BIMG-22. DEEP LEARNING SUPER-RESOLUTION MR SPECTROSCOPIC IMAGING TO MAP TUMOR METABOLISM IN MUTANT IDH GLIOMA PATIENTS

Metabolic imaging can map spatially abnormal molecular pathways with higher specificity for cancer compared to anatomical imaging. However, acquiring high resolution metabolic maps similar to anatomical MRI is challenging due to low metabolite concentrations, and alternative approaches that increase resolution by post-acquisition image processing can mitigate this limitation. We developed deep learning super-resolution MR spectroscopic imaging (MRSI) to map tumor metabolism in patients with mutant IDH glioma. We used a generative adversarial network (GAN) architecture comprised of a UNet neural network as the generator network and a discriminator network for adversarial training. For training we simulated a large data set of 9600 images with realistic quality for acquired MRSI to effectively train the deep learning model to upsample by a factor of four. Two types of training were performed: 1) using only the MRSI data, and 2) using MRSI and prior information from anatomical MRI to further enhance structural details. The performance of super-resolution methods was evaluated by peak SNR (PSNR), structure similarity index (SSIM), and feature similarity index (FSIM). After training on simulations, GAN was evaluated on measured MRSI metabolic maps acquired with resolution 5.2×5.2 mm² and upsampled to 1.3×1.3 mm². The GAN trained only on MRSI achieved PSNR = 27.94, SSIM = 0.88, FSIM = 0.89. Using prior anatomical MRI improved GAN performance to PSNR = 30.75, SSIM = 0.90, FSIM = 0.92. In the patient measured data, GAN super-resolution metabolic images provided clearer tumor margins and made apparent the tumor metabolic heterogeneity. Compared to conventional image interpolation such as bicubic or total variation, deep learning methods provided sharper edges and less blurring of structural details. Our results indicate that the proposed deep learning method is effective in enhancing the spatial resolution of metabolite maps which may better guide treatment in mutant IDH glioma patients.

Imaging Neurochemistry and Brain Structure Tracks Clinical Decline and Mechanisms of ALS in Patients

Background: Oxidative stress and protein aggregation are key mechanisms in amyotrophic lateral sclerosis (ALS) disease. Reduced glutathione (GSH) is the most important intracellular antioxidant that protects neurons from reactive oxygen species. We hypothesized that levels of GSH measured by MR spectroscopic imaging (MRSI) in the motor cortex and corticospinal tract are linked to clinical trajectory of ALS patients.

Objectives: Investigate the value of GSH imaging to probe clinical decline of ALS patients in combination with other neurochemical and structural parameters.

Methods: Twenty-four ALS patients were imaged at 3 T with an advanced MR protocol. Mapping GSH levels in the brain is challenging, and for this purpose, we used an optimized spectral-edited 3D MRSI sequence with real-time motion and field correction to image glutathione and other brain metabolites. In addition, our imaging protocol included (i) an adiabatic T1ρ sequence to image macromolecular fraction of brain parenchyma, (ii) diffusion tensor imaging (DTI) for white matter tractography, and (iii) high-resolution anatomical imaging.

Results: We found GSH in motor cortex (r = −0.431, p = 0.04) and corticospinal tract (r = −0.497, p = 0.016) inversely correlated with time between diagnosis and imaging. N-Acetyl-aspartate (NAA) in motor cortex inversely correlated (r = −0.416, p = 0.049), while mean water diffusivity (r = 0.437, p = 0.033) and T1ρ (r = 0.482, p = 0.019) positively correlated with disease progression measured by imputed change in revised ALS Functional Rating Scale. There is more decrease in the motor cortex than in the white matter for GSH compared to NAA, glutamate, and glutamine.

Conclusions: Our study suggests that a panel of biochemical and structural imaging biomarkers defines a brain endophenotype, which can be used to time biological events and clinical progression in ALS patients. Such a quantitative brain endophenotype may stratify ALS patients into more homogeneous groups for therapeutic interventions compared to clinical criteria.

Super-Resolution Whole-Brain 3D MR Spectroscopic Imaging for Mapping D-2-Hydroxyglutarate and Tumor Metabolism in Isocitrate Dehydrogenase 1-mutated Human Gliomas

Background Isocitrate dehydrogenase (IDH) mutations are highly frequent in glioma, producing high levels of the oncometabolite D-2-hydroxyglutarate (D-2HG). Hence, D-2HG represents a valuable imaging marker for IDH-mutated human glioma. Purpose To develop and evaluate a super-resolution three-dimensional (3D) MR spectroscopic imaging strategy to map D-2HG and tumor metabolism in IDH-mutated human glioma. Materials and Methods Between March and September 2018, participants with IDH1-mutated gliomas and healthy participants were prospectively scanned with a 3-T whole-brain 3D MR spectroscopic imaging protocol optimized for D-2HG. The acquired D-2HG maps with a voxel size of 5.2 × 5.2 × 12 mm were upsampled to a voxel size of 1.7 × 1.7 × 3 mm using a super-resolution method that combined weighted total variation, feature-based nonlocal means, and high-spatial-resolution anatomic imaging priors. Validation with simulated healthy and patient data and phantom measurements was also performed. The Mann-Whitney U test was used to check that the proposed super-resolution technique yields the highest peak signal-to-noise ratio and structural similarity index. Results Three participants with IDH1-mutated gliomas (mean age, 50 years ± 21 [standard deviation]; two men) and three healthy participants (mean age, 32 years ± 3; two men) were scanned. Twenty healthy participants (mean age, 33 years ± 5; 16 men) underwent a simulation of upsampled MR spectroscopic imaging. Super-resolution upsampling improved peak signal-to-noise ratio and structural similarity index by 62% (P < .05) and 7.3% (P < .05), respectively, for simulated data when compared with spline interpolation. Correspondingly, the proposed method significantly improved tissue contrast and structural information for the acquired 3D MR spectroscopic imaging data. Conclusion High-spatial-resolution whole-brain D-2-hydroxyglutarate imaging is possible in isocitrate dehydrogenase 1-mutated human glioma by using a super-resolution framework to upsample three-dimensional MR spectroscopic images acquired at lower resolution. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Huang and Lin in this issue.

