three-dimensional NAWM mask from the image voxels with the highest probability to pertain to the WM cluster.
Finally, the lesion- ﬁlling process is achieved as follows: for each axial slice composing the three-dimensional image, we compute the mean and standard deviation of the signal intensity of NAWM tissue. Axial sampling is motivated because after testing the sampling procedure on the coronal, axial and sagittal planes, we found that the best results were obtained when we sampled the axial plane. This was due to the fact that using the axial plane reduced the variability of possible existing WM intensities, when compared to coronal and sagittal sampling. The Fuzzy-C-means approach used to estimate the tissue probabilities is a simple method which in fact does not take into account neither spatial nor neighboring information, and hyper-intense signal intensities such as residual parts of the eyes or the neck produced in the skull- stripping process can bias signi ﬁcantly the clusters. The risk of adding these parts into the WM distribution is minimized in the axial plane be- cause we are reducing it to a certain slice where lesionvolume is usually lower than that in central slices. The computed mean and standard de- viation values are used to generate a normal distribution with mean equal to the computed NAWM mean intensity and standard deviation equal to half of the computed NAWM standard deviation. Standard de- viation is always ﬁxed to half of the WM mean independently of the dataset used. This value was chosen empirically with the aim of balancing the accuracy of the method with both 1.5 and 3 T images. Although a speci ﬁc tuning of this parameter could provide a better performance on certain cases, we decided to ﬁx it avoiding therefore the number of parameters to tune. Lesion voxel intensities from the current image slice are replaced by random values of the generated distribution. The procedure is repeated until all image slices are completed.
To measure cross-sectional differences and changes over time in GM volumes, accurate segmentation methods must be used. A variety of different approaches to braintissue segmentation has been described in the literature. Few algorithms rely solely on image intensity,  because these approaches are overly sensitive to image artifacts such as radiofrequencyinhomogeneity, and aliasing, and cannot adequately account for overlapping intensity distributions across structures. Therefore, to improve segmentation accuracy, most tissue segmentation algorithms combine intensity information with other techniques, such as the use of a priori anatomic information [15, 16] or edge information through deformable contours[17,18, 19]. Intensity information is analyzed differently in each approach, including Gaussian mixture models [20, 21, 22, 23], discriminate analysis ), k-nearest neighbor classification , and fuzzy c-means clustering [26, 27, 28, 29, 30]. The use of multiple images has significant advantages over a single image because the different contrasts can be enhanced between tissues. For example, fluid attenuated inversion recovery (FLAIR) images have desirable contrast between MS lesions and the normal- appearing braintissue and can be combined with other images to obtain gray/whitematter segmentation .
of indifferent medical images of tumors is motivated by the need for high precision when it comes to human life. In addition, IT assistance is required by medical institutions in that it could improve man's results in a domain where false negative cases should be at a very low level. It has been shown that double reading of medical images could lead to better detection of tumors. The implicit cost of Butte's dual reading is very high and for this reason good software to help humans in medical institutions is of great interest today. Conventional methods of surveillance and diagnosis of diseases are based on the detection of the presence of particular features by a human observer . Due to the high number of patients in intensive care units and the need to constantly observe these conditions, several automated diagnostic techniques have been developed in recent years to try and solve this problem. These techniques work by transforming the most qualitative diagnostic criteria into a more objective quantitative classification problem in this project. We propose the automated classification of magnetic resonance imaging using some prior knowledge such as pixel intensity and some anatomical features. There are currently no widely accepted methods, so automatic and reliable methods for tumor detection are of great need and interest. The PNN application in the classification of MR image data is not yet fully utilized. These include grouping and classification techniques especially for MR image problems with large data and power consumption and power times manually. Therefore, comprehensive understanding of recognition, classification or grouping techniques is essential for the development of neural network systems, particularly in medical problems. The segmentation of braintissue in gray matter, whitematter, and cancer in medical images is not only of great interest for the serial follow-up of
