Efficient Cross-Modality Abdominal Organ Segmentation Using nnU-Net and MIND Descriptors

Yannick Kirchhoff, Maximilian Rouven Rokuss, Benjamin Hamm, Ashis Ravindran, Constantin Ulrich, Klaus Maier-Hein, Fabian Isensee

Keywords: FLARE Challenge, Organ Segmentation, nnU-Net, MIND Descriptors

Abstract: Accurate segmentation of abdominal organs in magnetic resonance imaging (MRI) is essential for diagnosis and treatment planning. However, this task is challenging due to the scarcity of labeled MRI data and significant differences in appearance between MRI and computed tomography (CT) images. Task 3 of the FLARE 2024 challenge was launched to encourage the development of algorithms capable of transferring knowledge from labeled CT scans to unlabeled MRI scans for efficient abdominal organ segmentation under strict resource constraints. In this paper, we describe our contribution to this challenge by utilizing nnU-Net combined with modality-independent neighborhood descriptor (MIND) features to transfer labels from CT to MRI. Our method achieved an average Dice Similarity Coefficient (DSC) of 57.7\% and an average Normalized Surface Dice (NSD) of 59.8\% on the validation set, with an average running time of 20 seconds and an area under the GPU memory-time curve of 73,607 MB. These results demonstrate that our approach effectively addresses the challenges of cross-modality abdominal organ segmentation under resource constraints, highlighting the potential of modality-independent descriptors for label transfer in medical imaging.

Submission Number: 22