SPOCKMIP: Segmentation of Vessels in MRAs with Enhanced Continuity using Maximum Intensity Projection as Loss

TL;DR

This paper addresses the challenge of segmenting vessels of varying sizes in high-resolution 7T TOF MRA with limited annotations by introducing MIP-based loss signals to enforce vascular continuity. Building on a UNet-MSS backbone, the authors propose SPOCKMIP, which adds a Maximum Intensity Projection loss along the slice dimension and across multiple axes, combined via a weighted objective. Evaluations on the StudyForrest SF7-Tesla dataset show that MIP loss improves Dice, AUC, and continuity metrics, with the single-axis MIP providing stable gains and the multi-axis MIP yielding the strongest performance for the non-regularized baseline, while potentially over-regularizing some MSS configurations. The results suggest practical benefits for accurate, continuous vessel segmentation in neuroimaging, and point to future work on volumetric MIP loss and integration with semi-supervised learning frameworks.

Abstract

Identification of vessel structures of different sizes in biomedical images is crucial in the diagnosis of many neurodegenerative diseases. However, the sparsity of good-quality annotations of such images makes the task of vessel segmentation challenging. Deep learning offers an efficient way to segment vessels of different sizes by learning their high-level feature representations and the spatial continuity of such features across dimensions. Semi-supervised patch-based approaches have been effective in identifying small vessels of one to two voxels in diameter. This study focuses on improving the segmentation quality by considering the spatial correlation of the features using the Maximum Intensity Projection~(MIP) as an additional loss criterion. Two methods are proposed with the incorporation of MIPs of label segmentation on the single~(z-axis) and multiple perceivable axes of the 3D volume. The proposed MIP-based methods produce segmentations with improved vessel continuity, which is evident in visual examinations of ROIs. Patch-based training is improved by introducing an additional loss term, MIP loss, to penalise the predicted discontinuity of vessels. A training set of 14 volumes is selected from the StudyForrest dataset comprising of 18 7-Tesla 3D Time-of-Flight~(ToF) Magnetic Resonance Angiography (MRA) images. The generalisation performance of the method is evaluated using the other unseen volumes in the dataset. It is observed that the proposed method with multi-axes MIP loss produces better quality segmentations with a median Dice of $80.245 \pm 0.129$. Also, the method with single-axis MIP loss produces segmentations with a median Dice of $79.749 \pm 0.109$. Furthermore, a visual comparison of the ROIs in the predicted segmentation reveals a significant improvement in the continuity of the vessels when MIP loss is incorporated into training.

Figures (7)

• Figure 1: The Modified UNet MSS network is trained on 3D patches created from the input volume. MIPs of multi-level predictions are then compared against their corresponding patches on the MIP of the ground truth to compute MIP loss. The MSS loss is computed by comparing multi-level patch predictions with the corresponding label patches. A weighted sum of the two losses is then backpropagated.
• Figure 2: (a) MIP loss is calculated by comparing the MIP of the network's predictions at each level for a patch with the corresponding patch on the MIP of the ground truth segmentation. (b) Overall MIP loss is calculated by comparing the MIPs along three perceivable axes of the network's predictions at each level for a patch with the corresponding patches on the respective MIPs along multiple axes of the ground truth segmentation.
• Figure 3: Wrap-around artefact
• Figure 4: Label creation using Ilastik and 3D slicer
• Figure 5: Violin plots comparing the performance of proposed methods and baselines over three test volumes, excluding the volume with wrap-around artefacts, across five folds.