Automatic Segmentation of the Human Body from CT Scanograms : Evaluation of Various Segmentation Methods

Sammanfattning

The purpose of this study was to implement and evaluate segmentation methods for body contours in pre-existing two-dimensional CT scanograms. Segment Anything Model (SAM) was used for semi-automatic annotation of 500 CT scanograms, which were subsequently used as ground truth for training nnU-Net. The segmentation performance was evaluated using Dice score, Precision, Recall and Hausdorff distance, and compared with the conventional segmentation methods Otsu thresholding and K-means clustering. The results showed that nnU-Net achieved the highest segmentation performance compared with Otsu thresholding and K-means clustering. SAM enabled efficient semi-automatic annotation, although some patient cases required more user interaction than others. The study shows that nnU-Net is a robust tool for automatic segmentation of CT scanograms and has the potential to be used in future clinical applications within medical image analysis.