Evaluation of whole‐body MR to CT deformable image registration

2017

Medical Image Analysis

a b s t r a c tWe evaluate the accuracy of whole-body bone extraction from whole-body MR images using a number of atlas-based segmentation methods. The motivation behind this work is to find the most promising approach for the purpose of MRI-guided derivation of PET attenuation maps in whole-body PET/MRI. To this end, a variety of atlas-based segmentation strategies commonly used in medical image segmentation and pseudo-CT generation were implemented and evaluated in terms of whole-body bone segmentation accuracy. Bone segmentation was performed on 23 whole-body CT/MR image pairs via leave-one-out cross validation procedure. The evaluated segmentation techniques include: (i) intensity averaging (IA), (ii) majority voting (MV), (iii) global and (iv) local (voxel-wise) weighting atlas fusion frameworks implemented utilizing normalized mutual information (NMI), normalized cross-correlation (NCC) and mean square distance (MSD) as image similarity measures for calculating the weighting factors, along with other atlas-dependent algorithms, such as (v) shape-based averaging (SBA) and (vi) Hofmann's pseudo-CT generation method. The performance evaluation of the different segmentation techniques was carried out in terms of estimating bone extraction accuracy from whole-body MRI using standard metrics, such as Dice similarity (DSC) and relative volume difference (RVD) considering bony structures obtained from intensity thresholding of the reference CT images as the ground truth. Considering the Dice criterion, global weighting atlas fusion methods provided moderate improvement of whole-body bone segmentation (DSC = 0.65 ± 0.05) compared to non-weighted IA (DSC = 0.60 ± 0.02). The local weighed atlas fusion approach using the MSD similarity measure outperformed the other strategies by achieving a DSC of 0.81 ± 0.03 while using the NCC and NMI measures resulted in a DSC of 0.78 ± 0.05 and 0.75 ± 0.04, respectively. Despite very long computation time, the extracted bone obtained from both SBA (DSC = 0.56 ± 0.05) and Hofmann's methods (DSC = 0.60 ± 0.02) exhibited no improvement compared to non-weighted IA. Finding the optimum parameters for implementation of the atlas fusion approach, such as weighting factors and image similarity patch size, have great impact on the performance of atlas-based segmentation approaches. The voxel-wise atlas fusion approach exhibited excellent performance in terms of cancelling out the non-systematic registration errors leading to accurate and reliable segmentation results. Denoising and normalization of MR images together with optimization of the involved parameters play a key role in improving bone extraction accuracy.

Section: Pet/ct and Pet/mr Data Acquisitionmentioning

confidence: 99%

Section: Label Fusion Strategiesmentioning

confidence: 99%

Comparison of atlas-based techniques for whole-body bone segmentation

2017

Medical Image Analysis

“…The registration process was carried out through a combination of affine and nonrigid alignment based on the advanced Mattes mutual information 34 implemented in the Elastix package 35 as described in previous studies. 36,37 The following parameters were adopted: interpolate: B-spline, optimizer: standard gradient descent, image pyramid schedule: (16 8 4 2 2 in terms of Dice similarity coefficient. This procedure was repeated for the whole dataset and the optimal sizes of R and patch windows were chosen for the rest of the study (Fig.…”

Section: C Parameter Optimizationmentioning

confidence: 99%

Whole‐body bone segmentation from MRI for PET/MRI attenuation correction using shape‐based averaging

