The 5th International Conference on Biomedical Engineering and Biotechnology (ICBEB 2016)

01 The application value of three-dimensional rotational angiography of intracranial micro-aneurysms in diagnosis and treatment Shaoqing Wang, Xiancun Yang, Meixia Su, Qiang Liu Department of MRI, Shandong Medical Imaging Research Institute Affiliated to Shandong University, Jinan, Shandong, 250021, People's Republic of China; Department of Interventional Radiology, Shandong Provincial Hospital Affiliated to Shandong University, Jinan, Shandong, 250021, People's Republic of China Correspondence: Qiang Liu (2002md@163.com) BMC Medical Imaging 2016, 16(Suppl 1):01


Aims
Efficient improving contrast of cortical bone from surrounding long T2 tissues is important in ultarshort echo time (UTE) imaging. Methods UTE acquisition prepared by adiabatic inversion recovery were developed for this purpose. The effect of TI on cortical bone imaging was evaluated on mature bovine tibial mid-shafts using a 3-T clinical MR scanner. The imaging parameters were: TE/TR = 10 μs/300 ms, TI = 80,90,100,110,120,130,and 140ms, FA =45°, Bandwidth = ±62.5 kHz, FOV = 8cm, slice thickness = 7mm, NEX = 2, single slice.

Results
With TI = 90ms, excellent suppression of long T2 signals was achieved with the CNR cortical-muscle value of 13.49 ± 0.67, and the CNR cortical-marrrow value of 12.26 ± 0.86. Due to different T1s of muscle and fat, some residual signals from fat were presented. Therefore, the CNR marrow-muscle value was 1.24 ± 0.35. Furthermore, approximate 80% signals from muscle and fat were suppressed.

Conclusions
The 2D adiabatic inversion UTE sequence with a TI of 90 ms provided excellent contrast depiction of bovine cortical bone. Due to the T1 difference, muscle and fat longitudinal magnetizations cannot arrive to the null point at the same time. Therefore, simultaneous reduction of long T2 signals is complicated.
performance with an average accuracy of 98.80%, an average sensitivity of 99.50%, and an average specificity of 98.63%. The proposed "WE + KSVM" method is a promising brain tumor detection method for MRI scanning. Keywords brain tumor; detection; wavelet entropy; sequential minimal optimization; computer-aided diagnosis; radial basis function; cross validation; kernel support vector machine.

Aims
To investigate the dependence of geometric efficiency of a MDCT system on several exposure parameters such as tube voltage, collimation and pitch. Materials and methods Dose profiles in PMMA phantom for Siemens Definition Flash CT and GE Discovery CT750 HD were derived in helical mode using different tube voltages, collimations and pitches. Corresponding geometric efficiencies and weighted geometric efficiencies were calculated. Kruskal-Wallis test was performed to test the differences between weighted geometric efficiencies using different exposure parameters and the Spearman's correlation coefficient was calculated to determine the correlation between different exposure parameters and weighted geometric efficiencies.

Results
With larger collimation the weighted geometric efficiency can be improved by 30%, while combined with larger pitch the weighted geometric efficiency can be reached to about 70%. Weighted geometric efficiencies had positive correlation with beam collimation and pitch (p < 0.05) for both CT scanners, while there was no significant difference between weighted geometric efficiencies with different tube potentials (p > 0.05).

Conclusions
The decrease of geometric efficiency leads to the increase of patient radiation dose. It is necessary to improve the geometric efficiency and reduce the burden of patients by optimal setting beam collimation and pitch for CT scans. # Ning Mao, Xinnuan Mu contribute equally to this work. Aims To explore the changes of the white matter in adolescent depression by using method of Tract-Based Spatial Statistcs (TBSS).

Materials and methods
We have applied TBSS to 35 depressed adolescents and 40 matched control to exam WM microstructure. With TBSS, we have concluded the fractional anisotropy (FA), axial diffusivity (AD), radical diffusivity (RD) and mean diffusivity (MD) of adolescent patients with depression and controls.

Results
Research found unusual WM structure among adolescent depression. Our analysis showed that the FA values are lower (P < 0.01), the RD and MD values are elevated (P < 0.01), and the AD values are Invariant (P > 0.05) in the patients' body of the corpus callosum (CC). There is a contrary relationship between the severity of depression and FA values in the body of the CC(P < 0.01).

Conclusion
Our study showed that WM abnormalities are occurred in the pathophysiology of depression. What's more, our research suggested that these changes occurred in the early stages of the disease. Keywords Adolescent Depression; diffusion tensor imaging; white matter 08 A landmark-based approach for mid-sagittal plane detection in 3D brain MR images Ke Gan, Daisheng Luo College of Electronics and Information Engineering, Sichuan University, Chengdu, China Correspondence: Ke Gan (gankeonline@hotmail.com) BMC Medical Imaging 2016, 16(Suppl 1):08 Aims This paper presents a fully automated approach for the mid-sagittal plane (MSP) detection in 3D brain MR images. This method detects the MSP by accurately identifying highly-visible anatomical landmarks in the brain.

