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

Chethan Radhakrishna1, Karthikesh Varma Chintalapati1, Sri Chandana Hudukula Ram Kumar1, Raviteja Sutrave1, Hendrik Mattern2,4,5, Oliver Speck2,4,5, Andreas Nürnberger1,3,5, Soumick Chatterjee1,3,6
1Faculty of Computer Science, Otto von Guericke University Magdeburg, Germany
2Biomedical Magnetic Resonance, Otto von Guericke University Magdeburg, Germany
3Data and Knowledge Engineering Group, Otto von Guericke University, Magdeburg, Germany
4German Centre for Neurodegenerative Disease, Magdeburg, Germany
5Centre for Behavioural Brain Sciences, Magdeburg, Germany
6Genomics Research Centre, Human Technopole, Milan, Italy
Email: contact@soumick.com
arXiv:2407.08655
Part of the EL-AURIAN project
Preprint Code 🤗 Model Weights
MY ALT TEXT

Method overview

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. In this study, UNet and UNet MSS models with ReLU activation replaced by LeakyReLU are trained on the Study Forrest dataset. 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 ± 0.129. Also, the method with single-axis MIP loss produces segmentations with a median Dice of 79.749 ± 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.


MY ALT TEXT

Visual comparison of ROIs of MIPs of segmentations resulting from training all the models with MIP loss along the z-axis and MIP loss along multiple axes. Annotations on the MIPs show the missing continuity of the vessels and the improvements after using the z-axis and multi-axes MIP loss. Each image is an overlay of network predictions on the ground truth, where blue and red represent false positives and false negatives, respectively.

MY ALT TEXT

Comparison of False Positives between segmentations generated by UNet-MSS\_MIP and UNet MSS mMIP. Each image is an overlay of network predictions on the Ground Truth, where blue annotations represent false positives and red annotations represent false negatives.

MY ALT TEXT

Metrics comparisons of 15 segmentation results of test volumes, excluding the volume with wrap-around artefacts, over 5-fold cross-validation.

BibTeX


          @article{radhakrishna2024spockmip,
            title={SPOCKMIP: Segmentation of Vessels in MRAs with Enhanced Continuity using Maximum Intensity Projection as Loss},
            author={Radhakrishna, Chethan and Chintalapati, Karthikesh Varma and Kumar, Sri Chandana Hudukula Ram and Sutrave, Raviteja and Mattern, Hendrik and Speck, Oliver and N{\"u}rnberger, Andreas and Chatterjee, Soumick},
            journal={arXiv preprint arXiv:2407.08655},
            year={2024}
          }