Pilot Study Examines Machine Learning for Glioblastoma Radiation Treatment

A recent pilot study published in Nature Communications outlines the potential of a personalized approach to treating glioblastoma through machine-learning-guided radiation dose escalation. Glioblastoma, a notoriously aggressive brain tumor, frequently recurs despite the most extensive treatment protocols.

The study, identified under ClinicalTrials.gov (NCT03477513), involved integrating machine learning (ML) techniques to develop precision radiation therapy (PPRT) using tumor infiltration maps. This pilot investigation included 20 adult patients with newly diagnosed IDH-wildtype glioblastoma who had undergone gross total resection, followed by concurrent and adjuvant temozolomide chemotherapy. The principal aim was to determine the feasibility and safety of the PPRT approach compared to standard chemoradiotherapy.

The findings suggest that PPRT was both achievable and tolerated well, with patients experiencing no severe (grade ≥3) acute adverse events. However, a subset experienced milder grade 1 and 2 adverse events, with occurrences at rates of 47% and 53%, respectively. Within this cohort, increased radiation necrosis was observed (47% in the PPRT-temozolomide recipients) compared to a 12% necrosis rate in those receiving standard treatment (P < 0.001).

In terms of efficacy, those who received PPRT exhibited promising outcomes in terms of median progression-free survival (PFS) and overall survival (OS). The PPRT group demonstrated a median PFS of 24.4 months and OS of 35.4 months, outperforming a matched historical control group, whose figures were 11.6 and 17.7 months, respectively. The hazard ratios indicated a significant reduction in the risk of progression (HR 0.28, 95% CI 0.13–0.61; P = 0.001) and mortality (HR 0.34, 95% CI 0.17–0.69; P = 0.003).

While initial results appear promising, indicating possible survival benefits, these findings are considered preliminary and emphasize the need for further validation through randomized clinical trials. The study was partially funded by the Penn Abramson Cancer Center, with no influence on the experimental design, data analysis, or manuscript preparation.