Advanced Synergy between Perfusion Imaging, Artificial Intelligence, and Oncology: Towards Personalized Management of Radiotherapy Treatments in Grade 4 Astrocytoma IDH mutated and Glioblastoma.

Abstract: Background: In high-grade gliomas (HGG), predicting the site of relapse (L3R) from pre-radiotherapy (pre-RT) MRI is crucial for personalizing treatment strategies. Advanced perfusion biomarkers derived from the Cercare Medical Neurosuite software, such as Capillary Transit Time Heterogeneity (CTH) and Oxygen Extraction Fraction (OEF), may provide new insights beyond conventional MRI sequences. Material and Methods: We retrospectively analyzed a monocentric cohort of 186 adult HGG patients, all with pre-RT MRI including T2*-DSC perfusion imaging, and MRI at recurrence. Perfusion maps (rCBV, CTH, OEF) were extracted, and voxel-wise prediction models were developed to assess their ability to identify L3R. ROC-AUC values were calculated to quantify predictive performance. Relapse patterns were also classified in relation to the 95% isodose (in-field, marginal, distant). Clinical, imaging, surgical, and treatment-related prognostic factors and survival outcomes were included in the analysis. Results: In the overall cohort, 81% of recurrences occurred in-field. CTH showed the highest predictive performance for L3R (AUC 0.71), followed by OEF (AUC 0.61), while rCBV had poor discrimination (AUC 0.51). Selected Clinical Cases: Case 1 vs. Case 2: In a patient treated with subtotal resection, CTH and OEF accurately delineated L3R (AUC 0.96 and 0.89, respectively), while rCBV underperformed (AUC 0.78). A multifocal “button” visible on pre-RT rCBV and T1 post-contrast (T1Gd) but not on CTH or OEF did not correspond to relapse, suggesting low aggressiveness. Conversely, in Case 2, a region hypointense on T1Gd and normal on rCBV showed high CTH and OEF: relapse occurred precisely there. Case 3: In a biopsy patient, CTH and OEF maps revealed pre-RT hyperintensities extending toward the genu of the corpus callosum. rCBV and T1Gd were unremarkable. Relapse followed that pathway (CTH AUC 0.81, OEF AUC 0.86; rCBV AUC 0.64). Case 4:In a patient with no residual tumor after gross total resection, CTH and OEF detected subtle hyperintensities at the anterior-medial border of the surgical cavity on pre-RT imaging, where rCBV and T1 post-c.e. gave no signal. The tumor relapsed precisely from that region, with anterior growth, demonstrating predictive capacity (CTH AUC 0.84, OEF AUC 0.76, while rCBV 0.16). Conclusion: In this large cohort of HGG patients, advanced perfusion biomarkers significantly outperformed rCBV in predicting future relapse sites. Combining anatomical MRI, rCBV, and advanced perfusion maps enables a voxel-wise aggressiveness profile, opening new opportunities for personalized, risk-adapted radiotherapy strategies to improve disease control in HGG.