Fig. 2.

Overview of the data sets used for method development and evaluation. The nnU-Net employed for extraction of a single slice at L3/L4 level from a 3D CT scan and the CDFNet for tissue segmentation of the 2D CT slices were developed on two different datasets. Both methods were fivefold cross-validated and an ensemble of the cross-validated models was tested on the hold-out data. The regression models for integrated quality control (QC) were developed on the validation data of the cross-validated models and were also tested on the hold-out data. Finally, the entire pipeline of slice extraction, tissue segmentation, and quality control was evaluated end-to-end on the dual-center test data and compared against manual analyses