Search

Millions of parcels and pieces of luggage are scanned daily for threat detection at border control or high‐security buildings. Currently, the process is manually operated by security agents and it is slow and time‐consuming. The automation of this process will lift the burden from the security agents and will allow larger volumes of items to be scanned. The authors consider the problem of automatic threat detection, in particular firearms, in X‐rays. To achieve this goal, they propose a hybrid algorithm that combines two well‐established image segmentation algorithms into a two‐step clustering method. The first step is a semi‐supervised spectral clustering algorithm at the image level, which classifies whole images into benign or containing a threat. The images classified as threatening from the first step proceed to the second stage, where a variational image segmentation algorithm performs clustering at the pixel level to locate the threat if it exists. The hybrid algorithm is designed to scale‐up the processing of hundreds of images, in comparison to the academic literature where only a handful images are used for demonstration. Numerical experiments establish that the combination of two different algorithms produces better results than using individual algorithms. [ABSTRACT FROM AUTHOR]

Annotations are critical for machine learning and developing computer aided diagnosis (CAD) algorithms. Good performance of CAD is critical to their adoption, which generally rely on training with a wide variety of annotated data. However, a vast amount of medical data is either unlabeled or annotated only at the image-level. This poses a problem for exploring data driven approaches like deep learning for CAD. In this paper, we propose a novel crowdsourcing and synthetic image generation for training deep neural net-based lesion detection. The noisy nature of crowdsourced annotations is overcome by assigning a reliability factor for crowd subjects based on their performance and requiring region of interest markings from the crowd. A generative adversarial network-based solution is proposed to generate synthetic images with lesions to control the overall severity level of the disease. We demonstrate the reliability of the crowdsourced annotations and synthetic images by presenting a solution for training the deep neural network (DNN) with data drawn from a heterogeneous mixture of annotations. Experimental results obtained for hard exudate detection from retinal images show that training with refined crowdsourced data/synthetic images is effective as detection performance in terms of sensitivity improves by 25%/27% over training with just expert-markings.

Global abnormal events form unique and distinct motion characteristics and category of anomalies at image level rather than pixel level with less complexity compared to local abnormal events. However, traditional anomaly detection approaches focused more on pixel‐level feature extraction from foreground pixels and combine global and local anomaly detection in a single algorithm with equal degree of computational complexity. In this paper, we propose a novel framework for global anomaly detection via block‐level feature extraction using context location (CL) and motion‐rich STVs (MRSTVs). The histogram of optical flow orientation and motion magnitude features from spatio‐temporal volumes (STVs) are used as global feature descriptor to capture motion characteristics of normal and abnormal events. Simple and cost‐effective one‐class SVM classifier is employed to learn normal behaviour from MRSTVs during training and detect abnormal STVs from test data. Thereafter, a spatio‐temporal post‐processing technique detects frame‐level abnormal behaviour and reduces false alarm rate. We define CL to detect abnormal behaviour in an unexpected region. The proposed approach omits pixel‐level feature extraction and background modelling by considering MRSTVs, thus enhances detection rate and reduces computational complexity. We have conducted experiments on widely used UMN and PETS2009 datasets to compare the performance of proposed approach with existing methods. [ABSTRACT FROM AUTHOR]

Background: Amyloid β-protein (Aβ) plaque deposition is an important prevention and treatment target for Alzheimer's disease (AD). As a noninvasive, nonradioactive and highly cost-effective clinical imaging method, magnetic resonance imaging (MRI) is the perfect imaging technology for the clinical diagnosis of AD, but it cannot display the plaque deposition directly. This paper resolves this problem based on pixel feature selection algorithms at the image level. Methods and Results: Firstly, the brain region was segmented from mouse model brain MR images. Secondly, the pixels in the segmented brain region were extracted as a feature vector (features). Thirdly, feature selection was conducted on the extracted features, and the optimal feature subset was obtained. Fourthly, the various optimal feature subsets were obtained by repeating the same processing above. Fifthly, based on the optimal feature subsets, the final optimal feature subset was obtained by voting mechanism. Finally, using the final optimal selected features, the corresponding pixels on the MR images could be found and marked to show the information about Aβ plaque deposition. The MR images and brain histological image slices of twenty-two model mice were used in the experiments. Four feature selection algorithms were used on the MR images and six kinds of classification experiments are conducted, thereby choosing a pixel feature selection algorithm for further study. The experimental results showed that by using the pixel features selected by the algorithms in this paper, the best classification accuracy between early AD and control slides could be as high as 80 %. The selected and marked MR pixels could show information of Aβ plaque deposition without missing most of the Aβ plaque deposition compared with brain histological slice images. The hit rate is over than 90 %. Conclusions: According to the experimental results, the proposed detection algorithm of the Aβ plaque deposition based on MR pixel feature selection algorithm is effective. The proposed algorithm can detect the information of the Aβ plaque deposition on MR images and the information can be useful for improving the classification accuracy as assistant MR biomarker. Besides, these findings firstly show the feasibility of detection of the Aβ plaque deposition on MR images and provide reference method for interested relevant researchers in public. [ABSTRACT FROM AUTHOR]

Background: The purpose of the present study was to determine whether the image levels of the distal femur affected the measurement of the tibial tubercle-trochlear groove (TT-TG) distance. Methods: Thirty sets of computer tomography (CT) images and 30 sets of MR images of the knee were evaluated. The TT-TG distance was quantified at multiple image levels in 1.5-mm increments, covering the proximodistal range of the trochlear groove. The CT measurement was based on osseous landmarks; the magnetic resonance imaging (MRI) measurement was based on cartilaginous and osseous landmarks. Results: The average TT-TG distances measured with CT, with MRI based on cartilaginous landmarks, and with MRI based on osseous landmarks were 15.74 mm (SD 3.83 mm), 12.8 mm (SD 5.67 mm), and 12.36 mm (SD 5.58 mm), respectively. No significant difference was found across image levels in the CT measurement and the MRI measurement upon osseous landmarks (P = 0.64, = 0.11); yet, the difference was significant in the MRI measurement upon cartilaginous landmarks (P < 0.01). Large deviation was found between levels in individual P subjects in all the three sorts of measurement. The proximal levels were the most variable, while the mid levels were the least variable. Conclusions: Measurements of the TT-TG distance are not identical across the levels of the distal femur. Cautions should be taken when specific image slices were selected for evaluation. [ABSTRACT FROM AUTHOR]

The popularity of iris biometric has grown considerably over the past two to three years. It has resulted in the development of a large number of new iris encoding and processing algorithms. Since there are no publicly available large-scale and even medium-size data bases, neither of the newly designed algorithms has undergone extensive testing. The designers claim exclusively high recognition performance when the algorithms are tested on a small amount of data. In a large-scale setting, systems are yet to be tested. Since the issues of security and privacy slow down the speed of collecting and publishing iris data, an optional solution to the problem of algorithm testing is to synthetically generate a large scale data base of iris images. In this work, we describe a model-based method to generate iris images and evaluate the performance of synthetic irises by using a traditional Gabor filter-based iris recognition system. A comprehensive comparison of synthetic and real data is performed at three levels of processing: 1) image level, 2) texture level, and 3) decision level. A sensitivity analysis is performed to conclude on the importance of various parameters involved in generating iris images [ABSTRACT FROM PUBLISHER]