CT-based quantification of spatiotemporal heterogeneity for predicting response to neoadjuvant chemotherapy in locally advanced gastric cancer.

Objective: To explore the efficacy of quantitative measurements of temporal changes in CT-based spatial habitat to predict pathological responses after neoadjuvant chemotherapy (NAC) in patients with locally advanced gastric cancer (LAGC).

Methods: This retrospective study included 68 LAGC patients who underwent NAC followed by radical gastrectomy. Patients were stratified into a good response (GR) group or a poor response (PR) group based on tumor regression grading (TRG). Baseline and restaging portal venous-phase CT images, along with CT-based texture entropy maps, were grouped into three spatial habitats using K-means clustering. The voxel number and fraction (f) of each habitat and their changing trends (delta) at the baseline (pre) and restaging (post) stages were quantified. Differences between the GR and PR groups were compared. The spatiotemporal habitat (SH), clinical-radiological (CR) and laboratory (Lab) models and their model scores were generated through multivariate logistic regression. A combined nomogram integrating SH, CR, and Lab scores was developed. Diagnostic accuracy was assessed using the area under receiver operating characteristic curves (AUCs).

Results: Compared to the PR group, the GR group exhibited significantly higher post-f and delta-f values, and lower post-f and post-voxel number values (p < 0.01). The GR group showed lower incidences of Borrmann III-IV, reduced post-tumor thickness, lower post-CA724 levels, and smaller delta-CEA values (p < 0.05). The combined nomogram exhibited the highest diagnostic efficacy (AUC = 0.893), significantly outperforming the CR (0.711), Lab (0.766), and SH (0.833) models (p < 0.05).

Conclusions: CT-based spatiotemporal habitat imaging demonstrates clinical utility in predicting pathological response to NAC in LAGC.