The Impact of Arterial Input Function Determination Variations on Prostate Dynamic Contrast-Enhanced Magnetic Resonance Imaging Pharmacokinetic Modeling: A Multicenter Data Analysis Challenge, Part II

This multicenter study evaluated the effect of variations in arterial input function (AIF) determination on pharmacokinetic (PK) analysis of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) data using the shutter-speed model (SSM). Data acquired from eleven prostate cancer patients were shared among nine centers. Each center used a site-specific method to measure the individual AIF from each data set and submitted the results to the managing center. These AIFs, their reference tissue-adjusted variants, and a literature population-averaged AIF, were used by the managing center to perform SSM PK analysis to estimate K^trans (volume transfer rate constant), v_e (extravascular, extracellular volume fraction), k_ep (efflux rate constant), and τ_i (mean intracellular water lifetime). All other variables, including the definition of the tumor region of interest and precontrast T₁ values, were kept the same to evaluate parameter variations caused by variations in only the AIF. Considerable PK parameter variations were observed with within-subject coefficient of variation (wCV) values of 0.58, 0.27, 0.42, and 0.24 for K^trans, v_e, k_ep, and τ_i, respectively, using the unadjusted AIFs. Use of the reference tissue-adjusted AIFs reduced variations in K^trans and v_e (wCV = 0.50 and 0.10, respectively), but had smaller effects on k_ep and τ_i (wCV = 0.39 and 0.22, respectively). k_ep is less sensitive to AIF variation than K^trans, suggesting it may be a more robust imaging biomarker of prostate microvasculature. With low sensitivity to AIF uncertainty, the SSM-unique τ_i parameter may have advantages over the conventional PK parameters in a longitudinal study.

Contribution of Quantitative Amygdalar MR FLAIR Signal Analysis for Lateralization of Mesial Temporal Lobe Epilepsy

BACKGROUND AND PURPOSE

This study evaluates the contribution of an automated amygdalar fluid-attenuated inversion recovery (FLAIR) signal analysis for the lateralization of mesial temporal lobe epilepsy (mTLE).

METHODS

Sixty-nine patients (27 M, 42 F) who had undergone surgery and achieved an Engel class Ia postoperative outcome were identified as a pure cohort of mTLE cases. Forty-six nonepileptic subjects comprised the control group. The amygdala was segmented in T1-weighted images using an atlas-based segmentation. The right/left ratios of amygdalar FLAIR mean and standard deviation were calculated for each subject. A linear classifier (ie, discriminator line) was designed for lateralization using the FLAIR features and a boundary domain, within which lateralization was assumed to be less definitive, was established using the same features from control subjects. Hippocampal FLAIR and volume analysis was performed for comparison.

RESULTS

With the boundary domain in place, lateralization accuracy was found to be 70% with hippocampal FLAIR and 67% with hippocampal volume. Taking amygdalar analysis into account, 22% of cases that were found to have uncertain lateralization by hippocampal FLAIR analysis were confidently lateralized by amygdalar FLAIR. No misclassified case was found outside the amygdalar FLAIR boundary domain.

CONCLUSIONS

Amygdalar FLAIR analysis provides an additional metric by which to establish mTLE in those cases where hippocampal FLAIR and volume analysis have failed to provide lateralizing information.

Effect of Region of Interest Size on the Repeatability of Quantitative Brain Imaging Biomarkers

In the repeatability analysis, when the measurement is the mean value of a parametric map within a region of interest (ROI), the ROI size becomes important as by increasing the size, the measurement will have a smaller variance. This is important in decision-making in prospective clinical studies of brain when the ROI size is variable, e.g., in monitoring the effect of treatment on lesions by quantitative MRI, and in particular when the ROI is small, e.g., in the case of brain lesions in multiple sclerosis. Thus, methods to estimate repeatability measures for arbitrary sizes of ROI are desired. We propose a statistical model of the values of parametric map within the ROI and a method to approximate the model parameters, based on which we estimate a number of repeatability measures including repeatability coefficient, coefficient of variation, and intraclass correlation coefficient for an ROI with an arbitrary size. We also show how this gives an insight into related problems such as spatial smoothing in voxel-wise analysis. Experiments are conducted on simulated data as well as on scan-rescan brain MRI of healthy subjects. The main application of this study is the adjustment of the decision threshold based on the lesion size in treatment monitoring.

Comparative performance evaluation of automated segmentation methods of hippocampus from magnetic resonance images of temporal lobe epilepsy patients

PURPOSE

Segmentation of the hippocampus from magnetic resonance (MR) images is a key task in the evaluation of mesial temporal lobe epilepsy (mTLE) patients. Several automated algorithms have been proposed although manual segmentation remains the benchmark. Choosing a reliable algorithm is problematic since structural definition pertaining to multiple edges, missing and fuzzy boundaries, and shape changes varies among mTLE subjects. Lack of statistical references and guidance for quantifying the reliability and reproducibility of automated techniques has further detracted from automated approaches. The purpose of this study was to develop a systematic and statistical approach using a large dataset for the evaluation of automated methods and establish a method that would achieve results better approximating those attained by manual tracing in the epileptogenic hippocampus.

METHODS

A template database of 195 (81 males, 114 females; age range 32-67 yr, mean 49.16 yr) MR images of mTLE patients was used in this study. Hippocampal segmentation was accomplished manually and by two well-known tools (FreeSurfer and hammer) and two previously published methods developed at their institution [Automatic brain structure segmentation (ABSS) and LocalInfo]. To establish which method was better performing for mTLE cases, several voxel-based, distance-based, and volume-based performance metrics were considered. Statistical validations of the results using automated techniques were compared with the results of benchmark manual segmentation. Extracted metrics were analyzed to find the method that provided a more similar result relative to the benchmark.

RESULTS

Among the four automated methods, ABSS generated the most accurate results. For this method, the Dice coefficient was 5.13%, 14.10%, and 16.67% higher, Hausdorff was 22.65%, 86.73%, and 69.58% lower, precision was 4.94%, -4.94%, and 12.35% higher, and the root mean square (RMS) was 19.05%, 61.90%, and 65.08% lower than LocalInfo, FreeSurfer, and hammer, respectively. The Bland-Altman similarity analysis revealed a low bias for the ABSS and LocalInfo techniques compared to the others.

CONCLUSIONS

The ABSS method for automated hippocampal segmentation outperformed other methods, best approximating what could be achieved by manual tracing. This study also shows that four categories of input data can cause automated segmentation methods to fail. They include incomplete studies, artifact, low signal-to-noise ratio, and inhomogeneity. Different scanner platforms and pulse sequences were considered as means by which to improve reliability of the automated methods. Other modifications were specially devised to enhance a particular method assessed in this study.

Volumetric relationship between 2-hydroxyglutarate and FLAIR hyperintensity has potential implications for radiotherapy planning of mutant IDH glioma patients

BACKGROUND

Gliomas with mutant isocitrate dehydrogenase (IDH) produce high levels of 2-hydroxyglutarate (2HG) that can be quantitatively measured by 3D magnetic resonance spectroscopic imaging (MRSI). Current glioma MRI primarily relies upon fluid-attenuated inversion recovery (FLAIR) hyperintensity for treatment planning, although this lacks specificity for tumor cells. Here, we investigated the relationship between 2HG and FLAIR in mutant IDH glioma patients to determine whether 2HG mapping is valuable for radiotherapy planning.

METHODS

Seventeen patients with mutant IDH1 gliomas were imaged by 3 T MRI. A 3D MRSI sequence was employed to specifically image 2HG. FLAIR imaging was performed using standard clinical protocol. Regions of interest (ROIs) were determined for FLAIR and optimally thresholded 2HG hyperintensities. The overlap, displacement, and volumes of 2HG and FLAIR ROIs were calculated.

