Automated method for accurate abdominal fat quantification on water-saturated magnetic resonance images

PURPOSE: To introduce and evaluate the performance of an automated fat quantification method for water-saturated magnetic resonance images. MATERIALS AND METHODS: A fat distribution model is proposed for fat quantification on water saturated magnetic resonance images. Fat from both full- and partial-volume voxels are accounted for in this model based on image intensity histogram analysis. An automated threshold method is therefore proposed to accurately quantify total fat. This method is compared to a traditional full-volume-fat-only method in phantom and human studies. In the phantom study, fat quantification was performed on MR images obtained from a human abdomen oil phantom and was compared with the true oil volumes. In the human study, results of the two fat quantification methods of six subjects were compared on abdominal images with different spatial resolutions. RESULTS: In the phantom study, the proposed method provided significantly more accurate estimations of true oil volumes compared to the reference method (P < 0.0001). In human studies, fat quantification using the proposed method gave much more consistent results on images with different spatial resolutions, and on regions with different degrees of partial volume averaging. CONCLUSION: The proposed automated method is simple, rapid, and accurate for fat quantification on water-saturated MR images.     

Automated quality control of brain MR images

PURPOSE: To present a novel fully automated method for assessing the quality of magnetic resonance imaging (MRI) data acquired in a clinical trials environment. MATERIALS AND METHODS: This work was performed in the context of clinical trials for multiple sclerosis. Quality control (QC) procedures included were: (i) patient brain identity verification, (ii) alphanumeric parameter matching, (iii) signal-to-noise ratio estimation, (iv) gadolinium-enhancement verification, and (v) detection of ghosting due to head motion. Each QC procedure produces a quantitative measurement which is compared against an acceptance threshold that was determined based on receiver operating characteristic analysis of traditional manual and visual QC performed by trained experts. RESULTS: The automated QC results have high sensitivity and specificity when compared with the visual QC. CONCLUSION: Our automated objective QC procedure can replace many manual subjective procedures to provide increased data throughput while reducing reader variability.     

Simultaneous quantification of fat content and fatty acid composition using MR imaging

Not only the fat content but also the composition of fatty acids (FAs) in stored triglycerides might be of interest in the research on nonalcoholic fatty liver disease and nonalcoholic steatohepatitis. In this study, a novel reconstruction approach is proposed that uses theoretical knowledge of the chemical structure of FAs to simultaneously quantify the fat fraction (FF) and the FAs composition (chain length cl, number of double bonds ndb, and number of methylene‐interrupted double bonds nmidb) from multiple gradient echo images. Twenty phantoms with various fat contents (FF = 9–100%) and FA compositions (cl = 12.1–17.9, ndb = 0.23–5.10, and nmidb = 0.04–2.39) were constructed and imaged in a 3‐T Siemens scanner. In addition, spectra were acquired in each phantom. Slopes and “standard deviations from true values” were used to investigate the accuracy of the two methods. The imaging method holds well in a comparison to the previously suggested spectroscopy method and showed similar overall accuracy. The in vivo feasibility was demonstrated in the thigh adipose tissue of a healthy volunteer. In conclusion, our developed method is a promising tool for FF and FA composition quantification. Magn Reson Med, 2013. © 2012 Wiley Periodicals, Inc.     

Automatic vessel wall contour detection and quantification of wall thickness in in-vivo MR images of the human aorta

PURPOSE: To develop an automated technique to trace the contours of the lumen and outer boundary of the aortic wall, and measure aortic wall thickness in axial MR images. MATERIALS AND METHODS: The algorithm uses prior knowledge of vessel wall morphology. A geometrical model (ellipse) is deformed, translated and rotated to obtain a rough approximation of the contours. Model-matching is based on image gradient measurements. To enhance edges, the images were preprocessed using gray-level stretching. Refinement is performed by means of dynamic programming. Wall thickness is computed by measuring the distance between inner and outer contour of the aortic wall. RESULTS: The algorithm has been tested on high-resolution axial MR images from 28 human subjects of the descending thoracic aorta. The results demonstrate: High correspondence between automatic and manual area measurements: lumen (r = 0.99), outer (r = 0.96), and wall thickness (r = 0.85). CONCLUSION: Though further optimization is required, our algorithm is a powerful tool to automatically draw the boundaries of the aortic wall and measure aortic wall thickness in aortic wall devoid of major lesions. J. Magn. Reson. Imaging 2006. (c) 2006 Wiley-Liss, Inc.     