Materials and methods
The proposed method is landmark-based, this involves a training phase, which is performed once for a particular set of data, using some spatially aligned images with known anatomical landmark locations. The center points of the anterior commissure (AC), posterior commissure (PC) and midbrain-pons junction (MPJ) were manually delineated on the training images by an expert. In the detecting phase, the intensity of the testing image was normalized and transform into the same space as the training images. The image feature of AC, PC, MPJ obtained in the training stage were used to match the AC, PC, MPJ in the testing image. To accelerate the matching, the landmark detection was conducted in the neighborhood of the mean AC, PC, MPJ positions in the normalized space. An refinement procedure was carried out to further adjust the detected landmarks. Finally, the formulation of the 3D MSP equation was estimated by the detected landmarks.

Results
The proposed method was applied to 30 T1-weighted brain MR images. All testing results were visually inspected and judged to be correct without obvious error. The directional difference of plane normal (DDPN) between automated detection and manual labeling has been evaluated, the average DDPN we achieved was 2.83°.

Conclusions
The promising results indicate this method can be potentially useful in clinical applications.  Aims Medial image classification is a difficult task for its high similarity inter-class and low similarity intra-class. Traditional methods usually devise a single classifier. While, in this paper we focus on learning a multiple classifier for each type for medical image classification.

Materials and methods
The proposed method designs a boosted learning framework. First, an initial classifier is constructed according to the feature distribution of training image set. Then, more classifiers are trained in an iterative way. The overall performance can be enhanced successively. Moreover, the optimal weights can be gained on each individual classifier.

Results
The experiment uses the public available CT lung image dataset for evaluation. 379 lung CT images are contained in this dataset, which are collected by 50 different CT lung scans. The standard data is described by the instruction of an expert. Experimental evaluations show that the proposed method outperforms traditional methods with application to lung nodule CT image classification task. Conclusions This paper utilizes the boosted learning, combine multiple classifier for CT lung image classification. The proposed method exploits the feature representation distribution, and the experimental results are preferable.

10
Comparison between conventional and golden ratio based radial trajectories: an eddy currents study Jie Yang, Yanchun Zhu, Shuo Li, Nan Fu, Shaode Yu, Rongmao Li, Yaoqin Xie Institute of Biomedical and Health Engineering, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, 518055, China Correspondence: Yanchun Zhu (yc.zhu@siat.ac.cn) BMC Medical Imaging 2016, 16(Suppl 1):10 Aims Golden ratio based k-space trajectories are widely used in dynamic MRI since it provides approximate uniform k-space distribution. However, rapidly switching gradient induces eddy currents, generating artifacts in image. The golden ratio (GR) based radial strategy was compared with conventional radial strategy on 0.7T open superconducting MRI system.

Materials and methods
In conventional radial strategy, a constant angle increment Φuniform = 180°/ P between neighboring profiles. In GR based radial strategy, the azimuthal spacing is ΦGR = 180°/1.618 = 111.2°. First, a simulation was carried out to the comparison between ideal and net gradient of Shepp-Logan phantom with two radial strategies respectively. Second, a pure water phantom was sampled by Cartesian, conventional and GR based radial trajectory on the 0.7T open superconducting MRI system. Three orthogonal planes were acquired. SNR was compared between these three sampling trajectories, and images from Cartesian trajectory were used as reference.

Results
Result of simulation has illustrated that the impact of eddy currents on the ideal gradient with GR based radial strategy is more apparent. The result of a pure phantom shows that SNR values of both radial strategies (conventional: 40.76 GR:16.11) are far smaller than Cartesian strategy (145.15).

Conclusions
Eddy currents artifacts are more serious in GR based radial trajectory. Rapid switching gradient in GR based radial strategy induces more eddy currents than conventional radial strategy, which may limit the application of GR based trajectories especially in high magnet filed system.

Introduction
Accurate segmentation of Region of interest (Region of Interest, ROI) has an important place in medical image analysis, and still remains a challenge task because of the complex background and structure. Materials and methods This paper proposed a novel active contour model based on localizing region for ROI segmentation in capsule endoscopy images. Features in regions centered on an active contour were used to compute the local region descriptors. For calculating the local energies, the image was separated by the initial circular shape curve into two parts: interior and exterior. And then each local region is fitted with a model to optimize the energies.

Results
Experiments show that in term of the average over segmentation rate, the proposed method is 2.8%, whereas traditional snake and GVF snake model are 3.2% and 3%, respectively; In term of the average under segmentation rate, the proposed method is 2.4%, whereas traditional snake and GVF snake model are 2.9% and 2.6%, respectively. All results demonstrated the superior of the proposed method to other existing methods in ROI segmentation.

