An improved level set method for brain MR images segmentation and bias correction

Intensity inhomogeneities cause considerable difficulty in the quantitative analysis of magnetic resonance (MR) images. Thus, bias field estimation is a necessary step before quantitative analysis of MR data can be undertaken. This paper presents a variational level set approach to bias correction and segmentation for images with intensity inhomogeneities. Our method is based on an observation that intensities in a relatively small local region are separable, despite of the inseparability of the intensities in the whole image caused by the overall intensity inhomogeneity. We first define a localized K-means-type clustering objective function for image intensities in a neighborhood around each point. The cluster centers in this objective function have a multiplicative factor that estimates the bias within the neighborhood. The objective function is then integrated over the entire domain to define the data term into the level set framework. Our method is able to capture bias of quite general profiles. Moreover, it is robust to initialization, and thereby allows fully automated applications. The proposed method has been used for images of various modalities with promising results.     

Rapid estimation of split renal function in kidney donors using software developed for computed tomographic renal volumetry

Purpose

To evaluate the speed and precision of split renal volume (SRV) measurement, which is the ratio of unilateral renal volume to bilateral renal volume, using a newly developed software for computed tomographic (CT) volumetry and to investigate the usefulness of SRV for the estimation of split renal function (SRF) in kidney donors.

Method

Both dynamic CT and renal scintigraphy in 28 adult potential living renal donors were the subjects of this study. We calculated SRV using the newly developed volumetric software built into a PACS viewer (n-SRV), and compared it with SRV calculated using a conventional workstation, ZIOSOFT (z-SRV). The correlation with split renal function (SRF) using ^99mTc-DMSA scintigraphy was also investigated.

Results

The time required for volumetry of bilateral kidneys with the newly developed software (16.7 ± 3.9 s) was significantly shorter than that of the workstation (102.6 ± 38.9 s, p < 0.0001). The results of n-SRV (49.7 ± 4.0%) were highly consistent with those of z-SRV (49.9 ± 3.6%), with a mean discrepancy of 0.12 ± 0.84%. The SRF also agreed well with the n-SRV, with a mean discrepancy of 0.25 ± 1.65%. The dominant side determined by SRF and n-SRV showed agreement in 26 of 28 cases (92.9%).

Conclusion

The newly developed software for CT volumetry was more rapid than the conventional workstation volumetry and just as accurate, and was suggested to be useful for the estimation of SRF and thus the dominant side in kidney donors.     

Skull-stripping method for brain MRI using a 3D level set with a speedup operator

Purpose:

To extract the brain region from brain magnetic resonance (MR) images using a fast 3D level set method and a refinement process.

Materials and Methods:

The proposed method introduces a speedup operator to the conventional 3D level set method in order to accelerate the level set evolution. While the processing time for brain extraction is reduced by the speedup operator, the accuracy of brain extraction is also improved by adopting a refinement process.

Results:

The speedup operator yielded a 75% reduction in the total iteration numbers for the synthesized volume. The proposed method was applied to several datasets and compared with previous methods, ie, BrainVisa, BET, and FreeSurfer. The proposed method provided a Jaccard index of 0.971 ± 0.0114 for the BrainWeb dataset, 0.864 ± 0.035 for the IBSR dataset, and 0.9414 ± 0.0517 for a self‐produced dataset acquired with a 3T MRI system.

Conclusion:

Utilizing a speedup operator, the proposed method reduced the evolution time. Robust and accurate results for various datasets were obtained in experiments. J. Magn. Reson. Imaging 2011;. © 2011 Wiley‐Liss, Inc.     

Assessment of pulmonary vasculature volume with automated threshold-based 3D quantitative CT volumetry: in vitro and in vivo validation

Objectives

To validate the ability of threshold-based 3D CT volumetry to enable measurement of volume of visible pulmonary vessels on CT.

