X-ray has been widely used in medical diagnosis and other fields with the non-destructive merit. Recently, the cold cathode X-ray source based on field emission, which are different from traditional hot cathode X-ray source, have overcome the limitations of traditional X-ray source due to their small size, addressable, and integrated characteristics, making CT imaging methods can be further developed. Especially X-ray sources using ZnO nanowires as cold cathodes can densely integrate a large number of X-ray sources into flat panel devices. When we use this flat X-ray source to light up multiple sources to scan an object at the same time, small source spacing will cause inevitable projection aliasing. Therefore, we propose an image restoration algorithm for projection aliasing, and use the method of projection rearrangement to remove projection aliasing. In addition, we estimated the spectrum of the flat-panel X-ray source by solving the spectral model parameters from the attenuation data of materials of different thicknesses, and used the spectral information to simulate the effect of the proposed image restoration method on the new photon counting detector. The results of simulation experiments show the effectiveness of this method and the potential of this new imaging mode.
Genomic and proteomic techniques have provided new research ideas for the early diagnosis of diseases such as cleft lip and palate (CLP). However, a significant challenge is finding the best set of biomarkers for the disease’s clinical diagnosis in massive or high dimensions. Existing studies have focused on refinements and combinations of feature selection methods to improve classification accuracy without interpreting the results. In this paper, seven feature selection methods are firstly compared on eight publicly available genomics microarray data. Moreover, five classification models and two pre-processing methods are aslo compared to find the most appropriate method for processing the microarray data. The method is finally validated on a CLP dataset. The results indicate that the methods described herein can achieve high classification accuracy and provide better feature interpretation.
Alzheimer's disease (AD), the most common type of dementia, is a severe neurodegenerative disorder. Disease progression modeling (DPM) using longitudinal data is a challenging machine learning task. In this paper, we consider the problem of predicting disease progression measured by the cognitive scores and selecting biomarkers predictive of the progression. Most previous work on predicting AD progression ignore the issue of missing data. Missing data poses a major difficulty for modeling longitudinal data since most statistical models assume feature-complete data. We proposed and applied a minimal recurrent neural network model with skip to data from The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge, comprising longitudinal data of 1677 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI). We have performed extensive experiments to demonstrate the effectiveness of the proposed model using data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
Numerous studies showed autism spectrum disorder (ASD) is a socio-cognitive and neurodevelopmental disorder together with disturbed static brain activity. However, little is known about alterations of dynamic brain activity and its associations with social behavior in ASD. In this study, the dynamic amplitude of low-frequency fluctuation (dALFF) based on sliding-window technique was applied to the resting-state functional magnetic resonance imaging data of 91 adolescents with ASD and 122 age-matched typically developing controls (TDC group), which obtained from the Autism Brain Imaging Data Exchange (ABIDE) repository. We found that the increased dALFF variance in the right middle temporal gyrus (MTG) was detected in adolescent ASD relative to TDC, which is positively correlated with communication deficits. Our findings reveal that atypical dynamic brain activity in MTG is associated with social deficits in ASD. Results also highlight the crucial role of MTG in the social-cognitive deficits of ASD and provide a novel insight into understanding dynamic pathophysiologic mechanisms underlying ASD.
Computer Assisted Sperm Analysis (CASA) plays a crucial role in the diagnosis and treatment of male reproductive health. In recent years, with the development of computer industry, more and more effective algorithms and techniques have been applied in this field to help CASA obtain more objective and quantitative analysis results rapidly. As target detection is an important part in image processing which is the basic technique of CASA and also includes pre-processing, feature extraction and tracking, this survey comprehensively analyses studies focus on target detection in CASA since 1988.