RESULTS

In 8 of 17 (47%) patients, the 2HG volume was larger than FLAIR volume. Across the entire cohort, the mean volume of 2HG was 35.3 cc (range, 5.3-92.7 cc), while the mean volume of FLAIR was 35.8 cc (range, 6.3-140.8 cc). FLAIR and 2HG ROIs had mean overlap of 0.28 (Dice coefficients range, 0.03-0.57) and mean displacement of 12.2 mm (range, 3.2-23.5 mm) between their centers of mass.

CONCLUSIONS

Our results indicate that for a substantial number of patients, the 2HG volumetric assessment of tumor burden is more extensive than FLAIR volume. In addition, there is only partial overlap and asymmetric displacement between the centers of FLAIR and 2HG ROIs. These results may have important implications for radiotherapy planning of IDH mutant glioma.

The Impact of Arterial Input Function Determination Variations on Prostate Dynamic Contrast-Enhanced Magnetic Resonance Imaging Pharmacokinetic Modeling: A Multicenter Data Analysis Challenge

Dynamic contrast-enhanced MRI (DCE-MRI) has been widely used in tumor detection and therapy response evaluation. Pharmacokinetic analysis of DCE-MRI time-course data allows estimation of quantitative imaging biomarkers such as K^trans(rate constant for plasma/interstitium contrast reagent (CR) transfer) and v_e (extravascular and extracellular volume fraction). However, the use of quantitative DCE-MRI in clinical prostate imaging islimited, with uncertainty in arterial input function (AIF, i.e., the time rate of change of the concentration of CR in the blood plasma) determination being one of the primary reasons. In this multicenter data analysis challenge to assess the effects of variations in AIF quantification on estimation of DCE-MRI parameters, prostate DCE-MRI data acquired at one center from 11 prostate cancer patients were shared among nine centers. Each center used its site-specific method to determine the individual AIF from each data set and submitted the results to the managing center. Along with a literature population averaged AIF, these AIFs and their reference-tissue-adjusted variants were used by the managing center to perform pharmacokinetic analysis of the DCE-MRI data sets using the Tofts model (TM). All other variables including tumor region of interest (ROI) definition and pre-contrast T₁ were kept the same to evaluate parameter variations caused by AIF variations only. Considerable pharmacokinetic parameter variations were observed with the within-subject coefficient of variation (wCV) of K^trans obtained with unadjusted AIFs as high as 0.74. AIF-caused variations were larger in K^trans than v_e and both were reduced when reference-tissue-adjusted AIFs were used. The parameter variations were largely systematic, resulting in nearly unchanged parametric map patterns. The CR intravasation rate constant, k_ep (= K^trans/v_e), was less sensitive to AIF variation than K^trans (wCV for unadjusted AIFs: 0.45 for k_epvs. 0.74 for K^trans), suggesting that it might be a more robust imaging biomarker of prostate microvasculature than K^trans.

Standard chemoradiation for glioblastoma results in progressive brain volume loss

OBJECTIVE

To investigate the effects of chemotherapy and cranial irradiation on normal brain tissue using in vivo neuroimaging in patients with glioblastoma.

METHODS

We used longitudinal MRI to monitor structural brain changes during standard treatment in patients newly diagnosed with glioblastoma. We assessed volumetric and diffusion tensor imaging measures in 14 patients receiving 6 weeks of chemoradiation, followed by up to 6 months of temozolomide chemotherapy alone. We examined changes in whole brain, gray matter (GM), white matter (WM), anterior lateral ventricle, and hippocampal volumes. Normal-appearing GM, WM, and hippocampal analyses were conducted within the hemisphere of lowest/absent tumor burden. We examined diffusion tensor imaging measures within the subventricular zone.

RESULTS

Whole brain (F = 2.41; p = 0.016) and GM (F = 2.13; p = 0.036) volume decreased during treatment, without significant WM volume change. Anterior lateral ventricle volume increased significantly (F = 65.51; p < 0.001). In participants analyzed beyond 23 weeks, mean ventricular volume increased by 42.2% (SE: 8.8%; t = 4.94; p < 0.005). Apparent diffusion coefficient increased within the subventricular zone (F = 7.028; p < 0.001). No significant changes were identified in hippocampal volume.

CONCLUSIONS

We present evidence of significant and progressive treatment-associated structural brain changes in patients with glioblastoma treated with standard chemoradiation. Future studies using longitudinal neuropsychological evaluation are needed to characterize the functional consequences of these structural changes.

Repeatability of Cerebral Perfusion Using Dynamic Susceptibility Contrast MRI in Glioblastoma Patients

OBJECTIVES

This study evaluates the repeatability of brain perfusion using dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) with a variety of post-processing methods.

METHODS

Thirty-two patients with newly diagnosed glioblastoma were recruited. On a 3-T MRI using a dual-echo, gradient-echo spin-echo DSC-MRI protocol, the patients were scanned twice 1 to 5 days apart. Perfusion maps including cerebral blood volume (CBV) and cerebral blood flow (CBF) were generated using two contrast agent leakage correction methods, along with testing normalization to reference tissue, and application of arterial input function (AIF). Repeatability of CBV and CBF within tumor regions and healthy tissues, identified by structural images, was assessed with intra-class correlation coefficients (ICCs) and repeatability coefficients (RCs). Coefficients of variation (CVs) were reported for selected methods.

RESULTS

CBV and CBF were highly repeatable within tumor with ICC values up to 0.97. However, both CBV and CBF showed lower ICCs for healthy cortical tissues (up to 0.83), healthy gray matter (up to 0.95), and healthy white matter (WM; up to 0.93). The values of CV ranged from 6% to 10% in tumor and 3% to 11% in healthy tissues. The values of RC relative to the mean value of measurement within healthy WM ranged from 22% to 42% in tumor and 7% to 43% in healthy tissues. These percentages show how much variation in perfusion parameter, relative to that in healthy WM, we expect to observe to consider it statistically significant. We also found that normalization improved repeatability, but AIF deconvolution did not.

CONCLUSIONS

DSC-MRI is highly repeatable in high-grade glioma patients.

Simple and reproducible linear measurements to determine ventricular enlargement in adults

Background:

Recent studies have suggested that Evan's Index (EI) is not accurate and instead endorse volumetric measurements. Our aim was to evaluate the reproducibility of linear measurements and their correlation to ventricular volume.

Methods:

Using magnetic resonance (MR) images of 30 patients referred for normal pressure hydrocephalus (NPH), EI, frontal-occipital horn ratio (FOR), third ventricular width and height, frontal horn width (FHW), and callosal angle (CA) at the foramen of Monro and the posterior commissure (PC) were independently measured by residents in neurosurgery and radiology, a neurosurgeon and radiologist, and a medical student. Intraclass correlation coefficients (ICC) were calculated to establish inter-rater agreement among the reviewers. Pearson's correlation coefficients were done to assess the relationship of the linear measurements with total ventricular volume. Kappa analyses were performed to assess the degree of agreement between cutpoints determined by the ROC analysis for the linear measurements and reviewers’ gestalt impression about ventricular size with volumetric abnormality.