Materials and methods

In vivo, 3D CT volumetry was validated in seven phantoms that consisted of silicone tubes embedded in a foam block. With the true volume value as reference standard, the accuracy of CT measurement at various lower thresholds of −600 HU, −500 HU, −300 HU and −200 HU were compared. The volume measurements obtained when filled with varied concentration of iodinated contrast media (1:100, 1:200 and 1:500) were also compared. In vivo validation was performed in sixteen patients (9 men, 7 women; mean age, 52.1 years). Inter-scan and inter-observer agreement and reproducibility for pulmonary vasculature volume measurement were evaluated with Bland–Altman analysis.

Results

In vitro, the mean value measured under lower threshold of −300 HU (relative error = 1.5%) were the closest to the true values and have no signi?cant difference (P = 0.375). There were no signi?cant differences among the phantom measurement values with different filled concentration (1:100, 1:200 and 1:500). In vivo, the inter-scan reproducibility of volume measurements was good, with a correlation coefficient of 0.82 and ICC (intraclass correlation coefficient) of 0.86. Inter-observer agreement was excellent with a correlation coefficient of 0.91 and ICC of 0.95.

Conclusions

The threshold-based 3D quantitative CT volumetry enables accurate and reproducible measurement of pulmonary vessels volume.     

Segmentation of cardiac cine-MR images and myocardial deformation assessment using level set methods

In this paper, we present an original method to assess the deformations of the left ventricular myocardium on cardiac cine-MRI. First, a segmentation process, based on a level set method is directly applied on a 2D+t dataset to detect endocardial contours. Second, the successive segmented contours are matched using a procedure of global alignment, followed by a morphing process based on a level set approach. Finally, local measurements of myocardial deformations are derived from the previously determined matched contours. The validation step is realized by comparing our results to the measurements achieved on the same patients by an expert using the semi-automated HARP reference method on tagged MR images.     

A comparison between the accuracy of voxel-based morphometry and hippocampal volumetry in Alzheimer's disease

PURPOSE: To compare the accuracy of voxel-based morphometry (VBM) and region of interest (ROI)-based hippocampal volumetry to detect medial temporal lobe atrophy in Alzheimer's disease (AD). MATERIALS AND METHODS: A total of 27 AD patients (age 74 +/- 9 years; 22 women; Mini-Mental State Exam [MMSE] 21 +/- 4) and 25 controls (age 70 +/- 8; 16 women; MMSE 29 +/- 1) were studied. Accuracy of VBM to detect gray matter loss in those seven AD patients and 11 controls with similar ROI-based hippocampal measures and of ROI-based volumetry to detect gray matter loss in those four AD patients and five controls with similar VBM-based hippocampal measures was assessed. VBM was performed with statistical parametric mapping (SPM99). RESULTS: The area under the curve was 0.96 (95% C.I., 0.92-1.00) for VBM, 0.89 (95% C.I., 0.80-0.98) for ROI-based hippocampal measures, and 0.99 (95% C.I., 0.96-1.00) for both. In subjects with similar ROI-based hippocampal measures, VBM detected atrophy in AD patients at P < 0.0001, while in subjects with similar VBM-based hippocampal measure, volumetry was not significant (P = 0.11). Both measures independently contributed to discrimination (P = 0.004 and P = 0.032) in a logistic regression model. CONCLUSION: These results indicate that VBM is more accurate, but the combination of both methods provides the highest accuracy for detection of hippocampal atrophy in AD.     

Level set method with automatic selective local statistics for brain tumor segmentation in MR images

The level set approach is a powerful tool for segmenting images. This paper proposes a method for segmenting brain tumor images from MR images. A new signed pressure function (SPF) that can efficiently stop the contours at weak or blurred edges is introduced. The local statistics of the different objects present in the MR images were calculated. Using local statistics, the tumor objects were identified among different objects. In this level set method, the calculation of the parameters is a challenging task. The calculations of different parameters for different types of images were automatic. The basic thresholding value was updated and adjusted automatically for different MR images. This thresholding value was used to calculate the different parameters in the proposed algorithm. The proposed algorithm was tested on the magnetic resonance images of the brain for tumor segmentation and its performance was evaluated visually and quantitatively. Numerical experiments on some brain tumor images highlighted the efficiency and robustness of this method.     

