Physics-informed vessel segmentation for χ-separation (chi-separation)
Poster No:
1878
Submission Type:
Abstract Submission
Authors:
Taechang Kim1, Sooyeon Ji1, Kyeongseon Min1, Jonghyo Youn1, Minjun Kim1, Jiye Kim1, Jongho Lee1
Institutions:
1Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea, Republic of
First Author:
Taechang Kim
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Co-Author(s):
Sooyeon Ji
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Kyeongseon Min
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Jonghyo Youn
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Minjun Kim
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Jiye Kim
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Jongho Lee
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Introduction:
χ-separation offers distributions of paramagnetic and diamagnetic susceptibility sources, presenting valuable information.6,8 However, both maps show erroneously high susceptibility values in vessels by flow effects7. In most studies, vessels are not of interest, interfering accurate quantification of myelin and tissue iron. For a reliable analysis, it is advantageous to remove vessels. In this study, we propose a vessel segmentation method designed for χ-separation.
Methods:
The proposed method is illustrated in Fig. 1.
[Step 1: Seed generation]
For large vessels, we first applied a high pass filter to R2* to suppress non-vessel structures3. Then, vesselness1,2, likelihood of being a vessel, was calculated and thresholded with a high value.
For small vessels, maximum intensity projection (MIP) was applied for χpara⋅|χdia| to enhance the visibility with masking out large vessel seeds from MIP. Then vesselness was calculated and thresholded with a low value. These seeds were backprojected to the original location in 3D. Finally, both large and small vessel seeds are combined, generating the total seed.
[Step 2: Vessel geometry characteristics guided-region growing & non-vessel structure removal]
If a voxel adjacent to a seed has intensity higher than the upper limit, that voxel was added to the vessel mask. If the intensity is between the upper and lower limit, it was only incorporated into the mask if the condition was satisfied (see Fig. 1, where v is vesselness, Ω is directionality similarity4, R is intensity ratio between adjacent voxels, and λ2⋅λ3 is anisotropy (two largest eigenvalues from the Hessian matrix of voxel intensity2)).
To exclude non-vessel structures, connected components (CC) whose average value of anisotropy is lower than threshold (Α∈[0.0001,0.004]) were removed assuming high anisotropy in vessels.
Evaluation The proposed method was compared with a Frangi filter2 and REF 10 with and without R2* as input. To test the robustness, the method was applied to various resolutions and field strengths (1.5×1.5×1.5 mm3 and 1×1×1 mm3 at 3 T; 0.8×0.8×1.2 mm3 and 0.65×0.65×0.65 mm3 at 7 T) and χ-separation algorithms (COSMOS9, MEDI, iLSQR, and χ-sepnet5).
Applications (1) Using 6 subject data for χ-separation-COSMOS9, quantitative metrics (RMSE, PSNR and SSIM) were calculated to evaluate the performance of χ-sepnet5 with, without and within the vessel mask. (2) Using 106 subjects, an χ-separation atlas was developed with and without the vessel mask. Twenty ROIs were analyzed to quantify the proportion of vessel and the population average.
[Step 1: Seed generation]
For large vessels, we first applied a high pass filter to R2* to suppress non-vessel structures3. Then, vesselness1,2, likelihood of being a vessel, was calculated and thresholded with a high value.
For small vessels, maximum intensity projection (MIP) was applied for χpara⋅|χdia| to enhance the visibility with masking out large vessel seeds from MIP. Then vesselness was calculated and thresholded with a low value. These seeds were backprojected to the original location in 3D. Finally, both large and small vessel seeds are combined, generating the total seed.
[Step 2: Vessel geometry characteristics guided-region growing & non-vessel structure removal]
If a voxel adjacent to a seed has intensity higher than the upper limit, that voxel was added to the vessel mask. If the intensity is between the upper and lower limit, it was only incorporated into the mask if the condition was satisfied (see Fig. 1, where v is vesselness, Ω is directionality similarity4, R is intensity ratio between adjacent voxels, and λ2⋅λ3 is anisotropy (two largest eigenvalues from the Hessian matrix of voxel intensity2)).
To exclude non-vessel structures, connected components (CC) whose average value of anisotropy is lower than threshold (Α∈[0.0001,0.004]) were removed assuming high anisotropy in vessels.
Evaluation The proposed method was compared with a Frangi filter2 and REF 10 with and without R2* as input. To test the robustness, the method was applied to various resolutions and field strengths (1.5×1.5×1.5 mm3 and 1×1×1 mm3 at 3 T; 0.8×0.8×1.2 mm3 and 0.65×0.65×0.65 mm3 at 7 T) and χ-separation algorithms (COSMOS9, MEDI, iLSQR, and χ-sepnet5).