12
Accuracy and effectiveness of the respiratory self-gating signal in 3D cardiac cine MRI Shuo Li 1, 2 , Yanchun Zhu 2 , Jie Yang 2 , Song Gao 1 , Nan Fu 2 , Shaode Yu 2 , Yaoqin Xie 2 Aims Cardiac and respiratory self-gated free-breathing steady-state free procession (SSFP) has been proposed as an alternative to conventional SSFP for cardiac cine magnetic resonance imaging. In this technique, the acquired k-space data within the given respiratory gating windows are used for image reconstruction. Therefore, the accuracy of respiratory self-gating (RSG) signal is important.

Materials and methods
The self-gated free-breathing 3D SSFP technique was performed on a 1.5T GE HDx scanner. Twenty five healthy volunteers were included. The imaging parameters were: TR/TE = 3.5/1.3 ms, flip angle = 40°, bandwidth = ±125 kHz, slice thickness = 7 mm (no gap), number of slices = 12-14, number of profiles = 5000. RSG and respiratory bellow (RB) triggers were compared by correlation and t-test analyses. The percentage of respiratory signal intensity within gating windows was calculated.

Results
The respiratory cycle duration is 3314.7 ± 1072.6 ms. For all cases, the correlation coefficient between RSG and RB triggers is greater than 0.99, the P value of t-test is greater than 0.90. The percentage of RSG signal intensity within gating windows was 66.1 ± 4.1% compared with 60.1 ± 3.4% for RB.

Conclusions
There was an excellent correlation between RSG and RB triggers. There was no significant difference between two methods. RSG signal can well synchronize with RB signal and provide approximately the same respiratory cycle duration.

13
Changes in fiber bundles with aging: a tractography-based MRI study Yaping Wang 1,3 , Guixue Liu 1,3 , Wensheng Li 1,2,3 The aim of this work was to investigate changes in fiber bundles with aging by utilizing quantified diffusion magnetic resonance imaging (MRI).

Materials and methods
A total of 125 normal subjects were separated into 5 groups (group 1: 16-30 years old, n = 20; group 2: 31-45 years old, n = 34; group 3: 46-60 years old, n = 24; group 4: 61-75 years old, n = 22; group 5: 76-90 years old, n = 25). All subjects underwent diffusion tensor imaging (DTI) and T1-weighted MRI in a 3T scanner, and DTI Studio software was used to process all DTI data and for tracing fiber bundles. Statistics for the total fiber number of brain and for the fiber density (FD) of 3 regions of interest (ROIs), namely the corpus callosum, cingulate, mesencephalon were gathered and analyzed using SPSS software. Results Significant differences were observed in total fiber number among all age groups (p < 0.05). In group 1, a significant difference was found between the FD of left and right cingulate (p < 0.05). Significant differences were found in comparisons of the FD of left and right cingulate (p < 0.05), and the downward trend of the left cingulate was found to be faster than that of right cingulate. Furthermore, significant differences were found between the FD of corpus callosum and cingulate (p < 0.05).

Conclusions
Thus, we can use quantitative MR DTI to study changes in brain fiber bundles.
14 Ultrasonographic assessment of bony roof ratio in infant hip joints Changyu Tu (13969987078@163.com) Department of Ultrasound Diagnosis, Women and Children's Hospital of Linyi, Linyi City, Shandong 276001, China BMC Medical Imaging 2016, 16(Suppl 1):14 Aims This paper conducted research on a new method of ultrasound pediatric hip -bone top ratio measurement. Material and methods 390 cases of pediatric hip (hip) ultrasound examination were selected since March 2011 to August 2016 according to Graf method of measuring the size of the angle α. They were divided into three groups: Group 1, α angle ≥60°, 130 cases; Group 2, α angle <60°~≥43°, 130 cases; Group 3, α angle <43°, 130 cases. On the basis of Graf law, the ratio of the top bone was measured. We measure bony roof ratios in the following way. A tangent (referred as X axis) to the iliac bone was drawn across the transition point of iliac periosteum and perichondrium. A line vertical to X axis (referred as Y axis) was drawn across the vertex of acetabular cartilage roof. A vertical line to Y axis (line A) was drawn across the inferior boarder of acetabular labrum, and the other vertical line to Y axis was drawn from the inferior margin of ilium (the lowest point of ossificious iliac bone) (line B). The length of lines A and B was measured (cm) and the ratio between the length (A:B) was calculated as acetabular bony roof ratio (bony roof ratio).

Results
Group 1, α angle of 78.38°~60.00°, the top bone ratio 3.71 to 1.05; Group 2, α angle <60.00°~≥43.00°, the top bone ratio 1.21 to 0.54; Group 3, α angle <43.00°~30.00°, the top bone ratio from 0.58 to -0.36. Top bone ratio method classification criteria: I type, top-bone ratio > 1.2; II type, top-bone ratio of 1.2 to 0.6; eccentric, top bone ratio <0.6. Aims Dual-energy CT (DECT) has promising medical applications in differentiating bones and tissue. For full-view data, radiation dose is an important concern which makes few-view imaging research valuable. For few views, however, conventional filtered back-projection type methods fail to provide satisfying images, and direct decomposition is unstable due to noise boost. In order to obtain both high-quality CT and decomposition images for few views, this paper proposes a fast total variation (TV)-based image reconstruction algorithm and a practical image domain decomposition method for DECT.