Automatic arm removal in PET and CT images for deformable registration

Positron emission tomography (PET) imaging is rapidly expanding its role in clinical practice for cancer management. The high sensitivity of PET for functional abnormalities associated with cancer can be confounded by the minimal anatomical information it provides for cancer localization. Computed tomography (CT) provides detailed anatomical information but is less sensitive to pathologies than PET. Thus, combining (i.e., registering) PET and CT images would enable both accurate and sensitive cancer localization with respect to detailed patient anatomy. An additional application area of registration is to align CT–CT scans from serial studies on a patient on a PET/CT scanner to facilitate accurate assessment of therapeutic response from the co-aligned PET images. To facilitate image fusion, we are developing a deformable registration software system using mutual information and a B-spline model of the deformation. When applying deformable registration to whole body images, one of the obstacles is that the arms are present in PET images but not in CT images or are in different positions in serial CT images. This feature mismatch requires a preprocessing step to remove the arms where present and thus adds a manual step in an otherwise automatic algorithm. In this paper, we present a simple yet effective method for automatic arm removal. We demonstrate the efficiency and robustness of this algorithm on both clinical PET and CT images. By streamlining the entire registration process, we expect that the fusion technology will soon find its way into clinics, greatly benefiting cancer diagnosis, staging, therapy planning and treatment monitoring.     

Automated segmentation and analysis of vascular structures in magnetic resonance angiographic images.

The accurate assessment of the presence and extent of vascular disease, and planning of vascular interventions based on MRA requires the determination of vessel dimensions. The current standard is based on measuring vessel diameters on maximum intensity projections (MIPs) using calipers. In order to increase the accuracy and reproducibility of the method, automated analysis of the 3D MR data is required. A novel method for automatically determining the trajectory of the vessel of interest, the luminal boundaries, and subsequent the vessel dimensions is presented. The automated segmentation in 3D uses deformable models, combined with knowledge of the acquisition protocol. The trajectory determination was tested on 20 in vivo studies of the abdomen and legs. In 93% the detected trajectory followed the vessel. The luminal boundary detection was validated on contrast-enhanced (CE) MRA images of five stenotic phantoms. The results from the automated analysis correlated very well with the true diameters of the phantoms used in the in vitro study (r = 0.999, P < 0.001). MRA and x-ray angiography (XA) of the phantoms also correlated well (r = 0.895, P < 0.001). The average unsigned difference between the MRA and XA measurements was 0.08 +/- 0.05 mm. In conclusion, the automated approach allows the accurate assessment of vessel dimensions in MRA images.     

Influence of radiation dose and iterative reconstruction algorithms for measurement accuracy and reproducibility of pulmonary nodule volumetry: A phantom study

Purpose

To evaluate the influence of radiation dose settings and reconstruction algorithms on the measurement accuracy and reproducibility of semi-automated pulmonary nodule volumetry.

Materials and methods

CT scans were performed on a chest phantom containing various nodules (10 and 12 mm; +100, −630 and −800 HU) at 120 kVp with tube current–time settings of 10, 20, 50, and 100 mAs. Each CT was reconstructed using filtered back projection (FBP), iDose⁴ and iterative model reconstruction (IMR). Semi-automated volumetry was performed by two radiologists using commercial volumetry software for nodules at each CT dataset. Noise, contrast-to-noise ratio and signal-to-noise ratio of CT images were also obtained. The absolute percentage measurement errors and differences were then calculated for volume and mass. The influence of radiation dose and reconstruction algorithm on measurement accuracy, reproducibility and objective image quality metrics was analyzed using generalized estimating equations.

Results

Measurement accuracy and reproducibility of nodule volume and mass were not significantly associated with CT radiation dose settings or reconstruction algorithms (p > 0.05). Objective image quality metrics of CT images were superior in IMR than in FBP or iDose⁴ at all radiation dose settings (p < 0.05).

Conclusion

Semi-automated nodule volumetry can be applied to low- or ultralow-dose chest CT with usage of a novel iterative reconstruction algorithm without losing measurement accuracy and reproducibility.     

