A convolutional neural network for fully automated blood SUV determination in oncological FDG-PET

Ziel/Aim The standardized uptake value (SUV) is widely used for quantitative evaluation in oncological FDG-PET but has well-known shortcomings as a measure of the tumor’s glucose consumption. The standard uptake ratio (SUR) of tumor SUV and arterial blood SUV (BSUV) possesses an increased prognostic value but requires image-based BSUV determination, typically in the aortic lumen. However, accurate manual ROI delineation requires care and imposes an additional workload which makes the SUR approach less attractive for clinical routine. The goal of the present work was the development of a fully automated method for BSUV determination in whole-body PET/CT.

Methodik/Methods Automatic delineation of the aortic lumen was performed with a convolutional neural network (CNN), namely U-Net. 632 FDG PET/CT scans from 4 different sites were used for network training (N = 208) and testing (N = 424). For all scans, the aortic lumen was manually delineated, avoiding areas affected by motion-induced attenuation artifacts or potential spill-over from adjacent FDG-avid regions. Performance of the network was assessed using the fractional deviations of automatically and manually derived BSUVs in the test data.

Ergebnisse/Results The trained U-Net yields BSUVs in close agreement with those obtained from manual delineation. Notably, using both CT and PET data as input for network training allows the trained network to derive unbiased BSUVs by detecting and excluding aorta segments affected by attenuation artifacts or spill-over. Comparison of manually (M) and automatically (A) derived BSUVs shows excellent concordance: the mean paired M-A difference in the 424 test cases is (mean ± SD)=(0.2 ± 3.1)% with a 95% confidence interval of [−6.6, 5.7]%. For a single test case the M-A difference exceeded 10%.

Schlussfolgerungen/Conclusions CNNs offer a viable approach for automatic BSUV determination. Our trained network exhibits a performance comparable to an experienced human observer and might already be considered suitable for supervised clinical use.