Universiteit Maastricht

Sebastian Sanduleanu

The present document describes a training method of a machine learning data processing model for determining a hypoxia status of a neoplasm, in particular a random forest model. The method comprises obtaining, for a plurality of neoplasms, at least one data sample comprising 3D imaging data. A hypoxic volume fraction is determined for each data sample, as well as a set of image features associated with the neoplasm. The method further iterates a sequence of training steps and each iteration includes: selecting a subset of image features and eliminating, for each data sample, the subset of image features to yield a reduced set of image features. The iteration also includes generating decision trees, providing a momentary random forest model based thereon, and submitting a test set of image features to the momentary random forest model to yield a performance value. The iterations are continued until all image features have been selected for a subset at least once, and then a plurality of preferred image features are selected for providing a radiomics feature signature. The trained random forest data processing model based on decision trees associated with the preferred image features of the radiomics feature signature.

Method of training a machine learning data processing model, method of determining a hypoxia status of a neoplasm in a human or animal body, and system therefore

Growing community of inventors

Brunssum, Netherlands

Sebastian Sanduleanu