Materials and methods
The reconstruction optimization problem, containing TV regularization term and data fidelity term, utilizes the pre-conditioned Chambolle-Pock method to design the algorithm which shows fast convergence. On the other hand, for noise suppression, the image decomposition uses the penalized weighted least square estimation, and applies the smoothness regularization term enforcing on estimation images. We implemented the proposed joint method on real DECT projections and compared it with state-of-the-art methods.

Results
The experiments on an anthropomorphic head phantom show that our methods have advantages on noise suppression and edge reservation, without blurring the fine structures in the sinus area. Furthermore, the proposed methods consume much less time than the compared iterative reconstruction algorithms. Conclusions This paper proposes a fast joint reconstruction-and-decomposition method for DECT imaging. Compared to the existing approaches, our method achieves better performance in reconstruction accuracy and decomposition quality. Aims For many years, without software the medical imaging tools like X-ray, CT, Ultrasound, and MRI which only work in a monochromatic fashion. In order to display the colorful tissue image by hardware, we invented a color converter circuit to transfer the monochromatic tissue image into the color image to broaden the diagnostic and therapeutic possibilities.

Materials and methods
In order to colorize the monochromatic tissue images, the tissue reflection light is separated into the tissue image light and the bioluminescence fluorescence by a dichroic. Both lights are detected by two photo diodes and converted into the color signals by the color converter circuit.

Results
The two conditions of color changing on the color monitor are discussed as the following:(1) The black color on the monitor means there is no backscattered light coming from the tissue. (2) When there is fluorescent coming from the tissue, we can see the different color compounded showing on the color monitor.

Conclusions
By checking the color changed area on the color monitor to see the distribution of fluorescence on the tissue image, we can make the disease checking easier and clear. Up to now, little research has been done in the area of monitoring the fragmentation progress during Extracorporeal Shock Wave Lithotripsy (ESWL) treatment or renal stones while no method has been widely suggested for clinical practice. Experience of technicians and doctors are key factors for efficient treatments. In this study the use of 3D ultrasound imaging has been used for the estimation of the fragmentation level of renal stones during ESWL.

Materials and methods
Generally, in the course of lithotripsy, fragmentation of the stones produce changes in the intensity and texture of the images. First, simulated fragmented stones were used in a gelatin phantom. Furthermore, pig kidneys containing plaster of Paris stone models were exposed to shock waves. Gray level cooccurrence matrix texture features (contrast, entropy ASM, IDM and homogeneity) were calculated in 3D regions of interest containing the stones.

Results
The calculated texture values from the 3D regions containing the stone fragments show progressive changes which relate to the stone fragmentation level. The results from the gelatin phantom experiments show agreement with the pig kidney experiments.

Conclusions
Three dimensional ultrasound imaging could be used for monitoring of the progress of shock wave treatments of renal stones.

18
Combining a support vector machine with a convolutional neural network for fMRI data classification Xiaolong Sun 1 , Juyoung Park 1 , Soyeun Kim 2 , Kyungtae Kang 1 Aims Functional magnetic resonance imaging (fMRI) facilitates brain research and it is found to be widely used in clinic. Brain tasks can be related to regions of neural activity by classification of fMRI data. The design of previous systems has primarily been focused on classifier performance, whereas we focus on reliability.

Materials and methods
We use a hybrid system in which a convolutional neural network (CNN) is combined with a support vector machine (SVM). The CNN extracts features from the fMRI image and the SVM classifier finds patterns of features. Converting the dimension of the image and retaining its significant features increases processing speed and reduces the effect of noise, while the use of a simplified CNN reduces training time.

Results
Our system achieves a classification accuracy of 99.5% on Haxby's 2001 fMRI dataset, which is superior to decision tree, random forest, neural network, K-nearest neighbor, support vector machine, AdaBoost, and Haxby's models. We contend that this makes it suitable for clinical applications. Conclusions Our hybrid system combines the advantages of SVMs and CNNs, which are both widely used for image recognition. The salient features extracted by our hybrid system do not have to be hand-coded, unlike those used by most existing classifiers. We found that the complexity of our hybrid model increased much less during the classification process than that of Haxby's model. results. To address this problem, we have proposed a novel spiking cortical model (SCM) based structural representation method for the accurate NMIR.

Materials and methods
The reference and floating images are input into the SCM to generate the firing mapping images (FMI). The weighted mean of all the differences among Tchebichef moments of image patches centered at each considered pixel and other pixels in a neighbourhood in the FMI is used as a local descriptor to represent the image structure. The similarity metric is computed as the sum of squared differences between structural descriptors for the two images. By combining free-form deformation (FFD) with L-BFGS-B optimization method, the similarity metric is optimized to produce the registered image.

