Radiomics-Based Machine Learning Classification for Glioma Grading Using Diffusion- and Perfusion-Weighted Magnetic Resonance Imaging.

OBJECTIVE: The aim of this study was to evaluate various radiomics-based machine learning classification models using the apparent diffusion coefficient (ADC) and cerebral blood flow (CBF) maps for differentiating between low-grade gliomas (LGGs) and high-grade gliomas (HGGs). METHODS: Fifty-two glioma patients, including 18 LGGs (grade II) and 34 HGGs (grade III/IV), were examined using a 3.0-T magnetic resonance scanner. The ADC and CBF maps were obtained from diffusion-weighted imaging and pseudo-continuous arterial spin labeling perfusion-weighted imaging, respectively. A total of 91 radiomic features were extracted from each of the tumor volume on the ADC and CBF maps. We constructed 4 types of machine learning classifiers based on (1) least absolute shrinkage and selection operator regularized logistic regression (LASSO-LR), (2) random forest (RF), (3) support vector machine (SVM) with the radial basis function kernel (SVM-RBF), and (4) SVM with the linear kernel (SVM-L). A training set with 36 gliomas (70%) was used to select the important radiomic features and train each model using 5-fold cross-validation. The remaining 16 gliomas (30%) were used as a test set. Receiver operating characteristic analysis was performed to evaluate the model performance. RESULTS: A radiomic feature, ADC first-order-based skewness, was selected as an important variable in all classification models. According to the receiver operating characteristic analysis, the areas under the curve of the LASSO-LR, RF, SVM-RBF, and SVM-L models for the training set were 0.965, 1.000, 0.979, and 0.969, respectively. For the test set, the areas under the curve of the LASSO-LR, RF, SVM-RBF, and SVM-L models were 0.883, 0.917, 0.717, and 0.917, respectively. All classification models showed sufficient diagnostic performance on the test set. CONCLUSIONS: Radiomics-based machine learning classifiers using the quantitative ADC and CBF maps are useful for differentiating HGGs from LGGs.

Magnetic Resonance Elastography: Measurement of Hepatic Stiffness Using Different Direct Inverse Problem Reconstruction Methods in Healthy Volunteers and Patients with Liver Disease.

The purpose of this study was to compare the mean hepatic stiffness values obtained by the application of two different direct inverse problem reconstruction methods to magnetic resonance elastography (MRE). Thirteen healthy men (23.2±2.1 years) and 16 patients with liver diseases (78.9±4.3 years; 12 men and 4 women) were examined for this study using a 3.0 T-MRI. The healthy volunteers underwent three consecutive scans, two 70-Hz waveform and a 50-Hz waveform scans. On the other hand, the patients with liver disease underwent scanning using the 70-Hz waveform only. The MRE data for each subject was processed twice for calculation of the mean hepatic stiffness (Pa), once using the multiscale direct inversion (MSDI) and once using the multimodel direct inversion (MMDI). There were no significant differences in the mean stiffness values among the scans obtained with two 70-Hz and different waveforms. However, the mean stiffness values obtained with the MSDI technique (with mask: 2895.3±255.8 Pa, without mask: 2940.6±265.4 Pa) were larger than those obtained with the MMDI technique (with mask: 2614.0±242.1 Pa, without mask: 2699.2±273.5 Pa). The reproducibility of measurements obtained using the two techniques was high for both the healthy volunteers [intraclass correlation coefficients (ICCs): 0.840-0.953] and the patients (ICC: 0.830-0.995). These results suggest that knowledge of the characteristics of different direct inversion algorithms is important for longitudinal liver stiffness assessments such as the comparison of different scanners and evaluation of the response to fibrosis therapy.

Evaluation of Image Quality in Three-dimensional Fat-suppressed T1-weighted Images with Fast Acquisition Mode for Upper Abdomen.