Subvoxel processing: a method for reducing partial volume blurring with application to in vivo MR images of trabecular bone.

Partial volume blurring precludes accurate measurement of structural dimensions in the limited-resolution regime in which image voxel size is larger than the typical structural element to be resolved. Since acquiring images at increased resolution often exacts an unacceptable signal-to-noise ratio (SNR) penalty, methods to alleviate the adverse effects of partial volume blurring are instrumental for the accurate measurement of architectural parameters in applications such as predicting the mechanical competence of trabecular bone networks. In the current work, a novel post-processing method, referred to as "subvoxel processing," is described for increasing apparent image resolution. The method is applicable to volumes of interest containing material phases of two discrete signal intensities. The principal strategy consists of subdividing voxels and assigning voxel intensities to each subvoxel on the basis of local neighborhood criteria and strict mass conservation. In the current work, the method's accuracy has been evaluated using microcomputed tomography images (22 x 22 x 22 microm(3) voxel size) of human trabecular bone. The results demonstrate that subvoxel processing is significantly more accurate than trilinear interpolation in decreasing apparent voxel size, especially in the presence of noise. In addition, the method's effectiveness is illustrated with MR images of human trabecular bone acquired in vivo at 137 x 137 x 350 microm(3) voxel size. The subvoxel-processed images are shown to have architectural features characteristic of images acquired at higher spatial resolution.     

Accuracy and reliability of MRI vs. laboratory measurements in an ex vivo porcine model of arthritic cartilage loss

PURPOSE: To quantify the accuracy of magnetic resonance imaging (MRI) measurement of change in cartilage volume due to thin linear excisions, simulating arthritic cartilage losses, by comparison with laboratory volume measurements in an ex vivo porcine model. MATERIALS AND METHODS: We scanned 15 porcine patellae by T1-weighted spoiled gradient echo (SPGR) MRI at baseline and after excision of up to three thin layers of articular cartilage. Excised fragment volume was determined from density and weight. Postexcision scans were "fused" to the baseline scan by three-dimensional (3D) registration. This allowed automated recalculation of the remaining cartilage volume within a baseline region of interest (ROI) following each excision. We compared MRI estimates of change in cartilage volume to direct laboratory measurement of fragment volume. RESULTS: Our 38 excised fragments averaged 0.16 mL, or approximately 7% of cartilage volume. MRI and laboratory estimates of total cartilage volume loss differed by 1.6% +/- 13.2% (mean, coefficient of variation [CV]). Accuracy was +/-0.1 mL for 95% of scans. CONCLUSION: MRI estimates of small changes in porcine patellar cartilage volume were unbiased, reliable, and accurate to 0.1 mL. Despite a proportionately high error in the very thin fragments tested, achievement of similar accuracy in vivo would be adequate to detect approximately two years of osteoarthritic cartilage loss.     

Diagnostic accuracy and tolerability of contrast enhanced CT colonoscopy in symptomatic patients with increased risk for colorectal cancer

Objective

We compared the accuracy and tolerability of intravenous contrast enhanced spiral computed tomography colonography (CTC) and optical colonoscopy (OC) for the detection of colorectal neoplasia in symptomatic patients for colorectal neoplasia.

Methods

A prospective study was performed in 48 patients with symptomatic patients with increased risk for colorectal cancer. Spiral CTC was performed in supine and prone positions after colonic cleansing. The axial, 2D MPR and virtual endoluminal views were analyzed. Results of spiral CTC were compared with OC which was done within 15 days. The psychometric tolerance test was asked to be performed for both CTC and colonoscopy after the procedure.

Results

Ten lesions in 9 of 48 patients were found in CTC and confirmed with OC. Two masses and eight polyps, consisted of 1 tubulovillous, 1 tubular, 2 villous adenoma, 4 adenomatous polyp, 4 adenocarcinoma, were identified. Lesion prevalence was 21%. Sensitivity, specificity, accuracy, positive and negative predictive values were found 100%, 87%, 89%, 67% and 100%, respectively. Psychometric tolerance test showed that CTC significantly more comfortable comparing with OC (p = 0.00). CTC was the preferred method in 37% while OC was preferred in 6% of patients. In both techniques, the most unpleasant part was bowel cleansing.

