PURPOSE: This paper presents a deep learning-based multi-label segmentation network that extracts a total of three separate adipose tissues and five different muscle tissues in CT slices of the third lumbar vertebra and additionally improves the segmentation of the intermuscular fat. METHOD: Based on a self-created data set of 130 patients, an extended Unet structure was trained and evaluated with the help of Dice score, IoU and Pixel Accuracy. In addition, the interobserver variability for the decision between ground truth and post-processed segmentation was calculated to illustrate the relevance in everyday clinical practice. RESULTS: On average, the presented approach achieved 91.0±0.065% DSC for the muscle tissues and 88.9±0.062% DSC for the adipose tissues. It was shown that by post-processing the intermuscular fat tissue, physicians prefer the result of the algorithm presented in the paper to their segmentation by 91.51%. CONCLUSION: The algorithm provided more precise segmentations of muscles and adipose tissue, demonstrating high-quality performance in segmenting muscle tissue. In qualitative evaluations, physicians preferred the algorithm's segmentation over expert segmentations, with the preference quantified as 91.51%. This indicates that, based on their assessment, the algorithm's results were significantly favored. This qualitative feedback supports the algorithm's use in subsequent analyses of patient fitness, leveraging the detailed information it provides.