Table 1.

Overview of commonly extracted feature types in studies developing ML prediction models.

Feature Type	Explanation
Clinical	Describe patient demographics, e.g., gender and age.
Deep learning extracted	Derived from pre-trained deep neural networks.
First-order	Create a three-dimensional (3D) histogram out of tumor volume characteristics, from which mean, median, range, skewness, kurtosis, etc., can be calculated [35].
Higher-order	Identify repetitiveness in image patterns, suppress noise, or highlight details [35].
Qualitative	Describe visible tumor characteristics on imaging using controlled vocabulary, e.g., VASARI features (tumor location, side of lesion center, enhancement quality, etc.).
Second-order	Classify texture characteristics, e.g., contrast, correlation, dissimilarity, maximum probability, grey level run length features, etc. [35]
Shape and size	Describe the statistical inter-relationships between neighboring voxels, e.g., total volume or surface area, surface-to-volume ratio, tumor compactness, sphericity, etc. [35]