Glioblastoma: STRIDE distinguishes between true progression and pseudoprogression by taking into account the interval between two MRI scans

This model analyses pairs of post-treatment MRI scans and the time interval between them to classify tumour progression, with an AUC of 0.816 in the Burdenko cohort.

After treatment, glioblastoma, an aggressive primary brain tumor, is monitored by successive MRIs. However, a stable disease, a pseudoprogression (inflammatory reaction to treatment) and a real progression can look similar in the image, even though they call for very different clinical decisions. Researchers propose STRIDE, a model that takes two consecutive examinations and the real time interval between them.

Modeling evolution, not just the image

STRIDE combines a spatial representation guided by the location of the lesion, which preserves the surrounding context, and a transition module that estimates how the representation should evolve according to the time elapsed. The comparison between the predicted state and the state actually observed on the second examination is used to characterize the dynamics of the disease.

Developed on BraTS2024, pre-trained on the longitudinal LUMIERE database and then adapted to the Burdenko cohort, the model achieves a macro AUC of 0.816 and a macro F1 score of 0.796 on the three-class classification. The authors see this as a step towards more reliable post-treatment follow-up, which remains to be validated under clinical conditions.