correlation was 0.96 for BaMoS compared to 0.88 for EMS- C, 0.94 for LST-WML and 0.90 for TOADS. Application of BaMoS to larger clinical datasets would however be needed to properly assess the relevance of the detected lesion burden in terms of disease staging and progression. As the TLL is insuficient to evaluate segmentation accuracy, lesion shape, localization accuracy and overlap   , the eight segmentation evaluation measures defined in Table I were used to assess the automated segmentations and better analyse the origin of the segmentation errors. The study of the lesion count version of these measures would also be of interest for a more complete evaluation of the methods. When compared to the validation on clinical data, the main advantage of using synthetic images is the availability of a ground truth. However, the validation using the Brain- Web dataset was limited by multiple factors as previously underlined in . As only one phantom is available, no statistical analysis is possible. Also, synthetic images cannot be considered truly realistic. Moreover the range of lesion load is limited compared to the amount that can be found in clinical cases. Thus, the BrainWeb data was used here to test the algorithmic stability to different imaging modal- ity combinations, different degrees of image quality with varying noise level and intensity inhomogeneity. BaMoS was more stable across noise level when compared to the other methods especially TOADS and EMS. Naturally, the decrease in TPR with image quality was most important in the mildest case since subtle and small lesions are more affected by the noise level than more prominent lesions as observed in the severe case. Comparing three versions of the proposed methodology: BaMoS-static, BaMoS-NoCov
the same buffer for 24 hours and processed for paraffin embedding. Serial sections through the hippocampus (two 5-μm sections per slide, 100 μm apart) were stained with Masson’s trichrome. Digital virtual slides obtained with Aperio Scanscope CS-1 scanner were used for ex- tensive computer assisted morphometry in a Spectrum image analysis system (Aperio Technology Inc., 1700 Leider Lane, Buffalo Grove, IL, 60089, USA). Scanscope software and associated algorithms were applied for measurements of lesionvolume and the count of dead or viable neurons in the impact zone, penumbra and relevant area of the contralateral hemisphere control (internal control) as described by Krajewska and col- leagues . Whole brains were perfused with 4% parafor- maldehyde, cryoprotected with 30% sucrose and frozen for cryostat sectioning in optimal cutting temperature embedding media. Free floating sections (50 μm) were washed in phosphate-buffered saline, blocked and in- cubated overnight with primary antibodies followed by species-specific secondary antibodies. Species-specific fluroconjugated Alexa® (3175 Staley Road, Grand Island, NY, 14072, USA) secondary antibodies were used at a 1:500 dilution with DAPI in 10% goat blocking solu- tion. Sections were incubated for 1 to 2 hours at room temperature, gently rotating. We have previously charac- terized and optimized our immunofluorescence protocols for GFAP (glial fibrillary acidic protein), Iba1 (ionized calcium-binding adapter molecule 1) and MAP2 (micro- tubule associated protein 2) as previously described [19,20,24,25]. Incubation with 10% goat and no primary antibodies, with and without secondary antibodies, served as controls samples for these experiments. Coverslips or brain sections were mounted with an anti-fade solution and imaged; when appropriate, matched exposures were obtained. All other images were exposure and saturation optimized. All quantitation was done using NIH Image J.
Other automated tract segmentation tools have been developed using a range of strategies, such as incorporating prior information about nearby anatomical landmarks [ 7 ], or clustering whole-brain streamline sets in some feature space [ 8 , 9 ]. Tractography typically cannot be used for segmentation without manual or automatic refinement, however, due to the accumulation of errors along pathways [ 10 , 11 ], and the use of more sophisticated diffusion models, has led to more false positives [ 12 ]. Manual placement of regions of interest (ROIs) or seed points is a common and effective approach [ 13 ], although this method is very time-consuming and operator-dependent, for large studies. Automatic ROI or seed registration from an atlas might suffer from registration errors [ 6 ], particularly for datasets from the upper and lower extremes of age that might not be appropriately represented by current atlases [ 14 ]. Information used for the initialization of tractography should allow for the changes in anatomy due to age, and the PNT priors that are used for the current work seek to achieve this. A similar PNT approach has also been tested for the segmentation of tracts in infants [ 15 ].
Significant clusters obtained from both the WM and the GM were analyzed to evaluate the differences among the WML- VaD, WML-VCIND, and HC groups. Then, comparisons of significant clusters of the region of interest (ROI) were carried out. In order to clarify the correlations among all the predictor variable differences, GMD, WM volumes, and the cognitions, the multiple linear regression analysis was employed to evaluate the critical predictor variables, white and GM clusters, age, sex, and the levels of education. First, each outcome variable of edu- cation, hypertension, and age were regressed in this study. The relationship between WM volume and GMD was also tested.
