A scanner-inline software for performing permuted random forest for CEST frequency importance analysis

Motivation: CEST quantitation typically relies on model-based fitting and always performs off-scanner. Besides, model-based fitting requires collection a number of saturation frequencies, hindering the clinical applications. Goal(s): To facilitate CEST applications by implementing a scanner-inline software through model-free analysis. Approach: We implemented CEST frequency importance analysis on Philips pride platform, which could rank the acquired frequencies according their contribution to lesion classification, using a permuted random forest algorithm. Results: Without specific requirement for sampled frequencies, this software allows researchers to extract frequency importance feature, either between lesion voxels and control ones, or between two different time points or different subjects. Impact: Compared with the conventional analysis based on fitting line-shape of the spectra, this PRF method does not have specific requirement for sampled frequencies on spectra, but fully explore all acquired ones, which is user-friendly and facilitate CEST applications.