Automated registration of dynamic MR images for the quantification of myocardial perfusion

Cardiac dynamic magnetic resonance imaging (MRI) after contrast media injection suffers from motion induced by free breathing during acquisition. This work presents an automated approach for motion correction of the heart. The registration is based on the multipass/multiresolution iterative minimizing of intrinsic differences between each image and a reference image coupled to a two-dimensional/3 parameters rigid body correction. The efficiency of this correction method was evaluated with anatomical landmarks, various cost functions, and for a compartment model fit of the data with 2 parameters: K1, the blood to myocardium transfer coefficient; and Vd, the distribution volume of the contrast media. The variability of K1 and Vd, derived from the fit of the registered images (using the manual correction as a gold standard), was significantly reduced by comparison with the variability obtained from the uncorrected images (P < 0.04). This motion correction method also clearly improves the analysis of dynamic cardiac MRI after contrast media injection in comparison to manual correction.     

Comparison of visceral adipose tissue quantification on water suppressed and nonwater-suppressed MRI at 3.0 Tesla

Purpose:

To systematically evaluate and compare the performance of water‐saturated and nonwater‐saturated T1‐weighted 3.0 T magnetic resonance imaging (MRI) in the application of visceral adipose tissue (VAT) quantification.

Materials and Methods:

Forty‐five patients underwent abdomen MRI using two different sequences at 3.0 T: 1) a traditional T1‐weighted gradient echo sequence, and 2) the same sequence with water presaturation to enhance fat and nonfat contrast. VAT amounts from both water‐saturated and nonwater‐saturated images were quantified with a manual thresholding technique and an automated segmentation method to study quantification variability and consistency of the two imaging techniques.

Results:

Nonwater‐saturated MRI had significantly larger coefficient of variation than water‐saturated MRI in the imaging reproducibility study based on 112 slices from seven subjects (11.4% vs. 2.5%, P < 0.0001). VAT volumes measured from the nonwater‐saturation MRI sequence had significantly higher variability than those from water‐saturation images even when using a manual quantification method based on images from 38 subjects (1.76% vs. 1.08%, P < 0.001). In addition, the VAT volume amounts from nonwater‐saturation images and water‐saturated images quantified with the automatic and manual quantification methods were statistically consistent.

Conclusion:

Water‐saturated MRI sequences at 3.0 T for VAT quantification improve reproducibility and decrease variability compared with nonwater saturated sequences, especially with the use of automatic quantification methods. J. Magn. Reson. Imaging 2012;35:1445–1452. © 2012 Wiley Periodicals, Inc.     

Automatic quantification of subcutaneous and visceral adipose tissue from whole‐body magnetic resonance images suitable for large cohort studies

Purpose:

To develop an automated method with which to distinguish metabolically different adipose tissues in a large number of subjects using whole‐body magnetic resonance imaging (MRI) datasets for improving the understanding of chronic disease risk predictions associated with distinct adipose tissue compartments.

Materials and Methods:

In all, 314 participants were scanned using a 1.5T MRI‐scanner with a 2‐point Dixon whole‐body sequence. Image segmentation was automated using standard image processing techniques and knowledge‐based methods. Abdominal adipose tissue was separated into subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT) by statistical shape models. Bone marrow was removed to provide a more accurate measurement of adipose tissue. To assess segmentation accuracy, ground‐truth segmentations in 52 images were performed manually by one operator. Due to the high effort of manual delineation, manual segmentation was limited to seven slices per volume.

Results:

Volumetric differences were 3.30 ± 2.97% and 6.22 ± 5.28% for SAT and VAT, respectively. The systematic error shows an overestimation of 4.22 ± 7.01% for VAT and 0.37 ± 4.45% for SAT. Coefficients‐of‐variation from repeated measurements were: 3.50 ± 2.93% for VAT and 0.35 ± 0.26% for SAT. The approach of removing bone marrow worked well in most body regions. Only occasionally the method failed for knees and/or shinbone, which resulted in an overestimation of SAT by 3.14 ± 1.45%.

Conclusion:

We developed a fully automatic process to assess SAT and VAT in whole‐body MRI data. The method can support epidemiological studies investigating the relationship between excess body fat and chronic diseases. J. Magn. Reson. Imaging 2012; 36:1421–1434. © 2012 Wiley Periodicals, Inc.     