We compared the uniformity of fat-suppression and image quality using three-dimensional fat-suppressed T1-weighted gradient-echo sequences that are liver acquisition with volume acceleration (LAVA) and Turbo-LAVA at 3.0T-MRI. The subjects were seven patients with liver disease (mean age, 66.7±8.2 years). The axial slices of two LAVA sequences were used for the comparison of the uniformity of fat-suppression and image quality at a region-of-interest (ROI) of the liver dome, the porta, and the renal hilum. To yield a quantitative measurement of the uniformity of fat suppression, the percentage standard deviation (%SD) was calculated by comparing two sequences. For image signal to noise ratio (SNR), the contrast between the liver and fat (Cliver-fat), and the liver and muscle (Cliver-muscle), the other ROIs were placed in the superficial fat, liver, spleen, pancreas, and muscle. The %SD in Turbo-LAVA (28.1±16.8%) was lower than that in LAVA (41.5±13.4%). The SNRs in Turbo-LAVA (17.8±4.1 [liver], 12.5±3.0 [pancreas], 14.7±1.6 [spleen], 8.2±3.5 [fat]) were lower than those in LAVA (20.9±6.1 [liver], 16.8±4.1 [pancreas], 17.4±2.4 [spleen], 12.0±4.5 [fat]). While, the Cliver-fat in the Turbo-LAVA (0.72±0.06) was significantly higher than that in LAVA (0.59±0.07). Turbo-LAVA sequence offers superior and more homogenous fat-suppression in comparison to LAVA sequence.

Relationship between Apparent Diffusion Coefficient Distribution and Cancer Grade in Prostate Cancer and Benign Prostatic Hyperplasia.

The aim of this paper was to assess the associations between prostate cancer aggressiveness and histogram-derived apparent diffusion coefficient (ADC) parameters and determine which ADC parameters may help distinguish among stromal hyperplasia (SH), glandular hyperplasia (GH), and low-grade, intermediate-grade, and high-grade prostate cancers. The mean, median, minimum, maximum, and 10th and 25th percentile ADC values were determined from the ADC histogram and compared among two benign prostate hyperplasia (BPH) groups and three Gleason score (GS) groups. Seventy lesions were identified in 58 patients who had undergone proctectomy. Thirty-nine lesions were prostate cancers (GS 6 = 7 lesions, GS 7 = 19 lesions, GS 8 = 11 lesions, GS 9 = 2 lesions), and thirty-one lesions were BPH (SH = 15 lesions, GH = 16 lesions). There were statistically significant differences in 10th percentile and 25th percentile ADC values when comparing GS 6 to GS 7 (p < 0.05). The 10th percentile ADC values yielded the highest area under the curve (AUC). Tenth and 25th percentile ADCs can be used to more accurately differentiate lesions with GS 6 from those with GS 7 than other ADC parameters. Our data indicate that the major challenge with ADC mapping is to differentiate between SH and GS 6, and SH and GS 7.

A radiomics-based comparative study on arterial spin labeling and dynamic susceptibility contrast perfusion-weighted imaging in gliomas.

Radiomics has potential for reflecting the differences in glioma perfusion heterogeneity between arterial spin labeling (ASL) and dynamic susceptibility contrast (DSC) imaging. The aim of this study was to compare radiomic features of ASL and DSC imaging-derived parameters (cerebral blood flow, CBF) and assess radiomics-based classification models for low-grade gliomas (LGGs) and high-grade gliomas (HGGs) using their parameters. The ASL-CBF and DSC-relative CBF of 46 glioma patients were normalized (ASL-nCBF and DSC-nrCBF) for data analysis. For each map, 91 radiomic features were extracted from the tumor volume. Seventy-five radiomic features were significantly different (P < 0.00055) between ASL-nCBF and DSC-nrCBF. Positive correlations were observed in 75 radiomic features between ASL-nCBF and DSC-nrCBF. Even though ASL imaging underestimated CBF compared with DSC imaging, there were significant correlations (P < 0.00055) in the first-order-based mean, median, 90th percentile, and maximum. Texture analysis showed that ASL-nCBF and DSC-nrCBF characterized similar perfusion patterns, while ASL-nCBF could evaluate perfusion heterogeneity better. The areas under the curve of the ASL-nCBF and DSC-nrCBF radiomics-based classification models for gliomas were 0.888 and 0.962, respectively. Radiomics in ASL and DSC imaging is useful for characterizing glioma perfusion patterns quantitatively and for classifying LGGs and HGGs.