Results
Extensive experiments have performed on MR and CT images from BrainWeb database and Atlas database as well as ten real prostate MR and ultrasound images. Experimental results demonstrate that the proposed method can produce more similar registration results to the reference images and provide smaller target registration errors than the NMIR methods based on the normalized mutual information, entropy images, Weber local descriptor (WLD) and modality independent neighbourhood descriptor (MIND).

Conclusions
The proposed SCM based registration method provides an effective means for the accurate NMIR due to its robustness to image noise and rotational invariance. Ultrasound imaging has become a prevalent examination method in prenatal diagnosis. The most important precondition of routine US examination is the accurate acquisition of fetal standard plane from ultrasound videos. However, the labor-intensive and subjective assessment is too time-consuming and less reliability, and thus the development of the automatic fetal facial standard plane (FFSP) recognition method is urgently needed. In this paper, we propose to recognize FFSP using very deep CNN architecture (16 layer). In addition, a transfer learning strategy combine with special data augmentation technique is adopted to overcome overfitting problem and improves the classification performance. Also, a relatively shallower CNN architecture (8 layer With the rapid development of technology, Cloud Computing turned out to be a reliable method to construct flexible and resilient regional medical imaging services platform. It will be cost-efficient and high-performance. Materials and methods By using various media with Cloud Computing, this paper describes the construction of regional medical imaging services platform and analyzes the needs of regional medical image sharing and cooperation and technology progress, and then designs a FC SAN and HDFS combination of medical imaging. Scalable distributed processing is the key to the platform. Storage regional services for SAAS based on the research goes for architecture and parallel medical imaging services.

Results
In consideration of the features of Hadoop, the requirement of medical imaging cloud platform and the basic structure of medical imaging, Could Computing is particularly designed for rapid large volumes of data (petabytes) storing and processing.

Conclusions
The reconstruction of medial cloud computing imaging services platform confronted great challenges in funds and technology with traditional technology. While newly-developed cloud computing and service model is a practicable approach to construct an economical, reliable and scalable regional medical imaging services platform.

22
Modelling and simulation of small world in memory network Lanhua Zhang 1* , Chengxin Yan 2 , Huihui Yang 1 , Baoliang Sun 3* Aims In order to model and simulate the small world topology in brain network on memory function, and explore the corresponding relationship between memory phenomena and functional characters.

Materials and methods
We design the deterministic algorithms to simulate the memory network of brain referring to the theories of graph, control and networks combing with the functional magnetic resonance imaging data.

Results
We simulate a memory network with evolution algorithms. By computing of network, the model has the small-world characters in clustering and average path length in accordance with the functional magnetic resonance imaging data results. Conclusions From computational model algorithm and memory phenomena, brain memory functional network also can be simulated with the same results of functional magnetic resonance imaging data. The method of cross subject research can provide a feasible way for the study of brain memory function network.
Disruption of functional connectivity is increasingly considered to be associated with Alzheimer's disease (AD) patient and that at high risk for AD. In this paper, altered functional connectivities are explored in longitudinal participants of 34 patients with EMCI, 23 patients with LMCI and 15 patients with AD, compared with 31 healthy control subjects. We evaluate the altered function connectivities with the progression of disease based on a priori defined 273 regions of interest. Different levels of analysis based on functional connectivity are explored. Many inter-network connectivities and itra-network connectivities are found to be impaired in both MCI group and AD group. The longitudinal alterations of functional connectivity within DMN (Default Mode Network) were correlated with variation in cognitive ability, and the SAL (Salience Network) as well as the interaction between DMN and SAL was disrupted in MCI group. Importantly, the longitudinal alternation of functional connectivity in the earlier stage is greater than that in the late stage, and the increase of altered network connectivity pattern is associated with the increase of disease severity. The altered connectivities are correlated significantly with both MMSE scores and ADAS-Cog. This study indicates that altered connectivity might be a potential biomarker of AD progression. Aims In order to optimize the initial center value for the medical image segmentation, an improved fuzzy clustering algorithm was proposed on the immune computation.

Materials and methods
The immunity-based improved fuzzy clustering algorithm searched the suitable initial center value quickly, increased the efficiency and accuracy, and avoided the blind local optimum. The antigen represented the grayscale image data, and the antibodies were generated at random to create the clusters. We designed a similar concentration factor and adjusted this parameter dynamically.

Results
The simulation results show that our improved algorithm can adaptively calculate the centers of the medical image clusters. The C-mean clustering algorithm used the initial cluster center (10, 80) and got the final threshold 103 in 6.1 seconds. Our proposed immune algorithm used the initial cluster center (65, 220), and got better final threshold 214 in 0.4 seconds.

