Figure 6.

Two example cases with good and bad performance from the validation set at the end of the shape space learning (1. training phase) and their interpolation at different t_{Imaging → Treatment}∈[0, 5] values. The first row shows an axial slice of an accurate reconstruction of core at 0h as well as of the penumbra, and as a consequence the non-linear interpolation of both gives a reliable estimation of the true follow-up lesion at its actual t_{Imaging → Treatment} value (outlined in purple). The second row shows a case where the temporal progression of the stroke is different, such that the final lesion was much smaller than a interpolation between core and penumbra would suggest.