Amide proton transfer imaging of glioblastoma, neuroblastoma, and breast cancer cells on a 11.7 T magnetic resonance imaging system.

PURPOSE: The purpose of this study was (i) to determine the optimal magnetization transfer (MT) pulse parameter for amide proton transfer (APT) chemical exchange saturation transfer (CEST) imaging on an ultra-high-field magnetic resonance imaging (MRI) system and (ii) to use APT CEST imaging to noninvasively assess brain orthotopic and ectopic tumor cells transplanted into the mouse brain. METHODS: To evaluate APT without the influence of other metabolites, we prepared egg white phantoms. Next, we used 7-11-week-old nude female mice and the following cell lines to establish tumors after injection into the left striatum of mice: C6 (rat glioma, n = 8) as primary tumors and Neuro-2A (mouse neuroblastoma, n = 11) and MDA-MB231 (human breast cancer, n = 8) as metastatic tumors. All MRI experiments were performed on an 11.7 T vertical-bore scanner. CEST imaging was performed at 1 week after injection of Neuro-2A cells and at 2 weeks after injection of C6 and MDA-MB231 cells. The MT pulse amplitude was set at 2.2 µT or 4.4 µT. We calculated and compared the magnetization transfer ratio (MTR) and difference of MTR asymmetry between normal tissue and tumor (ΔMTR asymmetry) on APT CEST images between mouse models of brain tumors. Then, we performed hematoxylin and eosin (HE) staining and Ki-67 immunohistochemical staining to compare the APT CEST effect on tumor tissues and the pathological findings. RESULTS: Phantom study of the amide proton phantom containing chicken egg white, z-spectra obtained at a pulse length of 500 ms showed smaller peaks, whereas those obtained at a pulse length of 2000 ms showed slightly higher peaks. The APT CEST effect on tumor tissues was clearer at a pulse amplitude of 2.2 µT than at 4.4 µT. For all mouse models of brain tumors, ΔMTR asymmetry was higher at 2.2 µT than at 4.4 µT. ΔMTR asymmetry was significantly higher for the Neuro-2A model than for the MDA-MB231 model. HE staining revealed light bleeding in Neuro-2A tumors. Immunohistochemical staining revealed that the density of Ki-67-positive cells was higher in Neuro-2A tumors than in C6 or MDA-MB231 tumors. CONCLUSION: The MTR was higher at 4.4 µT than at 2.2 µT for each concentration of egg white at a pulse length of 500 ms or 2000 ms. High-resolution APT CEST imaging on an ultra-high-field MRI system was able to provide tumor information such as proliferative potential and intratumoral bleeding, noninvasively.

Comparison between manual and automatic image registration in image-guided radiation therapy using megavoltage cone-beam computed tomography with an imaging beam line for prostate cancer.