Conclusions
We proposed an immune fuzzy clustering algorithm based on the improved preliminary optimum initial cluster centers, to avoid falling into the local minimization area and accelerate the searching speed. The lung image segmentation can be improved with our algorithm to help the doctors to analyze the lung disease better. Image denoising is always an active research topic in the field of computer vision. The denoising performance had been greatly improved by the prior state-of-the-art methods based on non-local self-similar (NSS) patches. However, the prior NSS patch-based methods usually smoothed the edges and useful structures. The main reason is that the sparsity of the NSS patches cannot be correctly represented. In this paper, we propose a sparsely augmented Lagrangian image denoising (SALID) model over NSS patches via adaptive dictionary. With the adaptive dictionary, the sparsity of the NSS patches can be represented more efficiently. The results of widely synthetic experiments demonstrate that, with the single and effective alternating directions method of multipliers (ADMM) to solve the SALID model, the proposed denoising method can obtain highly competitive denoising performance and high-quality images, even superior to other advanced denoising methods. Moreover, the extensively experimental results of clinical X-ray angiogram images further verify that our method can obtain high-quality visual images. Keywords: Angiogram image denoising; Non-local self-similar patch; Alternating directions method of multipliers Aims In order to detect whether there are structural and functional changes of central nervous system in pain processing in classical trigeminal neuralgia (CTN) patients.

Materials and methods
Twenty-two patients with CTN and twenty-one age and sex-matched healthy volunteers were recruited. Whole brain 3D T1-weighted images and resting-state functional MRI datasets were obtained with SIMENS 3.0T MRI scanner. Voxel-based morphometry (VBM) based on DARTEL was used to identify the differences of gray matter volume and the Regional Homogeneity (ReHo) method was used to compare the brain spontaneous activity differences between patients and healthy controls.

Results
Compared with the health controls, the CTN patient presented with decreased GM volume in several brain regions including the right inferior temporal gyrus, inferior frontal gyrus, amygdala, thalamus, precuneus, cingulate gyrus and bilateral superior temporal gyrus, para-hippocampus, as well as increased GM volume in right frontal gyrus. Decreased ReHo values are in the left temporal and parahippocampus, as well as increased ReHo values were noted in the bilateral thalamus and left parietal lobe between CTN patients and healthy controls.

Aims
This study seeks to determine what alterations of the cortical brain network constructed using original regional gray matter volume (ORGMV) can be revealed for normal aging.

Materials and methods
The subjects are acquired from OASIS database. IBASPM toolkit was used to perform ORGMV measurement. Pearson correlation was used to build the network after linear regression. Network analysis was done by BCT. The attributes are interregional correlations, small-world configurations, nodal and modular properties. Lastly, statistical analysis was applied to verify the significant alterations. Results 1) Decreased interregional correlation was found between the superior temporal pole and middle temporal pole in the right hemisphere, whereas increased cases occurred mostly in the frontal lobe between bilateral regions. 2) The distributions of hubs exhibited left-lateralized and right-lateralized in the young and aging group, separately. The fusiform gyrus and Rolandic operculum were identified as hubs for the young and aging groups, respectively. 3) Only one connector-module was found in the aging group, and the inter-module connections of one module in the aging group is relatively sparse.

Conclusions
To our knowledge, this study is the first to realize brain network construction by ORGMV. The findings suggest that it could enhance the understanding of the underlying physiology of normal aging and serve as a supplement approach to help exploring the mechanism of the human brain.

Aims
Abdominal viscera have its own particularity, so the choice of imaging guidance shows more importance in the process of 125-Iodine seed implantation. The goal of this study was to classify the advantages of ultrasound in the treatment of abdominal malignant tumors by 125-Iodine seed implantation. Materials and methods 20 patients were evaluated in this study and they accepted 125-Iodine seed implantation guided by ultrasound. This paper aims to analyze the ultrasound guidance influence factors and measures and make clear its advantages.

Results
Among all 20 patients, 4 cases were liver malignant tumors. General anesthesia combined with breathing controlled could reduce the activity of liver, achieve fast and accurate punctures and the 4 cases had no serious complications; 7 cases were pancreatic cancer. Using ultrasonic probe to squeeze the stomach could reduce or avoid injury caused by multiple punctures, and the 7 cases had no gastric bleeding and fistula; 9 cases were retroperitoneal tumors. Using ultrasonic probe to press abdomen could reduce or avoid bowel and mesentery injury, gastrointestinal hemorrhage and the rates of intestinal fistula and abdominal infection, and the 9 cases had no melena; Real-time guidance was equivalent to direct vision, the puncture process could be monitored and showed its security.