The novel corona-virus disease (COVID-19) pandemic has caused a major outbreak in more than 200 countries around the world, leading to a severe impact on the health and life of many people globally. Accurate and rapid diagnosis of COVID-19 suspected cases plays a crucial role in timely quarantine and medical treatment. In this work, we present a deep learning based framework for automatic segmentation of pathologic COVID-19 associated tissue areas from clinical CT images available from a dataset with 108 cases in China. More specifically, we present an effective multi-scale feature fusion U-Net equipped with ResNet architecture and a deep supervision mechanism to increase the network's capacity for learning richer representations of infected tissue. Our experiments demonstrate that our model achieves an average dice score (0.674), sensitivity (0.733) and Precision (0.714) on the dataset. The experimental results have indicated the effectiveness of the proposed improvements and the potential of our proposed method for real clinical practice.
The traditional image registration methods use the iterative optimization to search the most optimal transformation parameter. However, they are all time-consuming. Therefore, we proposed a fast and unsupervised T1/T2 weighted breast images registration model based on the convolutional neural network. The proposed model estimate the spatial transformations from pairs of T1/T2 weighted breast images , and used the transformations to warp the moving images. There is no supervised information such as the ground truth of the deformation fields should be provided in this model. The registration of T1/T2 weighted breast MRI images can assist doctors to get better diagnosis results and provide better data resource for lesion detection and image prediction. We have evaluated the proposed model and the experimental results show that the model is robust and effective.
Electromagnetic tomography (EMT) is a process tomography based on the principle of electromagnetic induction. It has the advantages of non-intrusive, non-contact and low harm. In the process of image reconstruction, the sensitivity matrix reflects the influence of tiny disturbance on the measured value, which is an important index of image quality. Base on the traditional "O" sensor array structure in this article, Comsol Multiphysics finite element simulation software is used to divide the measured field into several tiny elements. A simulated perturbation method is used to investigate the effect of the number of EMT sensors' midline coils on sensitivity. The sensitivity matrix of sensor arrays with 4 coils ,8 coils ,12 coils ,16 coils and 20 coils is compared by using the method of separation variables. The sensitivity matrix of different coils is imaged by Matlab software. Through comparative analysis, it is concluded that the sensitivity band shape of the coil in the same position is same in different models. All of them are distributed along the arc type between the excitation and detection coils, which presents the middle strong and gradually weakens to the outside. With the increase of the number of coils, the sensitivity band is widened, but the sensitivity value of the central part is slightly weakened. According to this characteristic, it lays a foundation for the practical application of EMT technology and image reconstruction.
In this paper, we propose and validate a coarse-to-fine kidney segmentation method from Computed Tomography (CT) images, i.e., predicting a coarse label based on the entire image and a fine label based on the coarse segmentation and cropped image patches. A key difference between the two stages lies in how input images were preprocessed. For the coarse segmentation, each 2D CT slice was normalized to be of the same image size (but possible different pixel size), and for the fine segmentation, each 2D CT slice was first resampled to be of the same pixel size and then cropped to be of the same image size. In other words, the image inputs to the coarse segmentation were 2D CT slices of the same image size whereas those to the fine segmentation were 2D CT patches of the same image size as well as the same pixel size. In addition, we designed an abnormality detection method based on component analysis between two stages and used another 2D convolutional neural network to correct the abnormality regions. A total of 168 CT images were used to train the proposed framework and evaluations were conducted qualitatively on another 42 testing images. The proposed method showed promising results and achieved an average DSC of 94.53 % on the testing data.
Functional magnetic resonance imaging under naturalistic paradigm (NfMRI) is gaining increasing attraction, as it offers an ecologically-valid condition to understand brain function in real life. Characterizing the hierarchical organization of brain function while taking the nature of fMRI activities under naturalistic condition into account has been a critical issue in identifying naturalistic functional networks. Recent studies have made efforts on characterizing the brain's hierarchical organizations from fMRI data via a variety of deep learning models. However, most of those models have ignored the properties of group-wise consistency and inter-subject difference in brain function under naturalistic paradigm. Another critical issue is how to determine the optimal neural architecture of deep learning models, as manual design of neural architecture is time-consuming and less reliable. To tackle these problems, we proposed a two-stage deep belief network (DBN) with neural architecture search (NAS) combined framework (two-stage NAS-DBN framework) to model both the group-consistent and individual-specific naturalistic functional brain networks. Our results demonstrated that the optimized DBN-based framework can characterize meaningful group-wise and individual-level naturalistic functional networks, which reflected the hierarchical organization of brain function and the properties of brain functional activities under naturalistic paradigm.
