The project “Validity and Reliability of Importance Analyzes” is an international research collaboration between Prof. Dr. Dominik Heider and the University of Cagliari, funded by the DAAD. It is concerned with the identification of biomarkers for cardiovascular diseases, particularly for the prediction of heart attacks. This examines how the stenosis diameter (degree of vessel narrowing) can be predicted using non-invasively obtained parameters. This is intended to help replace invasive procedures such as catheter examination.

The project will use feature selection and feature extraction methods to determine the most relevant biological markers. Both univariate methods and multivariate models are used. In particular, the variable importance measures (VIMs) of random forests are in focus because they play a central role in biomarker analysis. However, since various implementations of these VIMs exist, such as the Gini index or the error rate, there is evidence that some methods may have biases. The aim is to analyze the limitations and weaknesses of existing methods and to develop an optimized procedure that delivers more reliable results.

The project aims to develop new algorithms for the non-invasive diagnosis of heart attack risks. By combining statistical methods, machine learning and medical data analysis, the results could contribute to better early detection of cardiovascular diseases in the long term.

Publications: doi: 10.1186/s13040-017-0142-8; doi: 10.1186/s13040-016-0114-4; doi: 10.5114/aoms.2018.75678; doi: 10.1097/MD.0000000000002807 

Contact: Univ.-Prof. Dr. Dominik Heider