Interim FDG PET analysis performed by Neural Networks to predict the outcome of patients with aggressive B-cell lymphoma

Ziel/Aim Patients with aggressive B-cell lymphoma typically undergo FDG-PETs prior to chemotherapy and after initial chemotherapy. Therapy response assessment is usually done by employing the Deauville score, which is rating the residual metabolic activity on a 5-scale sore by comparing baseline and interim PET. However, this coarse assessment might lead to inaccurate outcome prediction of patients. Therefore, we evaluated percentual changes of the metabolic tumor volume and PET uptake for more detailed response characterization. To this end, we employed a neural network to fully automatically segment all lymphoma lesions in baseline and interim FDG-PETs.

Methodik/Methods All patients who participated in the Positron Emission Tomography-guided Therapy of Aggressive non- Hodgkin Lymphomas (PETAL) trial were regarded in this analysis, if baseline and interim whole-body FDG-PET CT were present. The PARS neural network (Siemens) was employed for fully automated PET analysis. Metabolic tumor volume (MTV) was calculated using a 41 % relative threshold for each lesion. Progression free survival (PFS) was used as primary endpoint.

Ergebnisse/Results A total number of 533 patients was included in this analysis. The interim Deauville (>3 vs. =< 3) score was a statistically significant predictor of PFS (HR: 1.63; p = 0.01, log-rank p = 0.01). The ratio in MTV between baseline and interim PET was likewise a statistically significant predictor of PFS (HR: 1.35, p = 0.02; log-rank p < 0.0001). The same was true for mean SUV_max in the interim PET (HR: 1.27, p = 0.000005; log-rank p < 0.0001). Multivariate Cox Regression confirmed the superiority of interim PET mean SUV_max over the Deauville score (HR: 1.27, p = 0.000008 vs. HR: 1.61, p = 0.0141).

Schlussfolgerungen/Conclusions Fully automatically derived FDG-PET metrics are superior to the manually set Deauville score. Future studies should evaluate neural network-based lymphoma patient stratification.

Publication History

Article published online:
08 April 2021