Development and Validation of a Deep Learning-Based Automatic Brain Segmentation and Classification Algorithm for Alzheimer Disease Using 3D T1-Weighted Volumetric Images

BACKGROUND AND PURPOSE

Limited evidence has suggested that a deep learning automatic brain segmentation and classification method, based on T1-weighted brain MR images, can predict Alzheimer disease. Our aim was to develop and validate a deep learning-based automatic brain segmentation and classification algorithm for the diagnosis of Alzheimer disease using 3D T1-weighted brain MR images.

MATERIALS AND METHODS

A deep learning-based algorithm was developed using a dataset of T1-weighted brain MR images in consecutive patients with Alzheimer disease and mild cognitive impairment. We developed a 2-step algorithm using a convolutional neural network to perform brain parcellation followed by 3 classifier techniques including XGBoost for disease prediction. All classification experiments were performed using 5-fold cross-validation. The diagnostic performance of the XGBoost method was compared with logistic regression and a linear Support Vector Machine by calculating their areas under the curve for differentiating Alzheimer disease from mild cognitive impairment and mild cognitive impairment from healthy controls.

RESULTS

In a total of 4 datasets, 1099, 212, 711, and 705 eligible patients were included. Compared with the linear Support Vector Machine and logistic regression, XGBoost significantly improved the prediction of Alzheimer disease (P < .001). In terms of differentiating Alzheimer disease from mild cognitive impairment, the 3 algorithms resulted in areas under the curve of 0.758-0.825. XGBoost had a sensitivity of 68% and a specificity of 70%. In terms of differentiating mild cognitive impairment from the healthy control group, the 3 algorithms resulted in areas under the curve of 0.668-0.870. XGBoost had a sensitivity of 79% and a specificity of 80%.

CONCLUSIONS

The deep learning-based automatic brain segmentation and classification algorithm allowed an accurate diagnosis of Alzheimer disease using T1-weighted brain MR images. The widespread availability of T1-weighted brain MR imaging suggests that this algorithm is a promising and widely applicable method for predicting Alzheimer disease.

Gadolinium enhancement of atherosclerotic plaque in the middle cerebral artery: relation to symptoms and degree of stenosis

BACKGROUND AND PURPOSE

High-resolution MR imaging can depict intracranial arterial atherosclerotic plaques. Our aim was to evaluate the relationship between the degree of enhancement of MCA plaques on contrast-enhanced high-resolution MR imaging and ischemic stroke and stenosis severity.

MATERIALS AND METHODS

This study enrolled 36 patients diagnosed with moderate-to-severe atherosclerotic MCA stenosis. A contrast-enhanced T1-weighted volume isotropic turbo spin-echo acquisition sequence was acquired for assessing plaque enhancement. Plaque-to-CSF contrast ratio was calculated after the signal intensity of plaques at the stenotic segment was measured. Univariate comparison and multivariate logistic regression analyses were performed for symptomatic and asymptomatic groups to assess the relationship between symptomatic stenosis and independent variables, including plaque-to-CSF contrast ratio, degree of stenosis, and clinical risk factors. Plaque-to-CSF contrast ratio was compared between the moderate and severe stenosis groups.

RESULTS

Twenty-one patients had symptomatic MCA stenosis, and 15 had asymptomatic stenosis. The plaque-to-CSF contrast ratio was significantly higher in the symptomatic group than in the asymptomatic group (63.6 ± 10.6% versus 54.1 ± 13.5%, respectively; P < .05). The degree of stenosis also differed significantly between the 2 groups (P < .05). Multivariate analysis revealed that the degree of stenosis was the only independent predictor of ischemic stroke symptoms. The plaque-to-CSF contrast ratio of severe stenosis was significantly higher than that of moderate stenosis (66.8 ± 8.7% versus 55.9 ± 12.8%, respectively; P < .05).

CONCLUSIONS

Plaque enhancement was significantly higher in patients with symptomatic plaques and may have been affected by the degree of stenosis. A difference in plaque enhancement according to the degree of stenosis has implications for understanding the development of intracranial atherosclerotic plaques.