Hemodynamic segmentation of MR brain perfusion images using independent component analysis, thresholding, and Bayesian estimation.

Dynamic-susceptibility-contrast MR perfusion imaging is a widely used imaging tool for in vivo study of cerebral blood perfusion. However, visualization of different hemodynamic compartments is less investigated. In this work, independent component analysis, thresholding, and Bayesian estimation were used to concurrently segment different tissues, i.e., artery, gray matter, white matter, vein and sinus, choroid plexus, and cerebral spinal fluid, with corresponding signal-time curves on perfusion images of five normal volunteers. Based on the spatiotemporal hemodynamics, sequential passages and microcirculation of contrast-agent particles in these tissues were decomposed and analyzed. Late and multiphasic perfusion, indicating the presence of contrast agents, was observed in the choroid plexus and the cerebral spinal fluid. An arterial input function was modeled using the concentration-time curve of the arterial area on the same slice, rather than remote slices, for the deconvolution calculation of relative cerebral blood flow.     

Computer assessment of neurodegeneration in Parkinson disease using data fusion techniques with MR images

RATIONALE AND OBJECTIVES: Recently developed MR imaging techniques using inversion recovery are a sensitive tool to identify and quantify morphologic changes in the substantia nigra due to neurodegeneration. Using a semi-automated computer segmentation technique to isolate the substantia nigra pars compacta (SN(c)), we propose a colored image fusion technique to visually assess the sites of damage in the SN(c) and integrate the information obtained from two implemented inversion-recovery sequences. PATIENTS AND METHODS: Six patients and six age-matched control subjects were scanned using a combination of two MR imaging inversion-recovery (IR) pulse sequences. A subgroup of them was used to develop our technique. Images were blended together into a final (RGBA) image, where A stands for the alpha channel describing transparency. RESULTS: Abnormalities in the SN(c) can be accurately assessed in location, shape, and variations of signal intensities within the segmented SN(c) by varying the transparency (alpha) channel of the color fusion image. Several previous findings such as the lateral-medial gradient of signal change and a ventral-dorsal broadening of the pars compacta are accompanied by an overall mild-to-severe heterogeneity of neurodegeneration patterns. CONCLUSION: Color fusion techniques revealed subtle changes in the neurodegeneration of the substantia nigra in Parkinson disease, which can be helpful for an objective and hence effective visual assessment of disease progression.     

Non-invasive quantification of hepatic fat fraction by fast 1.0, 1.5 and 3.0 T MR imaging

INTRODUCTION: Even mild hepatic steatosis in a split liver donor may cause general liver failure and death in the donor. So far, CT density measurements or percutaneous biopsy is used to determine the presence of hepatic steatosis. Magnetic resonance imaging (MRI) may be an elegant method of non-invasive and non-radiation quantification of hepatic fat content. METHODS: Fast gradient echo (GRE) technique was used to discriminate between fat and water spins. Echo time (TE) was adjusted for field strength dependent in-phase and out-of-phase states at 1.0, 1.5 and 3.0 T. Continuous MR signal transition from 100% water to 100% fat was investigated using a wedge water-oil phantom, which was positioned in such a way, that no spatial resolution occurred, thereby combining water and fat in one slice. RESULTS: Using the phantom, a significant difference for a 5% difference in fat content was demonstrated in the range from 20 to 80% fat content (p<0.05) for all tested field strengths. In 25 patients MRI data were correlated with the percentage of fat determined by histologic evaluation of a CT-guided liver biopsy. Using the linear correlation calculated from the MRI phantom data at 1.0 T, we determined the liver fat from each patient's MRI measurements. Comparison of these data with the histologic quantified fat fraction of liver tissue showed a strong correlation (r(2)=0.93 for TE 6 ms and r(2)=0.91 for TE 10 ms). CONCLUSION: The described method can be used to determine the presence of hepatic steatosis of >10% with p<0.05.     

