Exploring Automated Contouring Across Institutional Boundaries: A Deep Learning Approach with Mouse Micro-CT Datasets

Image-guided mouse irradiation is essential to understand interventionsinvolving radiation prior to human studies. Our objective is to employ SwinUNEt Transformers (Swin UNETR) to segment native micro-CT and contrast-enhancedmicro-CT scans and benchmark the results against 3D no-new-Net (nnU-Net). SwinUNETR reformulates mouse organ segmentation as a sequence-to-sequenceprediction task, using a hierarchical Swin Transformer encoder to extractfeatures at 5 resolution levels, and connects to a Fully Convolutional NeuralNetwork (FCNN)-based decoder via skip connections. The models were trained andevaluated on open datasets, with data separation based on individual mice.Further evaluation on an external mouse dataset acquired on a differentmicro-CT with lower kVp and higher imaging noise was also employed to assessmodel robustness and generalizability. Results indicate that Swin UNETRconsistently outperforms nnU-Net and AIMOS in terms of average dice similaritycoefficient (DSC) and Hausdorff distance (HD95p), except in two mice ofintestine contouring. This superior performance is especially evident in theexternal dataset, confirming the model's robustness to variations in imagingconditions, including noise and quality, thereby positioning Swin UNETR as ahighly generalizable and efficient tool for automated contouring inpre-clinical workflows.