Applications (1) Using 6 subject data for χ-separation-COSMOS9, quantitative metrics (RMSE, PSNR and SSIM) were calculated to evaluate the performance of χ-sepnet5 with, without and within the vessel mask. (2) Using 106 subjects, an χ-separation atlas was developed with and without the vessel mask. Twenty ROIs were analyzed to quantify the proportion of vessel and the population average.
Results:
Our method successfully generated vessel masks for χpara and χdia excluding non-vessel structures (yellow arrows in Fig. 2 (a)), when compared to the conventional methods. When tested for different resolutions, field strengths and χ-separation algorithms, the proposed method yielded robust outcomes (Fig. 2 (b), (c)).
The quantitative metrics computed with the vessel mask reported improved accuracy, suggesting that vessels are the source of variability. (RMSE: 0.0147±0.0015 ppm to 0.0115±0.0010 ppm; PSNR: 36.7260±0.9111 to 38.8034±0.7878; SSIM: 0.9258±0.0071 to 0.9269±0.0070) In the χ-separation atlas, caudate in χpara and genu in χdia revealed the highest vessel occupation (≥1 % voxels), demonstrating statistically significant decrease in the susceptibility values after applying the mask (caudate: 49.6±7.0 ppb to 47.4±6.9 ppb; genu: 30.3±2.4 ppb to 29.3±2.3 ppb; p<0.05/20).
The quantitative metrics computed with the vessel mask reported improved accuracy, suggesting that vessels are the source of variability. (RMSE: 0.0147±0.0015 ppm to 0.0115±0.0010 ppm; PSNR: 36.7260±0.9111 to 38.8034±0.7878; SSIM: 0.9258±0.0071 to 0.9269±0.0070) In the χ-separation atlas, caudate in χpara and genu in χdia revealed the highest vessel occupation (≥1 % voxels), demonstrating statistically significant decrease in the susceptibility values after applying the mask (caudate: 49.6±7.0 ppb to 47.4±6.9 ppb; genu: 30.3±2.4 ppb to 29.3±2.3 ppb; p<0.05/20).
Conclusions:
The proposed vessel segmentation method shows excellent performance in generating a vessel mask. When the mask is applied for the analysis, it improved the results by reducing variability from vessels.
Modeling and Analysis Methods:
Methods Development 1
Segmentation and Parcellation 2
Keywords:
Data analysis
Segmentation
Other - Quantitative Susceptibility imaging, Vessel
1|2Indicates the priority used for review
Provide references using author date format
2. Frangi, A. et al. (1998), Multiscale Vessel Enhancement Filtering. Medical Image Computing and Computer-Assisted Intervention — MICCAI’98, First International Conference Cambridge, MA, USA, October 11–13, 1998 Proceedings.
3. Jin, Z. et al. (2014), Background-Suppressed MR Venography of the Brain Using Magnitude Data: A High-Pass Filtering Approach. Computational and Mathematical Methods in Medicine, 812785.
4. Kerkeni, A. et al. (2016), A coronary artery segmentation method based on multiscale analysis and region growing. Computerized Medical Imaging and Graphics 48, 49–61.
5. Kim, M. et al. (2022), Chi-sepnet: Susceptibility source separation using deep neural network. Proceedings of International Society of Magnetic Resonance in Medicine 30, 2464.
6. Kim, W. et al. (2023), χ-Separation Imaging for Diagnosis of Multiple Sclerosis versus Neuromyelitis Optica Spectrum Disorder. Radiology 307, e220941.
7. Larson, T. C. et al. (1990), Spatial misregistration of vascular flow during MR imaging of the CNS: cause and clinical significance. American Journal of Neuroradiology 11, 1041–8.
8. Shin, H.-G. et al. (2021), χ-separation: Magnetic susceptibility source separation toward iron and myelin mapping in the brain. NeuroImage 240, 118371.
9. Shin, H.-G. et al. (2022), chi-separation using multi-orientation data in invivo and exvivo brains: Visualization of histology up to the resolution of 350μm. Proceedings of International Society of Magnetic Resonance in Medicine 30, 0771.
10. Straub, S. et al. (2022), A novel gradient echo data based vein segmentation algorithm and its application for the detection of regional cerebral differences in venous susceptibility. NeuroImage 250, 118931.