\noutput model, it will raise some essential questions. How do we interpret the\nprognostic model for clinical implementation? How can we identify potential\ninformation structures within sets of radiomic features, in order to create\nclinically interpretable models? And how can we recombine or exploit potential\nrelationships between features towards improved interpretability? A number of\nstatistical techniques are explored to assess (possibly nonlinear)\nrelationships between radiological features from different angles.\n

Ähnliche Arbeiten