Cyclophilin B is involved in p300-mediated degradation of CHOP in tumor cell adaptation to hypoxia

The regulation of CCAAT/enhancer-binding protein-homologous protein (CHOP), an endoplasmic reticulum (ER) stress-response factor, is key to cellular survival. Hypoxia is a physiologically important stress that induces cell death in the context of the ER, especially in solid tumors. Although our previous studies have suggested that Cyclophilin B (CypB), a molecular chaperone, has a role in ER stress, currently, there is no direct information supporting its mechanism under hypoxia. Here, we demonstrate for the first time that CypB is associated with p300 E4 ligase, induces ubiquitination and regulates the proteasomal turnover of CHOP, one of the well-known pro-apoptotic molecules under hypoxia. Our findings show that CypB physically interacts with the N-terminal α-helix domain of CHOP under hypoxia and cooperates with p300 to modulate the ubiquitination of CHOP. We also show that CypB is transcriptionally induced through ATF6 under hypoxia. Collectively, these findings demonstrate that CypB prevents hypoxia-induced cell death through modulation of ubiquitin-mediated CHOP protein degradation, suggesting that CypB may have an important role in the tight regulation of CHOP under hypoxia.

Time-resolved MR angiography for detecting and grading ovarian venous reflux: comparison with conventional venography

OBJECTIVE

The purpose of this study was to compare the diagnostic accuracy of time-resolved MR angiography (TR-MRA) with that of conventional venography for the detection and grading of ovarian venous reflux, which aid in the diagnosis of pelvic venous congestion.

METHODS

We performed a retrospective analysis of 19 consecutive patients who underwent TR-MRA and conventional venography. The images were analysed by two radiologists in a randomised "blinded" manner. With the use of conventional venography as a gold standard, the images were reviewed to determine if differences in the detection and grading of ovarian venous reflux were seen between TR-MRA and conventional venography; the sensitivity, specificity and accuracy of TR-MRA compared with that of conventional venography were evaluated. The McNemar test was performed to determine the significance of any differences. Interobserver agreement was analysed using generalised κ statistics.

RESULTS

There was no significant difference between TR-MRA and conventional venography for grading ovarian venous reflux (p>0.05). The sensitivity, specificity and diagnostic accuracy of TR-MRA were found to be 66.7%, 100% and 78.9%, and 75%, 100% and 84.2%, respectively, for the two observers. The weighted κ-values indicated excellent agreement between the two observers for grading ovarian venous reflux on TR-MRA (κ = 0.894).

CONCLUSION

TR-MRA is an accurate method for accessing pelvic venous congestion.

The detection and discrimination of malignant and benign focal hepatic lesions: T2 weighted vs diffusion-weighted MRI

OBJECTIVE

The purpose of this study was to evaluate the use of diffusion-weighted imaging (DWI) for the detection and characterisation of focal hepatic lesions compared with the use of T(2) weighted imaging.

METHOD

45 patients with 97 hepatic lesions (51 malignant lesions and 46 benign lesions) were included in this retrospective study. Malignant hepatic lesions included 12 hepatocellular carcinomas, 26 metastases and 13 intrahepatic cholangiocarcinomas. Benign hepatic lesions included 19 haemangiomas and 27 cysts. The MRI protocol for the upper abdomen included T(2) weighted images, in- and opposed-phase T(1) weighted images and dynamic T(1) weighted images. Breath-hold fat-suppressed single-shot echo planar DWI was performed with the following parameters: 1338/66; b factors, 0, 50 and 800 s mm(-2). Two independent observers reviewed the T(2) weighted images and the DWI to detect and to characterise the hepatic lesions.

RESULTS

For detection of malignant hepatic lesions, the use of DWI showed a significantly higher detection rate than the use of T(2) weighted images (p<0.05). However, there was no significant difference between the use of DWI and T(2) weighted images for benign hepatic lesions. For the differentiation between malignant and benign hepatic lesions, there was no significant difference in sensitivity, specificity and accuracy between the use of T(2) weighted images and the use of DWI.

CONCLUSION

The use of DWI was better for the detection of malignant hepatic lesions than the use of T(2) weighted images. However, for detection of benign hepatic lesions and characterisation of hepatic lesions, the use of DWI was equivalent to the use of T(2) weighted images.

