Computed tomography radiomics to predict microsatellite instability status and immunotherapy response in gastric cancer.

Radscores were an independent predictor of progression-free survival in a separate cohort of advanced unresectable gastric cancer patients receiving first-line immunotherapy, but no significant OS difference was found between high- and low-Radscore groups.

In the immunotherapy cohort, patients achieving partial remission had higher baseline Radscores than those with stable or progressive disease, and the high-Radscore group had a markedly higher partial remission rate than the low-Radscore group.

Radscores correlated with CD8+ T-cell infiltration and M2 macrophage content in a public radiotranscriptomic dataset, raising the possibility that CT texture may reflect immune microenvironment composition—though this is exploratory and not yet actionable.

Radscores showed no significant correlation with established immune checkpoint gene expression markers, which the authors speculate may indicate complementary rather than redundant immune information—a hypothesis requiring prospective validation.

The immunotherapy outcome cohort was small and retrospective, and the overall survival analysis was non-significant; results may not generalise to larger or more diverse patient populations, and prospective validation is needed before any clinical translation.

The clinical-radiomics model combining Radscores with age, clinical stage, cT-stage, cN-stage, and tumor location showed inconsistent performance across external sets: it outperformed the radiomics-only model in the training set but underperformed it in external testing set 2, highlighting generalizability concerns.

The association between Radscores and immunotherapy response does not establish that CT radiomics can replace or guide molecular testing for MSI-H status; the two analyses were conducted in separate cohorts with different endpoints.