2016

Medical Physics

Purpose: The authors evaluate the performance of shape-based averaging (SBA) technique for whole-body bone segmentation from MRI in the context of MRI-guided attenuation correction (MRAC) in hybrid PET/MRI. To enhance the performance of the SBA scheme, the authors propose to combine it with statistical atlas fusion techniques. Moreover, a fast and efficient shape comparisonbased atlas selection scheme was developed and incorporated into the SBA method. Methods: Clinical studies consisting of PET/CT and MR images of 21 patients were used to assess the performance of the SBA method. In addition, the authors assessed the performance of simultaneous truth and performance level estimation (STAPLE) and the selective and iterative method for performance level estimation (SIMPLE) combined with SBA. In addition, a local shape comparison scheme (L-Shp) was proposed to improve the performance of SBA. The SIMPLE method was applied globally (G-SIMPLE) while STAPLE method was employed at both global (G-STAPLE) and local (L-STAPLE) levels. The evaluation was performed based on the accuracy of extracted whole-body bones, fragmentation, and computation time achieved by the different methods. The majority voting (MV) atlas fusion scheme was also evaluated as a conventional and commonly used method. MRI-guided attenuation maps were generated using the different segmentation methods. Thereafter, quantitative analysis of PET attenuation correction was performed using CT-based attenuation correction as reference. Results: The SBA and MV methods resulted in considerable underestimation of bone identification (Dice ≈ 0.62) and high factious fragmentation error of contiguous structures. Applying global atlas selection or regularization (G-STAPLE and G-SIMPLE) to the SBA method enhanced bone segmentation accuracy up to a Dice = 0.66. The best results were achieved when applying the L-STAPLE method with a Dice of 0.76 and the L-Shp method with a Dice of 0.75. However, L-STAPLE required up to five-fold increased computation time compared to the L-Shp method. Moreover, both L-STAPLE and L-Shp methods resulted in less than 3% SUV mean relative error and 6% SUV mean absolute error in bony structures owing to superior bone identification accuracy. The quantitative analysis using joint histograms revealed good correlation between PET-MRAC images using the proposed L-Shp algorithm and the corresponding reference PET-CT images. Conclusions: The performance of SBA was enhanced through application of local atlas weighting or regularization schemes (L-STAPLE and L-Shp). Bone recognition, fragmentation of the contiguous structures, and quantitative PET uptake recovery improved dramatically using these methods while the proposed L-Shp method significantly reduced the computation time. C 2016 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4963809]

“…Introducing a thorough concept that guarantees the accuracy of the registration procedure in non-rigid organs proved to be a difficult task ( Murphy et al, 2011 ). In this work, we relied on a registration procedure validated in a previous work by our group ( Akbarzadeh et al, 2013 ) using the elastix software ( Klein et al, 2009 ). The alignment was performed by combining rigid and non-rigid registration based on normalized mutual information criterion using B-spline interpolator with an adaptive stochastic gradient descent optimizer.…”

Section: Introductionmentioning

confidence: 99%

Magnetic resonance imaging-guided attenuation correction in whole-body PET/MRI using a sorted atlas approach

2016

Medical Image Analysis

a b s t r a c tQuantitative whole-body PET/MR imaging is challenged by the lack of accurate and robust strategies for attenuation correction. In this work, a new pseudo-CT generation approach, referred to as sorted atlas pseudo-CT (SAP), is proposed for accurate extraction of bones and estimation of lung attenuation properties. This approach improves the Gaussian process regression (GPR) kernel proposed by Hofmann et al. which relies on the information provided by a co-registered atlas (CT and MRI) using a GPR kernel to predict the distribution of attenuation coefficients. Our approach uses two separate GPR kernels for lung and non-lung tissues. For non-lung tissues, the co-registered atlas dataset was sorted on the basis of local normalized cross-correlation similarity to the target MR image to select the most similar image in the atlas for each voxel. For lung tissue, the lung volume was incorporated in the GPR kernel taking advantage of the correlation between lung volume and corresponding attenuation properties to predict the attenuation coefficients of the lung. In the presence of pathological tissues in the lungs, the lesions are segmented on PET images corrected for attenuation using MRI-derived three-class attenuation map followed by assignment of soft-tissue attenuation coefficient. The proposed algorithm was compared to other techniques reported in the literature including Hofmann's approach and the three-class attenuation correction technique implemented on the Philips Ingenuity TF PET/MR where CT-based attenuation correction served as reference. Fourteen patients with head and neck cancer undergoing PET/CT and PET/MR examinations were used for quantitative analysis. SUV measurements were performed on 12 normal uptake regions as well as high uptake malignant regions. Moreover, a number of similarity measures were used to evaluate the accuracy of extracted bones. The Dice similarity metric revealed that the extracted bone improved from 0.58 ± 0.09 to 0.65 ± 0.07 when using the SAP technique compared to Hofmann's approach. This enabled to reduce the SUV mean bias in bony structures for the SAP approach to -1.7 ± 4.8% as compared to -7.3 ± 6.0% and -27.4 ± 10.1% when using Hofmann's approach and the three-class attenuation map, respectively. Likewise, the three-class attenuation map produces a relative absolute error of 21.7 ± 11.8% in the lungs. This was reduced on average to 15.8 ± 8.6% and 8.0 ± 3.8% when using Hofmann's and SAP techniques, respectively. The SAP technique resulted in better overall PET quantification accuracy than both Hofmann's and the three-class approaches owing to the more accurate extraction of bones and better prediction of lung attenuation coefficients. Further improvement of the technique and reduction of the computational time are still required.