Deep learning-based PET/CT mixture-of-experts model for relapse risk stratification in relapsed/refractory classical Hodgkin lymphoma: a multicenter study

Cancer research · Hodgkin lymphoma

The study claims that MoEScore, the deep learning‑based imaging biomarker derived from 18F‑FDG PET/CT, can stratify relapse risk in patients with relapsed/refractory classical Hodgkin lymphoma, and that the MoE model may help inform risk‑adapted therapeutic strategies.

In a multicenter retrospective study of 161 patients with relapsed/refractory classical Hodgkin lymphoma (R/R cHL), a deep learning‐based mixture‑of‑experts (MoE) framework was developed to generate a quantitative imaging biomarker (MoEScore) from baseline 18F‑FDG PET/CT. The MoEScore achieved an AUC of 0.861 in the training cohort and 0.783 in the validation cohort, and was independently associated with relapse (HR = 11.18, 95 % CI = 2.48–350.45; P = 0.002). A multiparametric model incorporating MoEScore and clinical/metabolic predictors had a C‑index of 0.785 (training) and 0.717 (validation). The model was shown to stratify relapse risk across both relapsed and refractory subgroups, with interpretability analyses indicating that PET contributed more strongly than CT. The study concludes that this multimodal deep learning framework may capture tumor heterogeneity beyond conventional metrics and could inform risk‑adapted therapeutic strategies for R/R cHL.

Accurate relapse risk stratification in R/R cHL is critical to guide individualized treatment decisions and improve patient outcomes. If the model reliably predicts relapse, clinicians could tailor salvage therapies and surveillance more effectively.

Evidence level: Számítógépes vagy elméleti. Modellből vagy adatbányászatból származó jel.

Original source

Related signals