Manual delineation of stroke lesion is a time-consuming and tedious task. This paper applies the three-phase level set method[1] to the segmentation of stroke lesion from DWI images. First, we select a region of interest (ROI) around the stroke lesion, and then two random initial level set functions are generated in the ROI. By iterating the level set functions, a three-phase segmentation result is obtained as a partition of the ROI into three subregions with one subregion being the segmented stroke lesion. The experimental results show that, the dice coefficient of our method reached 0.84 and the hausdorff distance reached 5.89. Compared with a level set method and a deep learning method, our method has better segmentation performance, which can obtain expected effect of segmentation.
Generative adversarial networks (GANs) are powerful generative models that have led to breakthroughs in image generation. The aim of our work is to adapt the idea of GANs to the graph data based on fMRI, which can solve the issue of limited data and improve the classification performance in the domain of the Autism spectrum disorder (ASD) diagnosis. In this work, we present a data augmentation method that generates synthetic graph data with the proposed graph generation model named α-GCNGAN, which is able to handle the intrinsic challenge of generating brain network with considering flexible context-structure. Extensive experiments on Autism Brain Imaging Data Exchange (ABIDE) dataset demonstrate the effectiveness and generalizability of α-GCNGAN.
In recent years, the number of lung cancer patients has continued to increase. In the process of detecting lung cancer, accurate segmentation of lung parenchyma plays a key role. In this paper, we proposed a method of lung parenchyma segmentation based on FPN++Mask R-CNN neural network model. The model improved original Mask R-CNN networks and optimized the structure of FPN (Feature Pyramid Networks), which is the feature extraction model of Mask R-CNN, by expanding the scale and level of FPN to fuse and extract more picture feature information from different levels. The experimental results show that compared with original Mask R-CNN models, FPN++Mask R-CNN demonstrates better segmentation results.
As the development of ultrasound technology, ultrasound scanning 3D imaging has become one of the common routine examinations in clinic. However, the shortage of ultrasound doctor resources and the high dependence on the operator's technology limit the ability and application prospect of ultrasound imaging assisted clinical diagnosis. Although the volume probe has been developed, it`s expensive and the scanning region is limited in this paper we propose a novel flexible probe assisted control system which can realize the automatic uniform motion of ultrasound probe, so as to reduce the difficulty of operation. We designed a slide table controlled by a micro stepper motor, and fixed the ultrasonic probe on the slide table. In this way, when the motor is started, the slide table goes forward at a constant speed and the ultrasonic probe can also move forward at a constant speed, so as to capture high-quality scanning images for 3D reconstruction. As a result of the micro stepper motor, our device is different from the traditional mechanical scanning device which is difficult to operate and carry, our system is not only reduces the difficulty of operation, but also brings us the portability of the device. Besides, compared with volume probe, our cost is much lower, due to our slide table can fix different normal probes which are cheaper than volume. As a result, our novel system has a wide range of application prospects.
Colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second most lethal cancer in the world. The segmentation and staging of tumours in CT images are important for CRC diagnosis and treatment. Using deep learning to fulfil medical image diagnosis is a promising approach. Most current methods solve segmentation and staging tasks separately, while ignoring the associated information between tasks. Multi-task deep learning (MTL) can solve the two tasks simultaneously. In this study, we proposed a CRC_MTL framework applying multi-task deep learning to complete two tasks jointly for analysing CRC tumours in abdominal CT images. The CRC_MTL framework not only improves the accuracy of the classification task from 91.40% to 93.52% but also improves the dice coefficient of the segmentation task from 0.6900 to 0.7379.
