Noise modeling of tumor size measurements from animal experiments for virtual patient generation

Optimizing computer-generated therapy may be one of the most promising tools for future medical treatments. In silico experiments are crucial for therapy planning and testing, this requires a reliable virtual patient model. We present a noise model that can be used to model real measurement noise from mice experiments and add this noise to the virtual patient model. In the experiments, the measurements are performed with a digital caliper, and the tumor is measured through the skin, which is the reason why significant nonlinear measurement noise can be expected. We use data from preclinical experiments to estimate the measurement noise and use a nonlinear transformation to whiten the noise. Finally, we find the distributions that most accurately describe the transformed noises based on the Anderson-Darling test. We show that a virtually generated measurement with noise generated by our noise model is similar to actual measurements.