Results:

The overall inter-rater agreement among reviewers was almost perfect for EI (ICC = 0.913), FOR (ICC = 0.830), third ventricular width, FHW (ICC = 0.88), and CA at PC (ICC = 0.865), substantial for temporal horn width (ICC = 0.729) and CA at foramen of Monro (ICC = 0.779), and moderate for third ventricular height (ICC = 0.496). EI, FOR, third ventricular width, temporal horn width, and CA at PC measures correlated with total ventricular volume. There was fair-to-almost-perfect agreement of the individual reviewer's gestalt responses of abnormatility with volumetric abnormality. Gestalt responses were better for more senior raters.

Conclusion:

Linear measurements are reliable and reproducible methods for determining ventricular enlargement.

Abstract W P46: Specific Infarct Locations Define Troponin Elevation in Acute Ischemic Stroke

Introduction: Myocardial injury is often encountered in acute ischemic stroke (AIS). Specific infarct locations have been considered as involved in its pathophysiology but there has been no systematic study involving MRI volumetric analyses. We aimed to define the imaging factors related to elevated troponin levels in AIS.

Hypothesis: Specific infarct locations would be associated with AIS-related myocardial injury.

Methods: We analyzed prospectively collected stroke registry data of patients with AIS admitted in one calendar year and included all patients with clinical and diffusion-weighted imaging (DWI) proven diagnosis of AIS, serum troponin measured within 24 hours of onset, and DWI studies of sufficient quality to permit analyses. Serum troponin ≥ 0.04 was considered elevated. A brain atlas was co-registered to each DWI volume by affine transform with 12 parameters representing pre-specified brain regions, including temporal, parietal, frontal, occipital lobes, insula and brainstem. Lesion volumes were calculated and the percent of total lesion volume within each region was computed. The relationship between troponin status (elevated vs. normal) and presence of ischemic lesion in a brain region was assessed using χ2 test. Multivariate logistic regression models with stepwise selection were done to determine which combination of brain regions were associated with elevated troponin levels.

Results: We included 266 patients; 138 (52%) were men, mean age was 66(±14) years. Elevated troponin level was found in 72 (27%) and normal in 194 (73%) patients. There was no age difference between the two groups. Patients with elevated troponin levels had larger infarct volumes than those with normal levels [21,446(±35,414) mm3 vs. 13,058 (±38,100) mm3, p=0.001). Patients with elevated troponin levels were more likely to have left parietal (OR=2.74, 95%CI=1.48 to 5.07, p=0.001) and right frontal (OR=2.05, 95%CI=1.06 to 3.94, p=0.032) infarct location.

Conclusions: Myocardial injury is often encountered in AIS, especially in patients with larger infarct volumes. Left parietal and right frontal infarct locations are independent predictors of myocardial injury. Further study will help define the clinical and prognostic significance of our findings.

Lateralization of temporal lobe epilepsy by imaging-based response-driven multinomial multivariate models

We have developed response-driven multinomial models, based on multivariate imaging features, to lateralize the epileptogenicity in temporal lobe epilepsy (TLE) patients. To this end, volumetrics and statistical quantities of FLAIR intensity and normalized ictal-interictal SPECT intensity on left and right hippocampi were extracted from preoperative images of forty-five retrospective TLE patients with surgical outcome of Engel class l. Using multinomial logistic function regression, the parameters of various univariate and multivariate models were estimated. Among univariate response models, the response model with SPECT attributes and response model with mean FLAIR attributes achieved the lowest fit deviance (65.1±0.2 and 65.5±0.3, respectively). They resulted in the highest probability of detection (0.82) and lowest probability of false alarm (0.02) for the epileptogenic side. The multivariate response model with incorporating all volumetrics, mean and standard deviation FLAIR, and SPECT attributes achieved a significantly lower fit deviance than other response models (11.9±0.1, p <; 0.001). It reached probability of detection of 1 with no false alarms. We were able to correctly lateralize the fifteen TLE patients who had undergone phase II intracranial monitoring. Therefore, the phase II intracranial monitoring might have been avoided for this set of patients. Based on this lateralization response model, the side of epileptogenicity was also detected for all thirty patients who had preceded to resection with only phase I of EEG monitoring. In conclusion, the proposed multinomial multivariate response-driven model for lateralization of epileptogenicity in TLE patients can help in decision-making prior to surgical resection and may reduce the need for implantation of intracranial monitoring electrodes.

Degree of corticospinal tract damage correlates with motor function after stroke

OBJECTIVES

Direct injury to the corticospinal tract (CST) is a major factor defining motor impairment after stroke. Diffusion tensor imaging (DTI) tractography allows definition of the CST. We sought to determine whether DTI-based assessment of the degree of CST damage correlates with motor impairment at each phase of ischemic stroke.

METHODS

We evaluated patients at the acute (3-7 days), subacute (30 days), and chronic (90 days) phases of ischemic stroke with DTI and clinical motor scores (upper extremity Fugl-Myer test [UE-FM], motor items of the National Institutes of Health Stroke Scale [mNIHSS]). The CST was identified and virtual fiber numbers (FN) were calculated for the affected and contralateral CST. We used Spearman correlation to study the relationship of FN ratio (FNr) (affected/unaffected CST) with motor scores at each time point, and the regression model to study the association of the acute parameters with chronic motor scores.

RESULTS

We studied 23 patients. Mean age was 66.7 (±12) years. FNr correlated with UE-FM score in the acute (r = 0.50, P = 0.032), subacute (r = 0.57, P = 0.007), and chronic (r = 0.67, P = 0.0008) phase, and with mNIHSS in the acute (r = -0.48, P = 0.043), subacute (r = -0.58, P = 0.006), and chronic (r = -0.75, P = 0.0001) phase. The combination of acute NIHSS and FNr significantly predicted chronic UE-FM score (r = 0.74, P = 0.0001).

INTERPRETATION

DTI-defined degree of CST injury correlates with motor impairment at each phase of ischemic stroke. The combination of baseline FNr and NIHSS predicts motor outcome. DTI-derived CST assessment could become a surrogate marker of motor impairment in the design of neurorestorative clinical trials.