Water-saturated three-dimensional balanced steady-state free precession for fast abdominal fat quantification

PURPOSE: To compare the performance of a novel water-saturated b-SSFP sequence with that of a conventional T1-weighted turbo spin echo (T1W TSE) sequence for abdominal fat quantification. MATERIALS AND METHODS: A water-saturated, segmented, three-dimensional balanced steady-state free precession (b-SSFP) sequence and a traditional T1W TSE sequence were both employed on phantom and human studies. For phantom studies, a dual-layered phantom with known internal/external oil volumes was imaged using the two sequences. Images obtained by the two sequences were both processed using a computer-aided semiautomatic program for oil volume quantification. For human studies, six volunteers were scanned axially, centered at L2-L3 levels. Signal-to-noise ratio (SNR)(fat), contrast-to-noise ratio (CNR)(fat-muscle), CNR(fat-large bowel), and CNR(fat-small bowel) were calculated on hand-drawn regions of interest (ROIs), and averaged over all six slices for each subject. Statistical analyses were then performed to determine the SNR and CNR differences between images obtained by the two techniques. RESULTS: The phantom studies show that water-saturated b-SSFP offers a significantly closer estimation of true oil volumes compared with that of T1W TSE (P < 0.0001), as well as a more accurate internal/external volume ratio (P = 0.0001). In human studies, three-dimensional water-saturated b-SSFP images demonstrated higher CNR than that of T1W TSE (P < 0.0005), and very close SNR(fat) (P = 0.045). CONCLUSION: The proposed three-dimensional water-saturated b-SSFP sequence can generate high quality fat-only abdominal images with high CNR and SNR in shorter scan duration than the conventional T1W TSE approach. As images generated by this sequence suffer from no flow artifacts, and are less sensitive to bulk, respiratory, and bowel motion, three-dimensional water-saturated b-SSFP is a faster and more robust method for improving abdominal fat quantification using MRI.     

Practical quantification of necrosis in histological whole-slide images

Since the histological quantification of necrosis is a common task in medical research and practice, we evaluate different image analysis methods for quantifying necrosis in whole-slide images. In a practical usage scenario, we assess the impact of different classification algorithms and feature sets on both accuracy and computation time. We show how a well-chosen combination of multiresolution features and an efficient postprocessing step enables the accurate quantification necrosis in gigapixel images in less than a minute. The results are general enough to be applied to other areas of histological image analysis as well.     

Automated segmentation of the aorta from phase contrast MR images: Validation against expert tracing in healthy volunteers and in patients with a dilated aorta

Purpose:

To assess if segmentation of the aorta can be accurately achieved using the modulus image of phase contrast (PC) magnetic resonance (MR) acquisitions.

Materials and Methods:

PC image sequences containing both the ascending and descending aorta of 52 subjects were acquired using three different MR scanners. An automated segmentation technique, based on a 2D+t deformable surface that takes into account the features of PC aortic images, such as flow‐related effects, was developed. The study was designed to: 1) assess the variability of our approach and its robustness to the type of MR scanner, and 2) determine its sensitivity to aortic dilation and its accuracy against an expert manual tracing.

Results:

Interobserver variability in the lumen area was 0.59 ± 0.92% for the automated approach versus 10.09 ± 8.29% for manual segmentation. The mean Dice overlap measure was 0.945 ± 0.014. The method was robust to the aortic size and highly correlated (r = 0.99) with the manual tracing in terms of aortic area and diameter.

Conclusion:

A fast and robust automated segmentation of the aortic lumen was developed and successfully tested on images provided by various MR scanners and acquired on healthy volunteers as well as on patients with a dilated aorta. J. Magn. Reson. Imaging 2010;31:881–888. ©2010 Wiley‐Liss, Inc.     

Automated quantification of myocardial infarction from MR images by accounting for partial volume effects: animal, phantom, and human study

Ethics committees approved human and animal study components; informed written consent was provided (prospective human study [20 men; mean age, 62 years]) or waived (retrospective human study [16 men, four women; mean age, 59 years]). The purpose of this study was to prospectively evaluate a clinically applicable method, accounting for the partial volume effect, to automatically quantify myocardial infarction from delayed contrast material-enhanced magnetic resonance images. Pixels were weighted according to signal intensity to calculate infarct fraction for each pixel. Mean bias +/- variability (or standard deviation), expressed as percentage left ventricular myocardium (%LVM), were -0.3 +/- 1.3 (animals), -1.2 +/- 1.7 (phantoms), and 0.3 +/- 2.7 (patients), respectively. Algorithm had lower variability than dichotomous approach (2.7 vs 7.7 %LVM, P < .01) and did not differ from interobserver variability for bias (P = .31) or variability (P = .38). The weighted approach provides automatic quantification of myocardial infarction with higher accuracy and lower variability than a dichotomous algorithm.     

Combined magnitude and phase-based segmentation of the cerebral cortex in 7T MR images of the elderly

Purpose:

To propose a new method that integrates both magnitude and phase information obtained from magnetic resonance (MR) T*₂‐weighted scans for cerebral cortex segmentation of the elderly.