This study aimed to compare and assess the compatibility of the bone-structure-based manual and maximization of mutual information (MMI)-algorithm-based automatic image registration using megavoltage cone-beam computed tomography (MV-CBCT) images acquired with an imaging beam line. A total of 1163 MV-CBCT images from 30 prostate cancer patients were retrospectively analyzed. The differences between setup errors in three directions (left-right, LR; superior-inferior, SI; anterior-posterior, AP) of both registration methods were investigated. Pearson's correlation coefficients (r) and Bland-Altman agreements were evaluated. Agreements were defined by a bias close to zero and 95% limits of agreement (LoA) less than ± 3 mm. The cumulative frequencies of the absolute differences between the two registration methods were calculated to assess the distributions of the setup error differences. There were significant differences (p < 0.001) in the setup errors between both registration methods. There were moderate (SI, r = 0.45) and strong positive correlation coefficients (LR, r = 0.74; AP, r = 0.72), whereas the 95% LoA (bias ± 1.96 × standard deviation of the setup error differences) were - 1.61 ± 4.29 mm (LR), - 0.41 ± 5.45 mm (SI), and 0.67 ± 4.29 mm (AP), revealing no agreements in all directions. The cumulative frequencies (%) of the cases with absolute setup error differences within 3 mm in each direction were 80.83% (LR), 81.86% (SI), and 90.71% (AP), with all directions having large proportions of > 3-mm differences. The MMI-algorithm-based automatic registration is not compatible with the bone-structure-based manual registration and should not be used alone for prostate cancer.

Effects of Imaging Parameters on the Quality of Contrast-Enhanced MR Angiography of Cerebral Aneurysms Treated Using Stent-Assisted Coiling: A Phantom Study.

PURPOSE: To quantitatively investigate in vitro the effects of flip angle (FA), receiver bandwidth (BW), echo time (TE), and magnetic field strength (FS) on image noise and artifacts induced by stent-assisted coiling on contrast-enhanced MR angiography (CE-MRA) images, as a first step towards optimization of imaging parameters. METHODS: A phantom simulating a cerebral aneurysm treated using stent-assisted coiling was filled with diluted gadolinium contrast medium, and MR angiography were obtained using varied parameters: FA (10°-60°), BW (164-780 Hz/pixel), and FS (1.5 and 3.0T). The TE varied automatically with BW because the TE was set to the smallest value. Three kinds of indices were semi-automatically calculated to quantify the severity of stent- and coil-induced artifacts: artificial lumen narrowing (ALN) representing a decrease in the in-stent luminal area, and relative in-stent signal (RISS) and relative in-coil signal (RISC) representing an increase in the in-stent and in-coil signal intensities, respectively. We also measured the ratio of in-stent signal to noise (IS/N) for each parameter. The variation in these indices with variations in FA, BW (TE), and FS was analyzed. RESULTS: An increase in FA led to an increase of up to 65% in the RISS, while the IS/N increased by up to three times. The 1.5T scanner indicated fewer artifacts (71% lower ALN, two times higher RISS, and 40% higher RISC) than the 3.0T scanner. On the other hand, the 1.5T scanner worsened the IS/N compared with the 3.0T scanner, although the difference was relatively small. Variation in BW (and hence, TE) led to a trade-off between artifact severity and IS/N. CONCLUSION: A high FA and low FS should be used for improved artifact severity and IS/N on CE-MRA images of a stent-assisted coil. A wide BW (short TE) could improve artifact severity at the expense of the image noise.

Quantitative T1, T2, and T2* Mapping and Semi-Quantitative Neuromelanin-Sensitive Magnetic Resonance Imaging of the Human Midbrain.

PURPOSE: Neuromelanin is a dark pigment granule present within certain catecholamine neurons of the human brain. Here, we aimed to clarify the relationship between contrast of neuromelanin-sensitive magnetic resonance imaging (MRI) and MR relaxation times using T1, T2, and T2* mapping of the lower midbrain. METHODS: The subjects were 14 healthy volunteers (11 men and 3 women, mean age 29.9 ± 6.9 years). Neuromelanin-sensitive MRI was acquired using an optimized T1-weighted two-dimensional (2D)-turbo spin-echo sequence. To quantitatively evaluate the relaxation time, 2D-image data for the T1, T2, and T2* maps were also acquired. The regions of interest (substantia nigra pars compacta [SNc], superior cerebellar peduncles [SCP], cerebral peduncles [CP], and midbrain tegmentum [MT]) were manually drawn on neuromelanin-sensitive MRI to measure the contrast ratio (CR) and on relaxation maps to measure the relaxation times. RESULTS: The CR in the SNc was significantly higher than the CRs in the SCP and CP. Compared to the SCP and CP, the SNc had significantly higher T1 relaxation times. Moreover, the SNc had significantly lower T2 and T2* relaxation times than the other three regions (SCP, CP, and MT). Correlation analyses showed no significant correlations between the CRs in the SNc, SCP, and CP and each relaxation time. CONCLUSIONS: We demonstrated the relationship between the CR of neuromelanin-sensitive MRI and the relaxation times of quantitative maps of the human midbrain.