MRI upsampling using feature-based nonlocal means approach

In magnetic resonance imaging (MRI), spatial resolution is limited by several factors such as acquisition time, short physiological phenomena, and organ motion. The acquired image usually has higher resolution in two dimensions (the acquisition plane) in comparison with the third dimension, resulting in highly anisotropic voxel size. Interpolation of these low resolution (LR) images using standard techniques, such as linear or spline interpolation, results in distorted edges in the planes perpendicular to the acquisition plane. This poses limitation on conducting quantitative analyses of LR images, particularly on their voxel-wise analysis and registration. We have proposed a new non-local means feature-based technique that uses structural information of a high resolution (HR) image with a different contrast and interpolates the LR image. In this approach, the similarity between voxels is estimated using a feature vector that characterizes the laminar pattern of the brain structures, resulting in a more accurate similarity measure in comparison with conventional patch-based approach. This technique can be applied to LR images with both anisotropic and isotropic voxel sizes. Experimental results conducted on brain MRI scans of patients with brain tumors, multiple sclerosis, epilepsy, as well as schizophrenic patients and normal controls show that the proposed method is more accurate, requires fewer computations, and thus is significantly faster than a previous state-of-the-art patch-based technique. We also show how the proposed method may be used to upsample regions of interest drawn on LR images.

Perfusion and Diffusion Abnormalities of Multiple Sclerosis Lesions and Relevance of Classified Lesions to Disease Status

OBJECTIVE

Hemodynamic abnormality and disruption of white matter (WM) integrity are significant components in the pathophysiology of multiple sclerosis (MS) lesions. However, the roles of stratified lesions with distinct degrees of hemodynamic and structural injury in disease states remain to be explored. We tested the hypothesis that hemodynamic and structural impairment, as assessed by cerebral blood volume (CBV) and fractional anisotropy (FA), respectively, characterizes the extent of tissue injury, and the load of lesion with substantial tissue destruction would reflect the disease status and therefore, would be related to clinical disability.

METHODS

Seven relapsing-remitting MS patients and seven healthy controls underwent perfusion, diffusion and conventional MRI scans. Based on T2-FLAIR and T1-weighted image, WM plaques were classified. After image coregistration, values of CBV and FA were estimated in three distinct lesion types (active, T1-hypointense and T1-isointense lesion) and compared with those obtained in WM from controls. A total of 1135 lesions were evaluated. Brain volumetric measurement and correlative analysis between brain atrophy, lesion volume and clinical disability were also performed.

RESULTS

Compared with normal WM, significantly reduced CBV and FA were present in the T1-hypointense lesion, while insignificant changes in both parameters were exhibited in the T1-isointense lesion. However, increased CBV but significantly decreased FA was detected in the active lesion. A close spatial relationship between active and T1-hypointense lesion was observed. Lesion load represented by T1-hypointense plus active lesion volume significantly correlated with brain atrophy, which, in turn, significantly correlated with the severity of clinical disability.

CONCLUSION

A distinct combination of CBV and FA characterizes the status of a specific lesion type. A severe structural impairment does not solely occur in the T1-hypointense lesion, but is also associated with the active lesion. The burden of the lesion with extensive structural damage provides an image index, indicative of disease status.

Slice accelerated gradient-echo spin-echo dynamic susceptibility contrast imaging with blipped CAIPI for increased slice coverage

PURPOSE

To improve slice coverage of gradient echo spin echo (GESE) sequences for dynamic susceptibility contrast (DSC) MRI using a simultaneous-multiple-slice (SMS) method.

METHODS

Data were acquired on 3 Tesla (T) MR scanners with a 32-channel head coil. To evaluate use of SMS for DSC, an SMS GESE sequence with two-fold slice coverage and same temporal sampling was compared with a standard GESE sequence, both with 2× in-plane acceleration. A signal to noise ratio (SNR) comparison was performed on one healthy subject. Additionally, data with Gadolinium injection were collected on three patients with glioblastoma using both sequences, and perfusion analysis was performed on healthy tissues as well as on tumor.

RESULTS

Retained SNR of SMS DSC is 90% for a gradient echo (GE) and 99% for a spin echo (SE) acquisition, compared with a standard acquisition without slice acceleration. Comparing cerebral blood volume maps, it was observed that the results of standard and SMS acquisitions are comparable for both GE and SE images.

CONCLUSION

Two-fold slice accelerated DSC MRI achieves similar SNR and perfusion metrics as a standard acquisition, while allowing a significant increase in slice coverage of the brain. The results also point to a possibility to improve temporal sampling rate, while retaining the same slice coverage.

Confident Surgical Decision Making in Temporal Lobe Epilepsy by Heterogeneous Classifier Ensembles

In medical domains with low tolerance for invalid predictions, classification confidence is highly important and traditional performance measures such as overall accuracy cannot provide adequate insight into classifications reliability. In this paper, a confident-prediction rate (CPR) which measures the upper limit of confident predictions has been proposed based on receiver operating characteristic (ROC) curves. It has been shown that heterogeneous ensemble of classifiers improves this measure. This ensemble approach has been applied to lateralization of focal epileptogenicity in temporal lobe epilepsy (TLE) and prediction of surgical outcomes. A goal of this study is to reduce extraoperative electrocorticography (eECoG) requirement which is the practice of using electrodes placed directly on the exposed surface of the brain. We have shown that such goal is achievable with application of data mining techniques. Furthermore, all TLE surgical operations do not result in complete relief from seizures and it is not always possible for human experts to identify such unsuccessful cases prior to surgery. This study demonstrates the capability of data mining techniques in prediction of undesirable outcome for a portion of such cases.

Three-dimensional volumetric measurements and analysis of the maxillary sinus

BACKGROUND

Multiple chronic rhinosinusitis (CRS) staging systems have been developed in an attempt to correlate symptoms with radiological imaging results. Currently, no perfect system exists. We sought to analyze the maxillary sinus (MS) using three-dimensional volumetric measurements and advanced high-resolution CT imaging.

METHODS

We reviewed MS CT scans from 50 control subjects and 50 subjects with documented CRS involving at least one MS. The following measurements were recorded: (1) volume of MS free air, (2) MS mucosal thickening, and (3) MS lateral wall bony thickness. Average Hounsfield unit (HU) values for mucosal thickening among CRS subjects were also recorded. Values are expressed as mean ± SD and median. Values from the CRS patients were compared with healthy controls using Student's t-tests.

RESULTS

Among controls (n = 50), volumes (mL) of right and left MS were 24.1 ± 9.7 and 24.7 ± 9.0, respectively. Among CRS patients (n = 50), the portion of mucosal disease to total sinus volume was 51.8% (right) and 50.7% (left). Mean bony thickness (mm) in controls was 0.98 ± 0.2 (right) and 1.0 ± 0.3 (left). CRS patients had significantly greater bony thickness 1.9 ± 0.8 (right) and 2.0 ± 0.9 (left; p = 0.0001). HU for diseased MS were 30.1 ± 18.7 (right) and 35.7 ± 22.1 (left).

CONCLUSION

Three-dimensional volumetric analysis combined with HU calculations and bony thickness measurements represents a new and unique way to evaluate CT scans in patients with CRS. Additional studies correlating symptoms with imaging findings as well as analysis of all paranasal sinuses is the next step toward a novel staging system.

High inter-reviewer variability of spike detection on intracranial EEG addressed by an automated multi-channel algorithm

OBJECTIVE

The goal of this study was to determine the consistency of human reviewer spike detection and then develop a computer algorithm to make the intracranial spike detection process more objective and reliable.