Materials and Methods:

This method makes use of K‐means clustering on magnitude and phase images to compute an initial segmentation, which is further refined by means of transformation with reconstruction criteria. The method was evaluated against the manual segmentation of 7T in vivo MR data of 20 elderly subjects (age = 67.7 ± 10.9). The added value of combining magnitude and phase was also evaluated by comparing the performance of the proposed method with the results obtained when limiting the available data to either magnitude or phase.

Results:

The proposed method shows good overlap agreement, as quantified by the Dice Index (0.79 ± 0.04), limited bias (average relative volume difference = 2.94%), and reasonable volumetric correlation (R = 0.555, p = 0.011). Using the combined magnitude and phase information significantly improves the segmentation accuracy compared with using either magnitude or phase.

Conclusion:

This study suggests that the proposed method is an accurate and robust approach for cerebral cortex segmentation in datasets presenting low gray/white matter contrast. J. Magn. Reson. Imaging 2012;36:99–109. © 2012 Wiley Periodicals Inc.     

Accurate segmentation of subcutaneous and intermuscular adipose tissue from MR images of the thigh

Purpose

To describe and evaluate a computer‐assisted method for assessing the quantity and distribution of adipose tissue in thigh by magnetic resonance imaging (MRI).

Materials and Methods

Twenty obese subjects were imaged on a Philips Achieva 1.5T scanner by a fast spin‐echo (FSE) sequence. A total of 636 images were acquired and analyzed by custom‐made software. Thigh subcutaneous adipose tissue (SAT) and bone were identified by fuzzy clustering segmentation and an active contour algorithm. Muscle and intermuscular adipose tissue (IMAT) were assessed by identifying the two peaks of the signal histogram with an expectation maximization algorithm. The whole analysis was performed in an unsupervised manner without the need of any user interaction.

Results

The coefficient of variation (CV) was evaluated between the unsupervised algorithm and manual analysis performed by an expert operator. The CV was low for all measurements (SAT <2%, muscle <1%, IMAT <5%). Limited manual correction of unsupervised segmentation results (less than 10% of contours modified) allowed us to further reduce the CV (SAT <0.5%, muscle <0.5%, IMAT <2%).

Conclusion

The proposed approach allowed effective computer‐assisted analysis of thigh MR images, dramatically reducing the user work compared to manual analysis. It allowed routine assessment of IMAT, a fat‐depot linked with metabolic abnormalities, important in monitoring the effect of nutrition and exercise. J. Magn. Reson. Imaging 2009;29:677–684. © 2009 Wiley‐Liss, Inc.     

Body weight and subcutaneous fat patterning in elite judokas

Body weight and fat are major performance variables in many sports. Extreme weight reduction can lead to severe medical problems. Accurate body composition measurements are fundamental for both medical and performance optimization. Relative body weight in terms of mass index (MI₁ = 0.53 M/(hs)), and in terms of body mass index (BMI = M/h²) were determined (h:stature, s:sitting height, M:body mass). Subcutaneous adipose tissue (SAT) was measured using a recently standardized ultrasound (US) method. US thickness sums from eight body sites were measured in 26 female and 35 male judokas of various weight classes. Comparisons of US and skinfold results indicate that the latter can be severely misleading in competitive judokas. Mean MI₁ of females was 22.8 kg m^?2 (BMI:22.9 kg m^?2), males: 26.7 kg m^?2 (BMI:26.5 kg m^?2), but individual differences MI₁‐BMI were larger than 0.5 kg m^?2 in 13 and larger than 1.0 kg m^?2 in three cases. Medians of SAT thickness sums D_I were three‐times higher in females (66.1 mm) than in males (21.8 mm), and the fat patterning differed significantly. Females had 8.6% (median) fibrous structures embedded in SAT, and males 20.2%. Both MI and BMI were not correlated with SAT. Mean pre‐competition weight loss was 4.3% (ie, 3.0 kg), and maximum was 9.2% (7.4 kg), indicating that modifications of weigh‐in procedures are urgently needed. D_I‐values mirror the athletes' potential to reduce ballast fat instead of short‐term weight reduction by dehydration; however, weight loss and SAT measured some weeks before the competitions were not correlated. Further, US measurements and medical longitudinal observations are required for discussing the large individual variations and possible fat minimum demands.