Enhancing Neurofibroma Segmentation in Whole-Body MRI: Leveraging an Anatomy-Informed Approach

Georgii Kolokolnikov, Marie-Lena Schmalhofer, Inka Ristow, René Werner

Keywords: Neurofibroma, whole-body MRI, deep learning, medical image segmentation, anatomy-informed approach, TotalSegmentator

Abstract: This study presents an anatomy-informed segmentation approach for neurofibroma in fat-suppressed T2-weighted whole-body MRI (WB-MRI). By adapting TotalSegmentator for WB-MRI segmentation and employing dedicated Dynamic UNet models across four anatomical zones, we achieved improvements of 20\% in terms of the Dice coefficient on a test set. The proposed method promises to streamline neurofibroma segmentation, emphasizing future integration into interactive workflows.

Submission Number: 17