Background <p>Modern wound and surgical care often uses 3D body scans as the input, and automated machine learning assessment of these artefacts offer increased efficiency and quality of care especially for military traumas.</p> Method <p>While segmentation of two-dimensional images is a highly common deep learning task, there is less previous work on 3D segmentation. This paper chronicles our efforts to segment gunshot wounds from 3D scans of patients injured during armed conflict. Working on the outputs of 3D imaging software, we investigated various machine learning methods including 3D-UNet and conventional segmentation techniques. The method presented here is a novel color and material-based segmentation approach that utilizes 2D projection for 3D objects.</p> Results <p>Preliminary experiments demonstrate a segmentation accuracy of 95%, with a Dice coefficient of 45% (± 29% standard deviation), measured based on correctly identified vertices. It is important to note that the dataset was manually annotated by the authors, which may influence the evaluation metrics. Additionally, a web-based user interface has been developed to enable interactive exploration of the results, serving as a prototype for future integration into full-scale surgical planning software.</p> Conclusions <p>Creating accurate assessment tools for war wounds can help save precious times in contexts where doctors are already under a lot of pressure. This article presents segmentation methods as a first step towards assessment of 3D wounds.</p> <p>Level of Evidence: Not gradable</p>