The invention relates to a therapy planning device for planning a therapy to be applied to tissue of a subject (2). A tissue parameter distribution, which has been generated based on a magnetic resonance fingerprint scan of the tissue, and a therapy goal distribution, which defines a distribution of at least one parameter being indicative of a desired effect of the therapy, are provided. A machine learning module (13), which has been trained to output at least one therapy application parameter defining the application of the therapy based on an input tissue parameter distribution and an input therapy goal distribution, is used for planning the therapy by determining the at least one therapy application parameter based on the provided tissue parameter distribution and the provided therapy goal distribution. This allows for a consideration of the actual quantitative tissue parameter distriution of the patient, thereby improving planning quality.