Conclusion

Contrast enhanced CTC is a highly accurate method in detecting colorectal lesions. Since the technique was found to be more comfortable and less time consuming compare to OE, it may be preferable in management of symptomatic patients with increased risk for colorectal cancer.     

MR histological correlation: a method for cutting specimens along the imaging plane in animal or ex vivo experiments

PURPOSE: To assess a method aimed at cutting histological specimens along the magnetic resonance (MR) imaging plane. MATERIAL AND METHODS: The method is performed in two steps: the imaging plane (defined by three acrylic paint markers) is made horizontal under MR guidance by using a mobile platform that can be rotated in three directions (PlaneFinder device [PFD]); then, the specimen is embedded in wax and cut horizontally. Three-dimensional images parallel to the markers' plane were obtained on 31 pork muscles containing a central hole with a pyramidal shape, with a technique of reference (RT images) and with PFD (PF images), before and after fixation. The last 17 fixed specimens were cut in the markers' plane (tissue section [TS] images). The central hole area (CHA) in the markers' plane was used to compare RT, PF, and TS images. Using a workstation, PF images were rotated and translated to estimate the shift along each direction that could explain the entire CHA difference between RT, PF, and TS images (maximum error, worst-case scenario). RESULTS: Excellent correlation was found between RT and PF images (r = 0.989, slope = 1.0175), PF and TS images (r = 0.991, slope = 1.0058), and RT images on fresh specimens and TS images (r = 0.979, slope = 1.0732). For each step, the maximum angle error was < or = 3 degrees in 88-95% of the specimens. CONCLUSION: Our methodology can be used to cut specimens along the imaging plane with high accuracy.     

Comparison between CT volume measurement and histopathological assessment of response to neoadjuvant therapy in rectal cancer

Objectives

The aim of this study was to compare volume measurements on computed tomography (CT) images with histopathological assessments of chemoradiotherapy (CRT)-induced tumor regression in locally advanced rectal cancer (RC).

Methods

In 25 patients (13 males, 12 females; median age, 63 years; age range, 44–79 years) with locally advanced RC treated with preoperative CRT and surgery, two radiologists measured tumor volume on CT images before and after CRT. CT-based tumor volumetry and the modified response evaluation criteria in solid tumors (mRECISTs) were compared with T and N downstaging after CRT, and with the tumor regression grade (TRG).

Results

Tumor volumes were significantly smaller on CT images after CRT. The tumors regressed in 52% (13/25), 36% (9/25) and 40% (10/25) of patients, based on T downstaging, TRG and mRECIST findings, respectively. In terms of T downstaging, the pre- and post-CRT tumor volumes of responders and non-responders to the treatment differed statistically, while their tumor volume reduction rates and volume reductions according to the 65% mRECIST threshold did not. In terms of N downstaging and TRG, the differences between the responders’ and the non-responders’ pre- and post-CRT tumor volumes, tumor volume reduction rates, and mRECIST thresholds were never statistically significant.

Conclusion

Measuring tumor size on CT images is of limited value in predicting the histopathological response to preoperative CRT in RC patients, so it may be unwise to select surgical treatment strategies based on CT volumetry.     

Assessment of synovitis in the osteoarthritic knee: Comparison between manual segmentation,semiautomated segmentation,and semiquantitative assessment using contrast‐enhanced fat‐suppressed T1‐weighted MRI

