Fig. 1.

Schematic representation of the presented pipeline for autonomous body composition analysis. Input of the pipeline is a 3D CT scan. In the first part, a 3D convolutional neural network (CNN) was employed for slice extraction using nnU-Net. In the second part, a competitive dense fully connected CNN (CDFNet) is applied for segmentation of the body compartments. Classical machine learning methods were employed for integration of quality control steps. For the slice extraction part, a logistic regression model was developed that classifies the presence of L3/L4 lumbar level in the 3D CT scan. For segmentation of the different tissues, a linear regression model was established that predicts segmentation quality in terms of the Dice score