Iron concentrations of the thalami were determined by using syngo.via software (Siemens, Erlangen, Germany). First, QSM maps were fused to the 3D T2 FLAIR images for improved ana- tomic delineation. Then, with the Volume-of-Interest tool, the upper and lower borders of the thalami were manually delineated by a board-certified radiologist with subspecialty certification in neuroradiology (G.C.C.) (Fig 1). The Contour tool then identi- fied the appropriate margins for the thalami on the intervening slices in semiautomated fashion, and the Nudge tool was used to exclude major veins that could confound the QSM values, partic- ularly the internal cerebral veins and basal veins of Rosenthal. As a zero reference, a small circular ROI was also placed in the CSF in the atrium of each lateral ventricle, adjacent to the thalamus. The relative susceptibility within the thalami was calculated by sub- tracting the mean susceptibility in the thalami from that in the CSF ROI.
Age-related whitematter changes first occur in the frontal lobe and the precentral and postcentral sulci. These brain areas are associated with memory, cognition, and sensation. As one ages, the temporal lobe, especially the hippocampus, begins to show age-related changes, leading to impaired memory. Deeper brain structures like the brain stem that are related to basic life activities do not exhibit apparent age-related changes, or age-related changes develop much later in life. The relation between whitematter changes and neuropsychiatric disorders has been pursued by investigators. For example, whitematter changes have been documented during transition of the prodromal phase of schizophrenia. 29 Wright et al inves-
al.  proposed a CNN architecture that considered multi-scale patches and explicit location features while training, and later was extended to consider non-uniform patch sampling . Their best performing architecture shares a similar design with the architecture proposed by Kamnitsas et al. [38, 39], in which it trained independent paths of convolutional layers for each scale. Using multi-resolution inputs [39, 27, 28] can increase the field of view with smaller feature maps, while also allowing more non-linearities (more layers) to be used at higher resolution, both of which are desired properties. However, down-sampling patches has the drawback that valuable information is being discarded before any processing is done, and since filters learned by the first few layers of CNNs tend to be basic feature detectors, e.g. lines or curves, different paths risk capturing redundant information. Furthermore, although convolutions performed in 3D as in  intuitively make sense for 3D volumetric images, FLAIR image acquisitions are actually often acquired as 2D images with large slice thickness and then stacked into a 3D volume. Further to this, gold standard annotations, such as those generated by trained radiologists (e.g. WMH delineation or Fazekas scores) are usually derived by assessing images slice by slice. Thus, as pointed out by Ghafoorian et al. , 3D convolutions for FLAIR MR image segmentation are in fact less intuitive. Some other works on CNN segmentation which are relevant to our work, though not on brainlesion segmentation, include Long et al.  and Ron- neberger et al. . Long et al.  proposed to segment natural images using a fully convolutional network that supplemented the output of a grad- ually contracting network with features from several of its levels of contrac- tion through up-sampling. Similar to , Ronneberger et al.  used a U-shaped architecture (U-net) to segment microscopical cell images. The architecture symmetrically combined a contracting and expanding path via feature concatenations, in which up-sampling operations were realized with trainable kernels (deconvolution or transposed convolution). Both of these networks form the foundation of the architecture later proposed in this work. Challenges. There are several challenges being held on brainlesion segmen- tation in recent years. For instance, the MS lesion segmentation challenge 2008 (http://www.ia.unc.edu/MSseg/) had the goal of the direct compari- son of different 3D MS lesion segmentation techniques. Data used in this
measurement was assessed and the high RI probably reflected the fluctuation of this index. As the study included newborns with hypoxic-ischemic cerebral injury, the CBF must have been initially low, with a high RI on Doppler ultrasound. This assumption is also valid for the presence of hemorrhage among five (16.1%) newborns with a low RI and two (10.5%) with normal RI. This finding indicates the necessity of serial RI measurements for the diagnosis of CBF fluctuations, as only one abnormal RI result is not a sign of intraventricular hemorrhage; however, the occurrence of at least one high RI result within the first 72 hours of life indicates a higher probability for this complication. Nevertheless, since cranial ultrasound has low sensitivity for the diagnosis of mild cerebral hemorrhage and WML (50% and 58%, respectively), 25 the ultrasound
Siemens Symphony 1.5T. Brain MR imaging scans were obtained on a Symphony 1.5T scanner (Siemens, Erlangen, Germany) by using a proprietary 3D magnetization-prepared rapid acquisition of gradi- ent echo (3D MPRAGE) protocol. Images were obtained from the vertex of the skull to the foramen magnum and from the occipital poles to the temporal poles. Specifications for 3D MPRAGE were the following: 120 contiguous coronal sections with a 1.5-mm gap in thickness; section interval, 0.75 mm; TR, 2190 ms; TE, 4.38 ms; TI, 1100 ms; FA, 15°; NEX, 1; matrix, 256 ⫻ 256; FOV, 260 mm; band- width, 130 Hz/pixel; acquisition time, 9 minutes; phase-encoding direction, right to left. MTA was evaluated by atrophy rating of the hippocampus, entorhinal cortex, and the perirhinal cortex on a coro- nal section intersecting the mamillary bodies.
