Training for Mass Effect Model

[shiblyg]

Hi Ivan, I am wondering, inside the data loader the outputmodes used for different combination of parameter prediction for reaction diffusion model. But for the mass effect model in the paper table 2 you reported the same predicted parameter estimation. However to generate simulated tumor using mass effect it requires different parameter (like gamma, nu,dx and others are same). I am confused how you train your network for mass effect and reaction diffusion model and estimating the same combination of parameter. Did you just use constant velocity for reaction diffusion model and random velocity for mass effect. Is the code inside torchcode you provided is only work for reaction diffusion model only? Since in the evaluation code I only found you simulated the predicted parameter using glioma solver only there is no evaluation code using mass effect model? Thank you

[IvanEz]

We were interested in predicting reaction-diffusion dynamics. As we mention in the paper (page 12), the values for the elasticity model parameters λ, µ, γ were taken from (Subramanian et al., 2019).

[shiblyg]

Thank you for the clarification. And could you please clarify it: 1

1) In figure 1 during inference for real MRI the input is the (T1gd+Flair) morphed (binary segmentation) right! The steps I am following: -register patient MRI to atlas space (both flair and T1gd) to get atlas anatomy -morphed MRI (binary segmentation fromt1gd and flair )to get morphed segmentation -use this for inference input 2) I was a little bit confused for the training did you use the simulated tumor (channel 0)or simulated tumor with brain anatomy (channel 1 generated by Glioma solver). Thank you