A pilot study of brain activation in children with trichotillomania during a visual-tactile symptom provocation task: a functional magnetic resonance imaging study

OBJECTIVES

In order to reveal the etiology and pathophysiology of trichotillomania (TTM), it is necessary to investigate which brain regions are involved in TTM, but limited knowledge exists regarding the neurobiology of TTM and the available functional neuroimaging studies of TTM are little. The purpose of the present study was to investigate the specific brain regions involved in the pathophysiology of TTM with symptom provocation task using functional magnetic resonance imaging (fMRI) for children and adolescents with TTM.

METHODS

Pediatric subjects who met the DSM-IV TR criteria for TTM (n=9) and age-, sex-, handedness-, IQ matched healthy controls (HC) (n=10), ages 9 to 17 years, were recruited for two fMRI experiments; symptom provocation of Visual Only (VO) and Visual and Tactile (VT). They were scanned while viewing two alternating blocks of symptom provocation (S) and neutral (N) movies.

RESULTS

Random effects between-group analysis revealed significant activation in left temporal cortex(including middle and superior temporal gyrus), dorsal posterior cingulate gyrus, and putamen for the contrast S>N in TTM subjects versus HC subjects during the VO session. And TTM subjects demonstrated higher activity in the precuneus and dorsal posterior cingulate gyrus to the contrast S>N during the VT session.

CONCLUSIONS

This study provided an objective whole-brain-based analysis that directed researchers to areas that were abnormal in TTM. Using the symptom provocation tasks, we found significant differences in regional brain function between pediatric TTM and HC subjects. However, in the face of modest statistical power, our preliminary findings in TTM need to be replicated in a larger sample. As the functional neuroanatomic circuits involved in TTM remain largely unexplored, future functional neuroimaging studies using other various paradigms may help investigate the neuroanatomic abnormalities of TTM.

Combination of high-resolution susceptibility-weighted imaging and the apparent diffusion coefficient: added value to brain tumour imaging and clinical feasibility of non-contrast MRI at 3 T

The purpose of this study was to determine the benefit of high-resolution susceptibility-weighted imaging and the apparent diffusion coefficient for brain tumour imaging, and to assess the clinical feasibility of using a non-contrast MR protocol at 3 T. 73 patients with intra-axial tumours were enrolled into the study. Two experienced neuroradiologists reviewed three MRI sessions: (i) a non-contrast protocol including high-resolution susceptibility-weighted images and apparent diffusion coefficient; (ii) a contrast protocol including MR perfusion images; and (iii) combined contrast and non-contrast protocols. The two observers categorised tumours as glial or non-glial tumours, and then subcategorised the gliomas into low-grade or high-grade tumours. For semi-quantitative analysis, a scoring system based on the degree of intra-tumoral susceptibility signals and the visual apparent diffusion coefficient was used. The two observers diagnosed accurate tumour pathology in 52 (71%) of 73 tumours in the first review, 55 (75%) of 73 tumours in the second review and 61 (84%) of 73 tumours in the third review. The addition of the non-contrast protocol to the contrast protocol significantly differentiated glioblastoma multiforme and metastatic tumours, which was not possible with the contrast protocol alone. The sensitivity, specificity, positive predictive value and negative predictive value for glioma grading with the non-contrast protocol were 83.2%, 100%, 100% and 79.3%, respectively. The addition of both high-resolution susceptibility-weighted imaging and the apparent diffusion coefficient improved the diagnostic performance of the contrast MR protocol for brain tumour imaging and could be feasible in selected patients who cannot tolerate a contrast agent.

Added value and diagnostic performance of intratumoral susceptibility signals in the differential diagnosis of solitary enhancing brain lesions: preliminary study

BACKGROUND AND PURPOSE

It has been reported that high-resolution susceptibility-weighted imaging (HR-SWI) is a promising tool for assessing brain tumor characterization noninvasively. The purpose of this study was to determine the added value and diagnostic performance of HR-SWI for differentiating solitary enhancing brain lesions (SELs) by assessing intratumoral susceptibility signals (ITSSs).

MATERIALS AND METHODS

Sixty-four consecutive patients with SELs, without previous surgery, were retrospectively reviewed. We performed 2 consensus reviews, by using conventional MR images alone and with adjunctive HR-SWI. We applied an ITSS grading system based on the degree of the ITSS. Then, we compared the presence and grade of the ITSSs among specific pathologic types of SELs.