The gray and whitematter segmentation of brain image is vital in identifying disorders and treatment planning in the field of medicine. In this work we extracted gray matter and whitematter using two well-known clustering algorithms- k-Means and Fuzzy c-Means algorithms. In both cases we performed intensity based and statistical feature based clustering. In statistical features we have used mean, variance, standard deviation, kurtosis, skewness and range. Among them only mean feature produces the prominent results. Experimental results shows that the segmentation using statistical based (mean) clustering gives improved performance metric and improved classification accuracy rather than intensity based segmentation. Our work investigated features based on first order statistical parameters that give very less number of distinguishable features for classification of MR images. In the future, we can use second order statistical features for the better segmentation of GM and WM in MR images.
Several conclusions were derived from this study. Seven parameters, diameter, aspect ratio, current, pulse duration, voltage, area of tissue affected and impedance, were investigated for their relationships to the VTS. When modeled as the independent variable, the impedance and area of tissue affected increase in a direct linear relationship to the VTS while the voltage, current, electrode diameter and pulse duration show an inverse linear relationship with VTS. This study presents three mathematical models which accurately predict the VTS for electrode diameters reaching the micrometer level. The multivariate models presented in this study demonstrate the strong correlation between the VTS with the electrode diameter, pulse duration and applied voltage. The predicted values for the three equations were within the normal average distribution line ranging from 95-110%. The accuracy with which the VTS was predicted using the new mathematical equations will enable the researchers to generate preliminary data without performing actual experiments. Therefore, this study will be useful in gaining insights before performing clinical testing and in the design of the experiments.
measure of changes in macroscopic neural activity patterns that underlie cognition and behaviour (Hutchison et al., 2013). Given this, another future aim is to use dynamic functional metrics to see how WMH impacts temporal network dynamics.
Future efforts will also be directed at investigating how reliable functional and structural patterns relate to more specific abilities and processes that comprise fluid and crystallized cognition. This can be accomplished by examining the relationships between associations of whitematter and functional connectivity in relation to scores on subtests of fluid and crystallized intelligence (e.g., processing speed, executive functions, episodic memory). These methods can be further applied to mapping shifts in the entire neuropsychological profile in the face of age- related whitematter disruptions across older adults in different age cohorts (i.e., young, middle, and late adulthood).
Mobile phone dependence (MPD) is a behavioral addiction that has become an increasing public mental health issue. While previous research has explored some of the factors that may predict MPD, the underlying neural mechanisms of MPD have not been investigated yet. The current study aimed to explore the microstructural variations associated with MPD as measured with functional Magnetic Resonance Imaging (fMRI). Gray mattervolume (GMV) and whitematter (WM) integrity [four indices: fractional anisotropy (FA); mean diffusivity (MD); axial diffusivity (AD); and radial diffusivity (RD)] were calculated via voxel-based morphometry (VBM) and tract-based spatial statistics (TBSS) analysis, respectively. Sixty-eight college students (42 female) were enrolled and separated into two groups [MPD group, N = 34; control group (CG), N = 34] based on Mobile Phone Addiction Index (MPAI) scale score. Trait impulsivity was also measured using the Barratt Impulsiveness Scale (BIS-11). In light of underlying trait impulsivity, results revealed decreased GMV in the MPD group relative to controls in regions such as the right superior frontal gyrus (sFG), right inferior frontal gyrus (iFG), and bilateral thalamus (Thal). In the MPD group, GMV in the above mentioned regions was negatively correlated with scores on the MPAI. Results also showed significantly less FA and AD measures of WM integrity in the MPD group relative to controls in bilateral hippocampal cingulum bundle fibers (CgH). Additionally, in the MPD group, FA of the CgH was also negatively correlated with scores on the MPAI. These findings provide the first morphological evidence of altered brain structure with mobile phone overuse, and may help to better understand the neural mechanisms of MPD in relation to other behavioral and substance addiction disorders.