research-article Microscopic Image Augmentation Using an Enhanced WGAN Share on Authors: Hao Xu Northeastern University, PR China Northeastern University, PR ChinaSearch about this author , Chen Li Northeastern University, China Northeastern University, ChinaSearch about this author , Jinghua Zhang Northeastern University, PR China Northeastern University, PR ChinaSearch about this author , Zihan Li Northeastern University, PR China Northeastern University, PR ChinaSearch about this author , Changhao Sun Northeastern University, PR China Northeastern University, PR ChinaSearch about this author , Xin Zhao Northeastern University, PR China Northeastern University, PR ChinaSearch about this author Authors Info & Affiliations ISICDM 2020: The Fourth International Symposium on Image Computing and Digital MedicineDecember 2020 Pages 40–45https://doi.org/10.1145/3451421.3451431Published:05 December 2020 0citation0DownloadsMetricsTotal Citations0Total Downloads0Last 12 Months0Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access
Brain structures segmentation in Magnetic Resonance Imaging (MRI) is a challenging task because of the large anatomical variability in both shape and size between individuals. In this work, an atlas based level set method is proposed for brain tissues segmentation. First, a skull stripping method is utilized to remove the skull region of brain. A registration algorithm is then applied to obtain the initial six brain structure labels (thalamus, hippocampus, amygdala, putamen, pallidum, caudate) with a manually labeled atlas, which can be used as the initial level set contours. The final label is achieved by minimizing the energy function which consists of three terms: an image data term, a length regularization term and an intensity constrained term. Experimental results demonstrate that our method can obtain satisfactory results compared with other state-of-art methods.
In this paper, we propose and validate a novel neural network named Lesion Guided Network (LGN) for automatic diagnosis of Diabetic Retinopathy (DR) from fundus images. RetinaNet is first adopted and trained on a coarsely-annotated dataset for rough lesion detection. Lesion-Aware Module (LAM) in LGN is proposed to highlight regions of interest in fundus images utilizing the coarse lesion maps. Then, the outputs of LAM are fed into a covolutional neural network (CNN) for DR identification. The proposed method is evaluated on a private dataset consisting of 4465 fundus images. Experimental results demonstrate the superiority of the proposed LGN, achieving comparable performance with ophthalmologists.
Perfusion CT imaging is commonly used for the rapid assessment of patients presenting with symptoms of acute stroke. Maps of perfusion parameters such as cerebral blood volume (CBV), cerebral blood flow (CBF), and mean transit time (MTT) derived from the scan data provide crucial information for stroke diagnosis and treatment decisions. Most vendors implement singular value decomposition (SVD)-based methods on their scanners to calculate these parameters. However, SVD-based method is known to have issues of improperly handling the imperfect scan. For example, increasing the acquisition interval or decreasing the scan duration may introduce a bias in the estimated perfusion parameters. In this work, we propose a Bayesian inference algorithm, which can tolerate the imperfect scan conditions better than conventional method and is able to derive the uncertainty of a given perfusion parameter. We apply the variational technique to the inference problem, which becomes an expectation-maximization problem. The probability distribution (with Gaussian mean and variance) of each estimated parameter can be obtained. We perform evaluations in simulation studies both with full and incomplete data. The proposed method can obtain much less bias in estimation than the conventional method, and additionally providing the degree of the uncertainty in measurement.
In the rat’s liver cancer chemotherapy model, it is significant to quantify the process of the time-dependent release of any anticancer drug, for which an essential component is to extract drug-diffused vessels and analyze their topology. In this work, we proposed a novel pipeline for segmenting vessels and extracting their skeletons for subsequent topological analyses. First, we used UNet to obtain coarse vessel segmentation as an initial result. Due to the vital importance of the vascular topology, we manually corrected the coarse segmentation to a minimum degree. Then, level-set was used to further refine the vessel segmentation. Finally, we extracted the vascular skeleton to analyze its topological structure. Experimental results identified the robust performance of the proposed pipeline.