Liver acquisition with volume acceleration flex on 70-cm wide-bore and 60-cm conventional-bore 3.0-T MRI.

This study aimed to compare the uniformity of fat suppression and image quality between liver acquisition with volume acceleration flex (LAVA-Flex) and LAVA on 60-cm conventional-bore and 70-cm wide-bore 3.0-T magnetic resonance imaging (MRI). The uniformity of fat suppression by LAVA-Flex and LAVA was assessed as the efficiency of suppression of superficial fat at the levels of the liver dome, porta, and renal hilum. Percentage standard deviation (%SD) was calculated using the following equation: %SD (%) = 100 × SD of the regions of interest (ROIs)/mean value of the signal intensity (SI) in the ROIs. Signal-to-noise ratio (SNR) and contrast ratio (CR) were calculated. In the LAVA sequence, the %SD in all slices on wide-bore 3.0-T MRI was significantly higher than that on conventional-bore 3.0-T MRI (P < 0.01). However, there was no significant difference in fat signal uniformity between the conventional and wide-bore scanners when LAVA-Flex was used. In the liver, there were no significant differences in SNR between the two sequences. However, the SNR in the pancreas was lower for the wide-bore scanner than for the conventional-bore scanner for both sequences (P < 0.05). There were no significant differences in CR for the liver and fat between LAVA-Flex and LAVA in both scanners. The CR in the LAVA-Flex images obtained by wide-bore MRI was significantly higher than that in the LAVA-Flex images recorded by conventional-bore MRI (P < 0.001). LAVA-Flex offers more homogenous fat suppression in the upper abdomen than LAVA for both conventional and wide-bore 3.0-T MRI.

[Relationship between the effect of medial rotation of the foot axis by ankle dorsiflexion and the ability to visualize the femoral neck axis in the hip joint anterio-posterior radiography: evaluation by magnetic resonance images].

In scanning of the hip joint anterio-posterior radiography, by changing the lower extremities to the extension position and moving the foot axis (base line of the foot) by medial rotation, the angle of anteversion of the femoral neck is corrected. In this study, we assessed the effects on medial rotation of the femoral neck when keeping the planta vertically-positioned by ankle dorsiflexion (intermediate position of the ankle) and making change of the medial rotation angle of the foot axis by scanning the magnetic resonance (MR) images of knee joints and hip joints. The subjects in this study were 12 males (age: 37.9±13.8, weight: 67.3±5.5 kg) and 7 females (age: 27.6±5.1, weight: 50.0±4.5 kg). We measured the medial rotation angles of knee joints and femoral necks on MR images. Also, differences of these angles between males and females were compared. Although the gender differences were not found in medial rotation angle of both joints at all leg positions (P>0.05), the medial rotation angles increased by approximately 1.5 to 2.0 times larger by putting them at the intermediate position, and there were significant differences between the naturally plantar-flexed position and the intermediate position (P<0.05). In conclusion, our results showed that the optimal leg position for correcting the angle of anteversion was 20 degrees medial rotation of the foot axis at the naturally plantar-flexed position, or 10 degrees medial rotation of the foot axis at the intermediate position, regardless of gender.