RESULTS

Two observers diagnosed tumor pathology accurately in 43 (67%) of 64 SELs after reviewing the conventional images alone and 50 (78%) of 64 SELs after reviewing the adjunctive HR-SWI (P = .016, McNemar test). ITSSs were seen in 25 (100%) of 25 glioblastoma multiformes (GBMs), in 2 (40%) of 5 anaplastic astrocytomas, and in 11 (73%) of 15 metastatic tumors. Although the ITSSs were unable to distinguish between GBMs and solitary metastatic tumors, differentiation between GBMs and solitary metastatic tumors was achieved (P = .01) by using a high ITSS degree (grade 3). Moreover, the ITSSs could discriminate high-grade gliomas from lymphomas and nontumorous lesions with a specificity of 100% (P < .0001).

CONCLUSIONS

The use of ITSSs on HR-SWIs significantly improves the accuracy for the differential diagnosis of SELs compared with the use of conventional MR imaging alone.

Semiquantitative assessment of intratumoral susceptibility signals using non-contrast-enhanced high-field high-resolution susceptibility-weighted imaging in patients with gliomas: comparison with MR perfusion imaging

BACKGROUND AND PURPOSE

It has been reported that high-resolution susceptibility-weighted imaging (HR-SWI) may demonstrate brain tumor vascularity. We determined whether the degree of intratumoral susceptibility signal intensity (ITSS) on HR-SWI correlates with maximum relative cerebral blood volume (rCBVmax) and to compare its diagnostic accuracy for glioma grading with that of dynamic susceptibility contrast (DSC) perfusion MR imaging.

MATERIALS AND METHODS

Forty-one patients with diffuse astrocytomas underwent both non-contrast-enhanced HR-SWI and DSC at 3T. We correlated the degree and morphology of ITSS with rCBVmax within the same tumor segment. The degree of ITSS and rCBVmax were compared among 3 groups with different histopathologic grades. Spearman correlation coefficients were determined between the degree of ITSS, rCBVmax, and glioma grade. Receiver operating characteristic (ROC) curve analyses were performed to determine the diagnostic accuracy for glioma grading.

RESULTS

The degree of ITSS showed a significant correlation with the value of rCBVmax in the same tumor segments (r = 0.72, P < .0001). However, the areas of densely prominent ITSSs did not accurately correspond with those of rCBVmax. Spearman correlation coefficients between ITSS degree and glioma grade were 0.88 (95% confidence interval, 0.79-0.94). In the ROC curve analysis of histopathologic correlation by using the degree of ITSS, the optimal sensitivity, specificity, positive predictive value, and negative predictive value for determining a high-grade tumor were 85.2%, 92.9%, 95.8%, and 76.5%, respectively.

CONCLUSIONS

The degree of ITSS shows a significant correlation with the value of rCBVmax in the same tumor segments, and its diagnostic performance for glioma grading is comparable with that of DSC.

Diffusion tensor imaging of cingulum fibers in mild cognitive impairment and Alzheimer disease

BACKGROUND

Neuroimaging in mild cognitive impairment (MCI) and Alzheimer disease (AD) generally shows medial temporal lobe atrophy and diminished glucose metabolism and cerebral blood flow in the posterior cingulate gyrus. However, it is unclear whether these abnormalities also impact the cingulum fibers, which connect the medial temporal lobe and the posterior cingulate regions.

OBJECTIVE

To use diffusion tensor imaging (DTI), by measuring fractional anisotropy (FA), to test 1) if MCI and AD are associated with DTI abnormalities in the parahippocampal and posterior cingulate regions of the cingulum fibers; 2) if white matter abnormalities extend to the neocortical fiber connections in the corpus callosum (CC); 3) if DTI improves accuracy to separate AD and MCI from healthy aging vs structural MRI.

METHODS

DTI and structural MRI were preformed on 17 patients with AD, 17 with MCI, and 18 cognitively normal (CN) subjects.

RESULTS

FA of the cingulum fibers was significantly reduced in MCI, and even more in AD. FA was also significantly reduced in the splenium of the CC in AD, but not in MCI. Adding DTI to hippocampal volume significantly improved the accuracy to separate MCI and AD from CN.

CONCLUSION

Assessment of the cingulum fibers using diffusion tensor imaging may aid early diagnosis of Alzheimer disease.

