Search

Anomaly detection is a very effective method to extract useful information from abundant data. Most existing anomaly detection methods are based on normal region or some specific algorithms, which ignore the fact that many actual datasets are mainly imbalanced, resulting in not function properly or effectively in practical, especially in the medical field. On the other hand, imbalanced dataset is also a frequently encountered problem in the learning of neural network because the lack of data in a minority class may lead to uneven classification accuracy. In this paper, inspired by these observations, a novel anomaly detection approach by using classical echo state network (ESN), a brain-inspired neural computing model, is presented. The entire dataset of the proposed method obeys an extremely imbalanced distribution, that is, anomalies are much rarer than normal data. And the training dataset has only the normal data. When the ESN is well trained, the parameters in ESN are the memory of normal data. If the normal data are added into the well-trained network, the error between the input data and the corresponding output is smaller compared with the error between abnormal input data and its corresponding output. Then anomaly behavior is detected if the error between the input data and the corresponding predictive value exceeds a certain threshold. Different from setting an invariable threshold arbitrarily for all of the data, the threshold value used in the proposed method is determined from the analysis of information theory and can be adjust adaptively according to different datasets. Experiments on abnormal heart rate detection are conducted to demonstrate and verify the effectiveness of the proposed detection algorithm and theory.

Electron tomography in the scanning transmission electron microscope (STEM) can be performed on sections of stained plastic-embedded tissues or cells of 1 to 2 micrometer thickness without effects of chromatic aberration because there are no imaging lenses after the specimen. By using a small STEM probe convergence angle of 1-2 mrad the geometrical broadening of the probe is restricted, which enables a spatial resolution of a few nanometers. Furthermore, by using an axial bright-field detector instead of the standard high-angle annular dark-field detector, image blurring due to multiple elastic scattering can be reduced in the lower part of the specimen. Here, we have applied STEM tomography to elucidate the 3D ultrastructure of human blood platelets, which are small anucleate blood cells that aggregate to seal leaks at sites of vascular injury and are important in the pathology of atherosclerosis and other diseases. Of particular interest are the morphological changes that occur in alpha-granules, which contain important proteins released when platelets are activated. Axial bright-field STEM electron tomographic tilt series were acquired at an accelerating voltage of 300 kV from 1.5-micrometer thick sections of platelets that had been prepared by rapid freezing and freeze-substitution; and the tomograms were reconstructed from dual-axis tilt series. The tomographic reconstructions revealed changes in ultrastructure that occurred on platelet activation including release of alpha granules through channels connecting to the plasma membrane. The research was supported by the intramural program of the National Institute of Biomedical Imaging and Bioengineering, and the research in the Storrie laboratory was supported in part by NIH grant R01 HL119393.

Serial block face scanning electron microscopy (SBF-SEM) provides nanoscale 3D ultrastructure of entire cells and tissue volumes. In SBF-SEM, an ultramicrotome built into the SEM specimen stage successively removes thin sections from a plastic-embedded, heavy metal-stained specimen. After each cut, the freshly exposed block face is imaged at a low incident electron energy using a backscattered electron detector to provide 3D ultrastructure with a resolution of approximately 5 nm in the plane of the block face and around 25 nm in the perpendicular z-direction, as limited by the slice thickness. We have explored the feasibility of improving the z-resolution in SBF-SEM by recording images at multiple primary beam energies, thus sampling different depths below the block surface.A linear relationship was found between the depth of test structures, generated by Monte Carlo simulations, and the ratio of backscattered image intensities recorded at primary beam energies between 1.4 keV and 6.8 keV. This enabled us to reconstruct the 3D model within a 25-nm surface layer at a z-resolution of around 5 nm. We used a Zeiss Sigma-VP SEM equipped with a Gatan 3View SBF system to acquire 3D data from a specimen consisting of gold spheres embedded in carbon. Experiments were also performed on embedded blocks of stained biological tissues.Although damage of the block under electron irradiation limits the signal to noise ratio, the use of multiple primary beam energies, coupled with a physics-based Monte Carlo model, provides the possibility of obtaining cellular ultrastructure at nearly isotropic 3D spatial resolution.

Serial block face scanning electron microscopy (SBF-SEM) provides nanoscale 3D ultrastructure of tissue samples up to several hundred micrometers in size. In SBF-SEM, an ultramicrotome built into the SEM specimen stage successively removes thin sections from a plastic-embedded, heavy metal-stained specimen. After each cut, the freshly exposed block face is imaged at a low incident electron energy using the backscattered electron signal, which is sensitive to heavy atoms in the sample. Although the x-y resolution in the plane of the block face is approximately 5 nm, the resolution along the z-axis in SBF-SEM is limited by the minimum slice thickness of around 25 nm. We have explored the feasibility of improving the z-resolution in SBF-SEM by recording images at more than one primary beam energy, thus sampling different depths below the block surface. We used Monte Carlo simulations of SEM images from an epoxy block containing 5-nm diameter carbon spheres stained with 14% osmium positioned at different depths, as a model for small biological structures. A linear relationship was found between the depth of the spheres and the ratio of backscattered signals at primary beam energies of 1.4 keV and 6.8 keV, which allowed us to generate 3D tomograms with a depth resolution of around 5 nm. Experiments are in progress to test this technique using a Zeiss Sigma-VP SEM equipped with a Gatan 3View SBF system. Sub-surface SBF-SEM could potentially match focused ion beam (FIB) SEM in terms of z-resolution, but with the added advantage of providing higher throughput and larger tissue volumes. The research was supported by the intramural program of NIBIB.