METHODS

Three human reviewers marked interictal spikes on samples of intracranial EEGs from 10 patients. The sensitivity, precision and agreement in channel ranking by activity were calculated between reviewers. A computer algorithm was developed to parallel the way human reviewers detect spikes by first identifying all potential spikes on each channel using frequency filtering and then block scaling all channels at the same time in order to exclude potential spikes that fall below an amplitude and slope threshold. Its performance was compared to the human reviewers on the same set of patients.

RESULTS

Human reviewers showed surprisingly poor inter-reviewer agreement, but did broadly agree on the ranking of channels for spike activity. The computer algorithm performed as well as the human reviewers and did especially well at ranking channels from highest to lowest spike frequency.

CONCLUSIONS

Our algorithm showed good agreement with the different human reviewers, even though they demonstrated different criteria for what constitutes a 'spike' and performed especially well at the clinically important task of ranking channels by spike activity.

SIGNIFICANCE

An automated, objective method to detect interictal spikes on intracranial recordings will improve both research and the surgical management of epilepsy patients.

Predicting final extent of ischemic infarction using artificial neural network analysis of multi-parametric MRI in patients with stroke

In hemispheric ischemic stroke, the final size of the ischemic lesion is the most important correlate of clinical functional outcome. Using a set of acute-phase MR images (Diffusion-weighted - DWI, T₁-weighted – T1WI, T₂-weighted-T2WI, and proton density weighted - PDWI) for inputs, and the chronic T2WI at 3 months as an outcome measure, an Artificial Neural Network (ANN) was trained to predict the 3-month outcome in the form of a voxel-by-voxel forecast of the chronic T2WI. The ANN was trained and tested using 12 subjects (with 83 slices and 140218 voxels) using a leave-one-out cross-validation method with calculation of the Area Under the Receiver Operator Characteristic Curve (AUROC) for training, testing and optimization of the ANN. After training and optimization, the ANN produced maps of predicted outcome that were well correlated (r = 0.80, p<0.0001) with the T2WI at 3 months for all 12 patients. This result implies that the trained ANN can provide an estimate of 3-month ischemic lesion on T2WI in a stable and accurate manner (AUROC = 0.89).

Influence of brain-derived neurotrophic factor and apolipoprotein E genetic variants on hemispheric and lateral ventricular volume of young healthy adults

OBJECTIVE

Brain-derived neurotrophic factor (BDNF) and apolipoprotein E (ApoE) are thought to be implicated in a variety of neuronal processes, including cell growth, resilience to noxious stimuli and synaptic plasticity. A Val to Met substitution at codon 66 in the BDNF protein has been associated with a variety of neuropsychiatric conditions. The ApoE4 allele is considered a risk factor for late-onset Alzheimer's disease, but its effects on young adults are less clear. We sought to investigate the effects of those two polymorphisms on hemispheric and lateral ventricular volumes of young healthy adults.

METHODS

Hemispheric and lateral ventricular volumes of 144 healthy individuals, aged 19-35 years, were measured using high resolution magnetic resonance imaging and data were correlated with BDNF and ApoE genotypes.

RESULTS

There were no correlations between BDNF or ApoE genotype and hemispheric or lateral ventricular volumes.

CONCLUSION

These findings indicate that it is unlikely that either the BDNF Val66Met or ApoE polymorphisms exert any significant effect on hemispheric or lateral ventricular volume. However, confounding epistatic genetic effects as well as relative insensitivity of the volumetric methods used cannot be ruled out. Further imaging analyses are warranted to better define any genetic influence of the BDNF Val6Met and ApoE polymorphism on brain structure of young healthy adults.

Clustering method for estimating principal diffusion directions

Diffusion tensor magnetic resonance imaging (DTMRI) is a non-invasive tool for the investigation of white matter structure within the brain. However, the traditional tensor model is unable to characterize anisotropies of orders higher than two in heterogeneous areas containing more than one fiber population. To resolve this issue, high angular resolution diffusion imaging (HARDI) with a large number of diffusion encoding gradients is used along with reconstruction methods such as Q-ball. Using HARDI data, the fiber orientation distribution function (ODF) on the unit sphere is calculated and used to extract the principal diffusion directions (PDDs). Fast and accurate estimation of PDDs is a prerequisite for tracking algorithms that deal with fiber crossings. In this paper, the PDDs are defined as the directions around which the ODF data is concentrated. Estimates of the PDDs based on this definition are less sensitive to noise in comparison with the previous approaches. A clustering approach to estimate the PDDs is proposed which is an extension of fuzzy c-means clustering developed for orientation of points on a sphere. MDL (Minimum description length) principle is proposed to estimate the number of PDDs. Using both simulated and real diffusion data, the proposed method has been evaluated and compared with some previous protocols. Experimental results show that the proposed clustering algorithm is more accurate, more resistant to noise, and faster than some of techniques currently being utilized.

Quantitative multi-compartmental SPECT image analysis for lateralization of temporal lobe epilepsy

This study assesses the utility of compartmental analysis of SPECT data in lateralizing ictal onset in cases of a putative mesial temporal lobe epilepsy (mTLE). An institutional archival review provided 46 patients (18M, 28F) operated for a putative mTLE who achieved an Engel class Ia postoperative outcome. This established the standard to assure a true ictal origin. Ictal and interictal SPECT images were separately coregistered to T1-weighted (T1W) magnetic resonance (MR) image using a rigid transformation and the intensities matched with an l(1) norm minimization technique. The T1W MR image was segmented into separate structures using an atlas-based automatic segmentation technique with the hippocampi manually segmented to improve accuracy. Mean ictal-interictal intensity difference values were calculated for select subcortical structures and the accuracy of lateralization evaluated using a linear classifier. Hippocampal SPECT analysis yielded the highest lateralization accuracy (91%) followed by the amygdala (87%), putamen (67%) and thalamus (61%). Comparative FLAIR and volumetric analyses yielded 89% and 78% accuracies, respectively. A multi-modality analysis did not generate a higher accuracy (89%). A quantitative anatomically compartmented approach to SPECT analysis yields a particularly high lateralization accuracy in the case of mTLE comparable to that of quantitative FLAIR MR imaging. Hippocampal segmentation in this regard correlates well with ictal origin and shows good reliability in the preoperative analysis.