Two published trials have looked at MRI outcomes following vitamin D intervention in patients with symptomatic knee OA, though in only one of these, the Australian VIDEO study, did the authors look at synovitis . In this trial participants were rando- mised to vitamin D therapy (50,000 IU/monthly) or placebo. In a post-hoc analysis the authors reported a mean between-group difference in effusion-synovitis of − 1.94 mL (95% CI -3.54 to − 0.33) at 2-years follow-up with the mean increase in effusion-synovitis volume over follow-up significantly less in the vitamin D therapy group (0.26 mL, 95% CI − 0.82 to 1.34) compared to the placebo group (2.20 mL, 95% CI 1.01 to 3.38) . The data however, were obtained using non-contrast images – such images are less sensitive in distinguishing effusion from underlying synovitis. In the Australian VIDEO study there was evidence to suggest that the effect on effusion-synovitis may have been more marked in those with low vitamin D levels; i.e. those who were consistently sufficient (> 50 nmol/L) at both 3 and 24 months follow-up had a sig- nificantly smaller increase in mean effusion-synovitis volume compared to those who were consistently insuffi- cient (≤50 nmol/L) (− 2.5 mL, 95% CI -4.7 to − 0.2) . Because of relatively small numbers of those with low vitamin D levels, we were unable to confirm or refute the findings of the Australian study in terms of a potential moderating effect of vitamin D status on treatment effect.
onates, WM signal abnormalities are referred to as “periventricu- lar whitematter injuries of prematurity.” Because tissue contrast comparisons showed higher contrast with synthetic PSIR images, we expected the periventricular WM lesions to appear more prominently on synthetic PSIR than on FSPGR images. However, periventricular WM lesions were more readily and sensitively de- tected using FSPGR rather than synthetic PSIR imaging. We spec- ulate that this finding is most likely due to higher spatial resolu- tion in 3D-FSPGR compared with 2D synthetic PSIR sequences. This is because the mean cranial-to-caudal diameter of lesions that were not detectable with synthetic PSIR (1.57 mm) was smaller than the slice thickness of 2D synthetic images (3 mm) but larger than that of 3D FSPGR (1 mm). In addition, the number of lesions detected with synthetic T1WI and PSIR imaging were the same. Another possible reason is the pathologic characteristics of punctate WM lesions, which are different from normal braintissue. The high T1 signal intensities of punctate WM lesions in- FIG 3. A 7-day-old neonate who underwent brain imaging due to an
projection pathways. The superior longitudinal fasciculus showed wide- spread alterations with increased BMI in water diffusivity, R1 and PD*. The SLF is a major intra-hemispheric ﬁber tract connecting the posterior part of the brain with the prefrontal cortex. Different measures of whitematter microstructure affected separate components of the SLF. Axial and mean diffusivity increased with BMI in the occipital part, while R1 decreased with BMI in the frontal operculum and increased PD* was observed in the temporal part. All SLF components terminate in the dorsolateral prefrontal cortex, a prominent brain region highly relevant for decision making processes and executive function ( Makris et al., 2005 ). Moreover, recent anatomical as well as functional connectivity studies revealed an obesity-associated imbalance between reward and executive brain networks. A dissociable pattern of structural connectiv- ity was observed, using deterministic tractography, showing increased ﬁber tract density between reward network regions and decreased ﬁber tract density between prefrontal regions ( Gupta et al., 2015 ). In addition, functional connectivity brain networks of the limbic and prefrontal system are especially vulnerable to increased body weight ( Garcia-Garcia et al., 2013; Kullmann et al., 2012, 2013; Marques- Iturria et al., 2015 ). The prominent decrease in WM integrity of the MFB/ATR and SLF in obese adults could contribute to over-eating via the rich projections to and from the prefrontal cortex.
FIG 2. Thalamus volumemeasurements in patients with INCL. Due to lack of visible boundaries of the thalamus (secondary to disease-re- lated alterations in intrinsic contrast), an ellipsoidal approximation to the volume was made on the basis of brain surface landmarks. Thala- mus volumes were out of the normal range by the time of our earliest measurements and further decreased with time. The equation for the best curve ﬁt is shown. We did not detect a difference between INCL in boys and girls (P ⫽ .98). The normal curve reﬂects thalamus volumes measured on 23 healthy volunteers 1.1–9.6 years of age participating in other studies at our institution; for the healthy volunteers, we did not ﬁnd a statistical difference between boys and girls (P ⫽ .89) or be- tween right and left (P ⫽ .86) (dark line ⫽ mean, light lines ⫽ ⫾2 SDs). For comparison, the onset ages of major clinical ﬁndings (mean and 95% CI) observed in our patients are plotted below the volume curve. A indicates developmental regression; B, cerebral atrophy noted in clinical MR imaging report; C, myoclonic jerks and seizures; D, loss of vision; E, deceleration of head growth; F, isoelectric visual-evoked potentials; and G, isoelectric EEG or electroretinogram.