Conclusion
This study found that ultrasound guided 125-Iodine seed implantation in the treatment of abdominal malignant tumors could achieve accurate puncture, reduce the puncture injury safely and effectively. Keywords Ultrosound, Radioactive seed implantation, 125-Iodine, Abdominal malignant tumor 29 Medical images classification using Mapreduce based convolutional neural network and its application for big data processing Binquan Li 1 , Xiaohui Hu 2

Materials and methods
We take advantage of distributed computing and design the system on clusters. Different from traditional approach using single machine, we propose Mapreduce based CNN (MRCNN) significantly increasing training speed and reducing computation cost simultaneously. The feature vector is sent to Naïve Bayes classifier (NBC) for classifying interstitial lung disease and other medical HRCT. To avoid overfitting and local minima, we utilize genetic algorithm and Bayesian regularization (GABR) pre-training networks and initializing the weights. Furthermore, we design Mapreduce based NBC to increase efficiency of training classifier.

Results
Compared with other methods in past years, our method achieves benchmark performance. It's capable of enhancing recognition accuracy and suitable for medical imagery big data processing. Conclusions MRCNN is efficient for storage and processing of HRCT big data. Due to the powerful feature representation ability of CNN, distributed MRCNN framework can be applied to other medical imaging big data analytics. Aims This study seeks to develop an atlas-based beamformer approach of dynamic functional connectivity (FC) for resting-state MEG data, which is comparable with other imaging methods.

Materials and methods
The dynamic FC was calculated based on atlas with a slidingwindow method. We first segmented the MEG data with a 2seconds window, then performed a whole brain LCMV beamformer scan. The source with maximum power of each ROI was selected to compute the FC matrices. To avoid the problem of volume conduction, PLI was used to quantify the phase synchronization. Lastly, a k-means algorithm was applied to the windowed correlation matrices, each cluster centroid putatively reflects a stable FC state. We studied 16 healthy subjects of the Human Connectome Project (HCP) database.

Results
From the k-means analysis, we got 7 FC states. State 1 accounts for about 52% over all windows, which has the homologous pattern with the mean connectivity matrix. States 2-7 occur less frequently (ranging from 7.1% to 24.4%), but represent substantial different patterns with the mean connectivity.

Conclusions
In this work we proposed a novel framework to study the temporal variability of FC for resting-state MEG data through an atlas-based beamformer approach. Preliminary experimental results showed that even in the resting-state recording, FC changes among serval distinct states.

Results
The original spatial resolution of FLIR C2 is 0.37 cycles/mm (at 10% of MTF) and the noise is 0.53% at 23°C and 0.60% at 50°C. For test thermal images, TD achieved the best performance among three image enhancement methods on both edge recovery and noise suppression. Quantitatively, the TD method improves the spatial resolution of FLIR C2 to 0.68 cycles/mm and the noise slightly to 0.51% at 23°C and 0.58% at 50°C.

Conclusions
The TV-based method can significantly improve the resolution (84%) of FLIR C2 and enable temperature monitoring of the millimeter-size photothermal treatment spot with less than one percent variation. Beamformed RF signals reconstruction in ultrasound imaging using sparse FFT Yifei Liu 1 , Mingyue Ding 2 , Yanhong Zhou 3 , Huihong Gong 1 , Wei Peng 1 Aims Sparse fast Fourier transform (SFFT) is a class of sub-linear time algorithms for estimating the k largest frequency coefficients of a signal which is sparse in the frequency domain. As with the compressive sensing, it is possible for SFFT to acquire sparse signals far below the Nyquist rate, but SFFT is considerably simpler because the original signal can be simply restored by inverse fast Fourier transform. In addition, SFFT can get better reconstructed signal quality from fewer frequency coefficients. The purpose of this paper was to study the SFFT reconstruction for ultrasound beamformed signals.

Materials and methods
We used Field II ultrasound program with a linear array probe and cyst phantom to simulate the beamformed signals. The SFFT algorithms utilized 3% of signal length as the sparsity parameter k. A total of 50 rebuilt radio frequency (RF) scan lines were applied to assess the impact of SFFT reconstruction on associated beamformed image.

Results
The SFFT reconstruction quality was evaluated by keeping 10%-50% candidate frequency coefficients of the original beamformed signal. The results show SFFT can restore the normalized original beamformed scan lines with a mean absolute error range of [0.007-0.015]. The normalized root mean square error ranges for associated beamformed image is [0.052-0.095]. Conclusions This paper shows the feasibility of SFFT for reconstructing the ultrasound beamformed signal. The future works include the reconstruction of raw channel RF signals and extending the study in 3D ultrasound imaging system. Aims In order to solve the slow convergence and early mature problems in searching for the image threshold with the traditional genetic algorithm and Otsu method, the genetic algorithm was improved for better image segmentation in the medical imaging of the lungs.

Materials and methods
According to the different evolution generation and the individual fitness, this improved genetic algorithm adjusted the strategies of elite selection and genetic operator dynamically, so it not only can speed up the convergence and the diversity of community, but also can get the best image segmentation threshold finally in a stable range. We used the dynamic crossover probability with the formula p c ¼ k1 1þe αÂgen þ λ , and we set the parameters with k 1 = 1, α = 0.055, λ = 0.3.