Osteoarthritic joints regularly exhibit synovitis, which is ideally assessed on contrast‐enhanced MRI. Manual segmentation is the reference standard for volumetric analysis but is labor intensive. The aim was to evaluate alternative semiautomated approaches of targeted thresholding and gaussian deconvolution. Volumetric and semiquantitative synovitis assessment was compared in addition. Thirty‐two knees with osteoarthritis were scanned on a 1.5‐T system. Synovitis volumes were plotted against each other and distributions fit with linear functions. The relationship between semiquantitative scores and synovitis volumes was assessed using Spearman's correlation coefficient. Semiautomated volume measurement was more time efficient than manual segmentation and showed a high correlation with manual analysis (R² = 0.88 and 0.82). Manual segmentation was correlated with summed and with maximum semiquantitative synovitis scores (ρ = 0.71 and 0.47). In conclusion, semiautomated analysis provides comparable quantitative results when compared to manual segmentation but is approximately five times more time efficient. Semiquantitative assessment adds anatomic information on synovitis distribution. Magn Reson Med, 2010. © 2010 Wiley‐Liss, Inc.     

Automated quantification of muscle and fat in the thigh from water-, fat-, and nonsuppressed MR images

Purpose:

To introduce and validate an unsupervised muscle and fat quantification algorithm based on joint analysis of water‐suppressed (WS), fat‐suppressed (FS), and water and fat (nonsuppressed) volumetric magnetic resonance imaging (MRI) of the mid‐thigh region.

Materials and Methods:

We first segmented the subcutaneous fat by use of a parametric deformable model, then applied centroid clustering in the feature domain defined by the voxel intensities in WS and FS images to identify the intermuscular fat and muscle. In the final step we computed volumetric and area measures of fat and muscle. We applied this algorithm on datasets of water‐, fat‐, and nonsuppressed volumetric MR images acquired from 28 participants.

Results:

We validated our tissue composition analysis against fat and muscle area measurements obtained from semimanual analysis of single‐slice mid‐thigh computed tomography (CT) images of the same participants and found very good agreement between the two methods. Furthermore, we compared the proposed approach with a variant that uses nonsuppressed images only and observed that joint analysis of WS and FS images is more accurate than the nonsuppressed only variant.

Conclusion:

Our MRI algorithm produces accurate tissue quantification, is less labor‐intensive, and more reproducible than the original CT‐based workflow and can address interparticipant anatomic variability and intensity inhomogeneity effects. J. Magn. Reson. Imaging 2012;35:1152‐1161. © 2011 Wiley Periodicals, Inc.     

Three-dimensional method for comparing in vivo interventional MR images of thermally ablated tissue with tissue response

PURPOSE: To investigate the ability of magnetic resonance (MR) to monitor radio-frequency (RF) ablation treatments by comparing MR images of thermal lesions to histologically assayed cellular damage. We developed a new methodology using three-dimensional registration for making spatial correlations. MATERIALS AND METHODS: A low-field, open MRI system was used to guide an ablation probe into rabbit thigh muscle and acquire MR volumes after ablation. After fixation, we sliced and photographed the tissue at 3-mm intervals, using a specially designed apparatus, to obtain a volume of tissue images. Histologic samples were digitized using a video microscopy system. For our three-dimensional registration method, we used the tissue images as the reference, and registered histology and MR images to them using two different computer alignment steps. First, the MR volume was aligned to the volume of tissue images by registering needle fiducials placed near the tissue of interest. Second, we registered the histology images with the tissue images using a two-dimensional warping technique that aligned internal features and the outside boundary of histology and tissue images. RESULTS: The MR and histology images were very well aligned, and registration accuracy, determined from displacement of needle fiducials, was 1.32 +/- 0.39 mm (mean +/- SD), which compared favorably to the MR voxel dimensions (0.70 mm in-plane and 3.0 mm thick). A preliminary comparison of MR and tissue response showed that the region inside the elliptical hyperintense rim in MR closely corresponds to the region of necrosis as established by histology, with a mean absolute distance between MR and histology boundaries of 1.17 mm, slightly smaller than the mean registration error. The MR region slightly overestimated the region of necrosis, with a mean signed distance between boundaries of 0.85 mm. CONCLUSION: Our results suggest that our methodology can be used to achieve three-dimensional registration of histology and in vivo MR images. In MR lesion images, the inner border of the hyperintense region corresponds to the border of irreversible cell damage. This is good evidence that during RF ablation treatments, iMRI lesion images can be used for real-time feedback.