Hypoperfusion in frontotemporal dementia and Alzheimer disease by arterial spin labeling MRI

OBJECTIVES

To test if arterial spin labeling (ASL) MRI could detect a pattern of hypoperfusion in frontotemporal dementia (FTD) vs cognitively normal (CN) control subjects; to determine the regional difference of perfusion between FTD and Alzheimer disease (AD); and to determine whether hypoperfusion in FTD correlates with cognitive impairment.

METHODS

We included 21 patients with FTD, 24 patients with AD, and 25 CN subjects in this cross-sectional MRI study. All subjects had MRI scans including T1-weighted structural images and ASL-MR images.

RESULTS

ASL-MRI detected a pattern of hypoperfusion in right frontal regions in patients with FTD vs CN subjects, similar to PET and SPECT. FTD had higher perfusion than AD in the parietal regions and posterior cingulate. Frontal hypoperfusion in FTD correlated with deficits in judgment and problem solving. Adding frontal perfusion to gray matter (GM) atrophy significantly improved the classification of FTD from normal aging to 74%, and adding parietal perfusion to GM atrophy significantly improved the classification of FTD from AD to 75%. Combining frontal and parietal lobe perfusion further improved the classification of FTD from AD to 87%.

CONCLUSION

Frontotemporal dementia and Alzheimer disease display different spatial distributions of hypoperfusion on arterial spin labeling MRI. With further development and evaluation, arterial spin labeling MRI could contribute to the differential diagnosis between frontotemporal dementia and Alzheimer disease.

The combination of X-ray-mediated radiosurgery and gene-mediated immunoprophylaxis for advanced intracerebral gliosarcomas in rats

Rats with advanced, imminently lethal, approximately 4 mm diameter, left-sided intracerebral 9L gliosarcoma (9LGS), a well characterized malignant tumor with some similarities to human high-grade astrocytomas, were used as a therapy model 14 days post-implantation of 10(4) cells. Such tumor-bearing rats die within two weeks (median, 6 days) thereafter if untreated. However, if these tumors are exposed on day 14 to 12-25 Gy of an electron-equilibrated 6 MV photon beam (radiosurgery), survival is extended about 5-6 fold to a median of 34 days, but long-term survival (> 1 year) is increased only to approximately 18%. Multiple subcutaneous inoculations of radiation-disabled 9LGS cells post-radiosurgery (immunoprophylaxis) extended lifespan and long-term (> 1 year) survival minimally (median, 37 days; 25%, respectively). In sharp contrast, radiosurgery followed by multiple subcutaneous inoculations of radiation-disabled 9LGS cells that had been transfected with granulocyte macrophage colony stimulating factor (GMCSF), a cytokine with demonstrated immune-enhancing properties (i.e. gene-mediated immunoprophylaxis, GMIMPR) increased long-term survival to approximately 67%. To our knowledge, these results are the first to show that the combination of photon radiosurgery and GMIMPR is effective for an advanced, imminently lethal brain tumor in a mammal. These data raise the possibility that GMIMPR following radiation therapy might prove effective for the treatment of some human malignant gliomas.

FAIR exempting separate T (1) measurement (FAIREST): a novel technique for online quantitative perfusion imaging and multi-contrast fMRI

A new pulse sequence, dubbed FAIR exempting separate T(1) measurement (FAIREST) in which a slice-selective saturation recovery acquisition is added in addition to the standard FAIR (flow-sensitive alternating inversion recovery) scheme, was developed for quantitative perfusion imaging and multi-contrast fMRI. The technique allows for clean separation between and thus simultaneous assessment of BOLD and perfusion effects, whereas quantitative cerebral blood flow (CBF) and tissue T(1) values are monitored online. Online CBF maps were obtained using the FAIREST technique and the measured CBF values were consistent with the off-line CBF maps obtained from using the FAIR technique in combination with a separate sequence for T(1) measurement. Finger tapping activation studies were carried out to demonstrate the applicability of the FAIREST technique in a typical fMRI setting for multi-contrast fMRI. The relative CBF and BOLD changes induced by finger-tapping were 75.1 +/- 18.3 and 1.8 +/- 0.4%, respectively, and the relative oxygen consumption rate change was 2.5 +/- 7.7%. The results from correlation of the T(1) maps with the activation images on a pixel-by-pixel basis show that the mean T(1) value of the CBF activation pixels is close to the T(1) of gray matter while the mean T(1) value of the BOLD activation pixels is close to the T(1) range of blood and cerebrospinal fluid.