Results
The simulations were implemented with Matlab and some medical images of the lungs. The lung images showed that the left half pulmonary part was not healthy, the health condition of the right half part was better than the left one. Our approach got the better result and cost less computing time than both the simple Otsu method and the traditional genetic algorithm.

Conclusions
The proposed algorithm effectively segmented the original medical image, which made the target and background well separated. Our approach not only minimized the noise disturbance, but also enhanced the medical images of the left and right lungs, which was conducive to the subsequent medical diagnosis. The segmentation of overlapped cervical cells from Pap smears is one of the most challenging problems in medical image processing field. We present a novel automated system for the cervical cell segmentation of cytoplasms and nuclei from multi-cell images. First, this system employed a graph cuts method including two algorithms, namely unsupervised k-means initialization and max-flow/min-cut optimization for scene segmentation. The segmented clumps were considered as foreground regions. Then, we used Voronoi diagrams to divide every clump of overlapping cells into individual non-overlapped regions, each containing one nucleus. Finally, to improve the overlapped segmentation, each individual cell in a clump was fitted by a minimum enclosing ellipse and the overlapped region was replaced by the corresponding area in this ellipse. This overlapped area and the connected freelying region were combined to form a region of one complete cell. The experiments were conducted on two publicly released databsets downloaded from websites of University of Adelaide. The quantitative performance presents the average Dice Coefficient (DC) higher than 0.85. According to the explanation of evaluation metric in databsets, the "good" segmentation is evaluated with the DC > 0.7. Thus, the result of our proposed system outperforms 0.7 and achieves the state-of-art performance.

Keywords
Medical image processing, cervical smear image, overlapping segmentation, graph cuts, Voronoi diagrams. Intensity inhomogeneity correction algorithm for brain MRI image segmentation Wei Liu 1 , ZhiJun Gao 2 , LiSha Tan  Aims To overcome the difficulty in accurate segmentation of brain magnetic resonance imaging (MRI), and reduce noise, partial volume effect and intensity inhomogeneity in MRI images. Materials and methods A brain MRI image segmentation strategy considering the intensity inhomogeneity is discussed in this paper. We improve basic fuzzy Cmeans clustering algorithm and propose a new algorithm using anisotropic diffusion for image segmentation. The correction of intensity inhomogeneity is also studied to be implemented to the actual work of MRI image segmentation.

Results
The improved algorithm has better restraint effect on intensity inhomogeneity and noise in brain MRI images, so the segmentation accuracy is enhanced. It can also well estimate the information of intensity inhomogeneity.

Conclusions
Our scheme can effectively estimate the intensity inhomogeneity information of images. We can get clearer segmentation images by removing intensity inhomogeneity, using the estimation of intensity inhomogeneity when performing image segmentation.

37
A voxel-based approach for sulci detection in 3D brain MR images Ke Gan, Daisheng Luo College of Electronics and Information Engineering, Sichuan University, Chengdu, China Correspondence: Ke Gan (gankeonline@hotmail.com) BMC Medical Imaging 2016, 16(Suppl 1):37 Aims This paper describes a fully automated method for sulci detection in 3D Magnetic Resonance (MR) images of human brain.

Materials and methods
To detect the brain sulci in 3D MR images, several consecutive steps were adopted. Brain image was automated segmented into white matter, gray matter and cerebrospinal fluid (CSF). A topological correction was carried out to make sure there is no direct contact between white matter and CSF. The gray matter/white matter boundary was extracted, and the Euclidean distance transformation was calculated from the boundary to the external regions. The gradient vector field of the distance map was calculated and diffused. Based on the diffused vector field, a divergence map was calculated. Candidate locations for the brain sulci were extracted from the divergence map by a thresholding procedure. Finally, morphological correction, including 3D connected component analysis and morphological operations, were applied to refine the detection results.

Results
This algorithm was implemented in C++ on a Windows platform and applied to label the brain sulci in 10 T1-weighted 3D brain MR images. Visual inspection indicates brain sulci in all testing images are correctly identified. Additionally, two major sulci (precentral sulcus and postcentral sulcus) in each hemisphere of the brain were manually labeled by a trained expert. And the overlap ratio between the manual labeling and the results of automated detection was calculated. Six imaging slices were selected from each testing image for the comparison. The average overlap ratio is 0.987±0.021 (the standard deviation).

Conclusions
The experimental results indicate this approach can be applied to extract the brain sulci in volumetric brain MR images.

Aims
To observe the configuration of the circle of Willis (CW), measure the size of posterior communicating artery (PcomA). Materials and methods 73 cases CTA data of head and neck (without any lesion in the CW and its surrounding structures) were selected from our hospital. 3D images of CW were obtained, then CW was observed as two classifications and PcomA were measured.

Conclusion
Most of CW is the types of complete anterior circulation with incomplete posterior or developmental dysplasia. Diameter of PcomA is 1-2mm, and no PcomA is 53.18%.