Hippocampal volumetry for lateralization of temporal lobe epilepsy: automated versus manual methods

The hippocampus has been the primary region of interest in the preoperative imaging investigations of mesial temporal lobe epilepsy (mTLE). Hippocampal imaging and electroencephalographic features may be sufficient in several cases to declare the epileptogenic focus. In particular, hippocampal atrophy, as appreciated on T1-weighted (T1W) magnetic resonance (MR) images, may suggest a mesial temporal sclerosis. Qualitative visual assessment of hippocampal volume, however, is influenced by head position in the magnet and the amount of atrophy in different parts of the hippocampus. An entropy-based segmentation algorithm for subcortical brain structures (LocalInfo) was developed and supplemented by both a new multiple atlas strategy and a free-form deformation step to capture structural variability. Manually segmented T1-weighted magnetic resonance (MR) images of 10 non-epileptic subjects were used as atlases for the proposed automatic segmentation protocol which was applied to a cohort of 46 mTLE patients. The segmentation and lateralization accuracies of the proposed technique were compared with those of two other available programs, HAMMER and FreeSurfer, in addition to the manual method. The Dice coefficient for the proposed method was 11% (p<10(-5)) and 14% (p<10(-4)) higher in comparison with the HAMMER and FreeSurfer, respectively. Mean and Hausdorff distances in the proposed method were also 14% (p<0.2) and 26% (p<10(-3)) lower in comparison with HAMMER and 8% (p<0.8) and 48% (p<10(-5)) lower in comparison with FreeSurfer, respectively. LocalInfo proved to have higher concordance (87%) with the manual segmentation method than either HAMMER (85%) or FreeSurfer (83%). The accuracy of lateralization by volumetry in this study with LocalInfo was 74% compared to 78% with the manual segmentation method. LocalInfo yields a closer approximation to that of manual segmentation and may therefore prove to be more reliable than currently published automatic segmentation algorithms.

Differentiation of glioma and radiation injury in rats using in vitro produce magnetically labeled cytotoxic T-cells and MRI

Background

A limitation with current imaging strategies of recurrent glioma undergoing radiotherapy is that tumor and radiation injury cannot be differentiated with post contrast CT or MRI, or with PET or other more complex parametric analyses of MRI data. We propose to address the imaging limitation building on emerging evidence indicating that effective therapy for recurrent glioma can be attained by sensitized T-cells following vaccination of primed dendritic cells (DCs). The purpose of this study was to determine whether cord blood T-cells can be sensitized against glioma cells (U-251) and if these sensitized cytotoxic T-cells (CTLs) can be used as cellular magnetic resonance imaging probes to identify and differentiate glioma from radiation necrosis in rodent models.

Methodology/Principal Findings

Cord blood T and CD14+ cells were collected. Isolated CD14+ cells were then converted to dendritic cells (DCs), primed with glioma cell lysate and used to sensitize T-cells. Phenotypical expression of the generated DCs were analyzed to determine the expression level of CD14, CD86, CD83 and HLA-DR. Cells positive for CD25, CD4, CD8 were determined in generated CTLs. Specificity of cytotoxicity of the generated CTLs was also determined by lactate dehydrogenase (LDH) release assay. Secondary proliferation capacity of magnetically labeled and unlabeled CTLs was also determined. Generated CTLs were magnetically labeled and intravenously injected into glioma bearing animals that underwent MRI on days 3 and 7 post- injection. CTLs were also administered to animals with focal radiation injury to determine whether these CTLs accumulated non-specifically to the injury sites. Multi-echo T2- and T2*-weighted images were acquired and R2 and R2* maps created. Our method produced functional, sensitized CTLs that specifically induced U251 cell death in vitro. Both labeled and unlabeled CTLs proliferated equally after the secondary stimulation. There were significantly higher CD25 positive cells (p = <0.006) in CTLs. In addition, T2- and T2*-weighted MR images showed increased low signal intensity areas in animals that received labeled CTLs as compared to the images from animals that received control cells. Histological analysis confirmed the presence of iron positive cells in sites corresponding to MRI low signal intensity regions. Significant differences (p = <0.001) in tumor R2 and R2* values were observed among the groups of animals. Animals with radiation injury exhibited neither MRI hypointense areas nor presence of iron positive cells.

Conclusion

Our results indicate that T-cells can be effectively sensitized by in vitro methods and used as cellular probes to identify and differentiate glioma from radiation necrosis.

FLAIR signal and texture analysis for lateralizing mesial temporal lobe epilepsy

Standard magnetic resonance (MR) imaging analysis in several cases of mesial temporal lobe epilepsy (mTLE) either fail to show an identifiable hippocampal asymmetry or provide only subtle distinguishing features that remain inconclusive. A retrospective analysis of hippocampal fluid-attenuated inversion recovery (FLAIR) MR images was performed in cases of mTLE addressing, particularly, the mean and standard deviation of the signal and its texture. Preoperative T1-weighted and FLAIR MR images of 25 nonepileptic control subjects and 36 mTLE patients with Engel class Ia outcomes were analyzed. Patients requiring extraoperative electrocorticography (ECoG) with intracranial electrodes and thus judged to be more challenging were studied as a separate cohort. Hippocampi were manually segmented on T1-weighted images and their outlines were transposed onto FLAIR studies using an affine registration. Image intensity features including mean and standard deviation and wavelet-based texture features were determined for the hippocampal body. The right/left ratios of these features were used with a linear classifier to establish laterality. Whole hippocampal within-subject volume ratios were assessed for comparison. Mean and standard deviation of FLAIR signal intensities lateralized the site of epileptogenicity in 98% of all cases, whereas analysis of wavelet texture features and hippocampal volumetry each yielded correct lateralization in 94% and 83% of cases, respectively. Of patients requiring more intensive study with extraoperative ECoG, 17/18 were lateralized effectively by the combination of mean and standard deviation ratios despite a ratio of mean signal intensity near one in some. The analysis of mean and standard deviation of FLAIR signal intensities provides a highly sensitive method for lateralizing the epileptic focus in mTLE over that of volumetry or texture analysis of the hippocampal body.

Magnetically-labeled sensitized splenocytes to identify glioma by MRI: a preliminary study

This study investigated the feasibility of imaging the migration and incorporation of magnetically-labeled sensitized splenocytes in an experimental 9L glioma brain tumor model. Splenocytes collected from tumor-bearing (sensitized splenocytes) or control (nonsensitized splenocytes) host rats were analyzed to determine the population of different cells, labeled with ferumoxides-protamine sulfate (FePro) and injected intravenously to recipient rats (N=4, for each group) bearing intracranial 9L tumors. Day 3 postinjection of splenocytes multiecho T2*-weighted and three-dimensional (3D) gradient echo MRI were obtained using a 7 Tesla MR system. R2* (1/T2*) maps were created from the T2*-weighted images. Signal intensities (SIs) and R2* values in the tumors and contralateral brain were determined by hand drawn regions of interest (ROIs). Brain sections were stained for the evidence of administered cells. Both 3D and T2*-weighted MRI showed low signal intensity areas in and around the tumors in rats that received labeled sensitized splenocytes. Prussian blue (PB), CD45- and CD8-positive cells were present in areas at the corresponding sites of low signal intensities seen on MRI. Rats that received labeled nonsensitized splenocytes did not show low signal intensity areas or PB positive cells in or around the implanted tumors. In conclusion, the immunogenic reaction can be exploited to delineate recurrent glioma using MRI following systemically delivered magnetically labeled sensitized splenocytes or T-cells.

Automated segmentation and classification of high throughput yeast assay spots

Several technologies for characterizing genes and proteins from humans and other organisms use yeast growth or color development as read outs. The yeast two-hybrid assay, for example, detects protein-protein interactions by measuring the growth of yeast on a specific solid medium, or the ability of the yeast to change color when grown on a medium containing a chromogenic substrate. Current systems for analyzing the results of these types of assays rely on subjective and inefficient scoring of growth or color by human experts. Here, an image analysis system is described for scoring yeast growth and color development in high throughput biological assays. The goal is to locate the spots and score them in color images of two types of plates named "X-Gal" and "growth assay" plates, with uniformly placed spots (cell areas) on each plate (both plates in one image). The scoring system relies on color for the X-Gal spots, and texture properties for the growth assay spots. A maximum likelihood projection-based segmentation is developed to automatically locate spots of yeast on each plate. Then color histogram and wavelet texture features are extracted for scoring using an optimal linear transformation. Finally, an artificial neural network is used to score the X-Gal and growth assay spots using the extracted features. The performance of the system is evaluated using spots of 60 images. After training the networks using training and validation sets, the system was assessed on the test set. The overall accuracies of 95.4% and 88.2% are achieved, respectively, for scoring the X-Gal and growth assay spots.

Automated segmentation and classification of high throughput yeast assay spots

Several technologies for characterizing genes and proteins from humans and other organisms use yeast growth or color development as read outs. The yeast two-hybrid assay, for example, detects protein-protein interactions by measuring the growth of yeast on a specific solid medium, or the ability of the yeast to change color when grown on a medium containing a chromogenic substrate. Current systems for analyzing the results of these types of assays rely on subjective and inefficient scoring of growth or color by human experts. Here, an image analysis system is described for scoring yeast growth and color development in high throughput biological assays. The goal is to locate the spots and score them in color images of two types of plates named "X-Gal" and "growth assay" plates, with uniformly placed spots (cell areas) on each plate (both plates in one image). The scoring system relies on color for the X-Gal spots, and texture properties for the growth assay spots. A maximum likelihood projection-based segmentation is developed to automatically locate spots of yeast on each plate. Then color histogram and wavelet texture features are extracted for scoring using an optimal linear transformation. Finally, an artificial neural network is used to score the X-Gal and growth assay spots using the extracted features. The performance of the system is evaluated using spots of 60 images. After training the networks using training and validation sets, the system was assessed on the test set. The overall accuracies of 95.4% and 88.2% are achieved, respectively, for scoring the X-Gal and growth assay spots.

An image analysis approach for automatic malignancy determination of prostate pathological images

BACKGROUND

Determining malignancy of prostate pathological samples is important for treatment planning of prostate cancer. Traditionally, this is performed by expert pathologists who evaluate the structure of prostate glands in the biopsy samples. However, this is a subjective task due to inter- and intra-observer differences among pathologists. Also, it is time-consuming and difficult to some extent. Therefore, automatic determination of malignancy of prostate pathological samples is of interest.

METHODS

A texture-based technique is first used to segment the prostate glands in the image. Features related to size and shape of these glands are then extracted and combined to generate an index, which is proportional to malignancy of cancer. A linear classifier is employed to classify the specimens into benign (low potential for malignancy) and malignant.

RESULTS

The leave-one-out technique is employed to evaluate the method using two datasets. The first has 91 images with similar magnifications and illuminations while the second has 199 images with different magnifications and illuminations. In the experiments, accuracies of about 98 and 95% have been achieved for these two datasets, respectively.

CONCLUSIONS

An image analysis approach is employed to evaluate prostate pathological images. Experimental results show that the proposed method can successfully classify the prostate biopsy samples into benign and malignant. They also show that the proposed method is robust to variations in magnification and illumination.

Radon transform orientation estimation for rotation invariant texture analysis

This paper presents a new approach to rotation invariant texture classification. The proposed approach benefits from the fact that most of the texture patterns either have directionality (anisotropic textures) or are not with a specific direction (isotropic textures). The wavelet energy features of the directional textures change significantly when the image is rotated. However, for the isotropic images, the wavelet features are not sensitive to rotation. Therefore, for the directional textures, it is essential to calculate the wavelet features along a specific direction. In the proposed approach, the Radon transform is first employed to detect the principal direction of the texture. Then, the texture is rotated to place its principal direction at 0 degrees. A wavelet transform is applied to the rotated image to extract texture features. This approach provides a features space with small intraclass variability and, therefore, good separation between different classes. The performance of the method is evaluated using three texture sets. Experimental results show the superiority of the proposed approach compared with some existing methods.

Rotation-invariant multiresolution texture analysis using radon and wavelet transforms

A new rotation-invariant texture-analysis technique using Radon and wavelet transforms is proposed. This technique utilizes the Radon transform to convert the rotation to translation and then applies a translation-invariant wavelet transform to the result to extract texture features. A kappa-nearest neighbors classifier is employed to classify texture patterns. A method to find the optimal number of projections for the Radon transform is proposed. It is shown that the extracted features generate an efficient orthogonal feature space. It is also shown that the proposed features extract both of the local and directional information of the texture patterns. The proposed method is robust to additive white noise as a result of summing pixel values to generate projections in the Radon transform step. To test and evaluate the method, we employed several sets of textures along with different wavelet bases. Experimental results show the superiority of the proposed method and its robustness to additive white noise in comparison with some recent texture-analysis methods.

Multiwavelet grading of pathological images of prostate

Histological grading of pathological images is used to determine level of malignancy of cancerous tissues. This is a very important task in prostate cancer prognosis, since it is used for treatment planning. If infection of cancer is not rejected by non-invasive diagnostic techniques like magnetic resonance imaging, computed tomography scan, and ultrasound, then biopsy specimens of tissue are tested. For prostate, biopsied tissue is stained by hematoxyline and eosine method and viewed by pathologists under a microscope to determine its histological grade. Human grading is very subjective due to interobserver and intraobserver variations and in some cases difficult and time-consuming. Thus, an automatic and repeatable technique is needed for grading. Gleason grading system is the most common method for histological grading of prostate tissue samples. According to this system, each cancerous specimen is assigned one of five grades. Although some automatic systems have been developed for analysis of pathological images, Gleason grading has not yet been automated; the goal of this research is to automate it. To this end, we calculate energy and entropy features of multiwavelet coefficients of the image. Then, we select most discriminative features by simulated annealing and use a k-nearest neighbor classifier to classify each image to appropriate grade (class). The leaving-one-out technique is used for error rate estimation. We also obtain the results using features extracted by wavelet packets and co-occurrence matrices and compare them with the multiwavelet method. Experimental results show the superiority of the multiwavelet transforms compared with other techniques. For multiwavelets, critically sampled preprocessing outperforms repeated-row preprocessing and has less sensitivity to noise for second level of decomposition. The first level of decomposition is very sensitive to noise and, thus, should not be used for feature extraction. The best multiwavelet method grades prostate pathological images correctly 97% of the time.