Yinpeng Jin

Contrast enhancement by multi-scale adaptive histogram equalization

An approach for contrast enhancement utilizing multi-scale analysis is introduced. Sub-band coefficients were modified by the method of adaptive histogram equalization. To achieve optimal contrast enhancement, the sizes of sub-regions were chosen with consideration to the support of the analysis filters. The enhanced images provided subtle details of tissues that are only visible with tedious contrast/brightness windowing methods currently used in clinical reading. We present results on chest CT data, which shows significant improvement over existing state-of-the-art methods: unsharp masking, adaptive histogram equalization (AHE), and the contrast limited adaptive histogram equalization (CLAHE). A systematic study on 109 clinical chest CT images by three radiologists suggests the promise of this method in terms of both interpretation time and diagnostic performance on different pathological cases. In addition, radiologists observed no noticeable artifacts or amplification of noise that usually appears in traditional adaptive histogram equalization and its variations.

Methodology for evaluating image-segmentation algorithms

The purpose of this paper is to describe a framework for evaluating image segmentation algorithms. Image segmentation consists of object recognition and delineation. For evaluating segmentation methods, three factors - precision (reproducibility), accuracy (agreement with truth, validity), and efficiency (time taken) - need to be considered for both recognition and delineation. To assess precision, we need to choose a figure of merit, repeat segmentation considering all sources of variation, and determine variations in figure of merit via statistical analysis. It is impossible usually to establish true segmentation. Hence, to assess accuracy, we need to choose a surrogate of true segmentation and proceed as for precision. In determining accuracy, it may be important to consider different landmark areas of the structure to be segmented depending on the application. To assess efficiency, both the computational and the user time required for algorithm and operator training and for algorithm execution should be measured and analyzed. Precision, accuracy, and efficiency are interdependent. It is difficult to improve one factor without affecting others. Segmentation methods must be compared based on all three factors. The weight given to each factor depends on application.

Hybrid Segmentation of Anatomical Data

We propose new hybrid methods for automated segmentation of radiological patient data and the Visible Human data. In this paper, we integrate boundary-based and region-based segmentation methods which amplifies the strength but reduces the weakness of both approaches. The novelty comes from combining a boundary-based method, the deformable model-based segmentation with region-based segmentation methods, the fuzzy connectedness and Voronoi Diagram-based segmentation, to develop hybrid methods that yield high precision, accuracy and efficiency. This work is a part of a NLM funded effort to provide a fully implemented and tested Visible Human Project Segmentation and Registration Toolkit (Insight).

State of the Art of Level Set Methods in Segmentation and Registration of Medical Imaging Modalities

Segmentation of medical images is an important step in various applications such as visualization, quantitative analysis and image-guided surgery. Numerous segmentation methods have been developed in the past two decades for extraction of organ contours on medical images. Low-level segmentation methods, such as pixel-based clustering, region growing, and filter-based edge detection, require additional pre-processing and post-processing as well as considerable amounts of expert intervention or information of the objects of interest. Furthermore, the subsequent analysis of segmented objects is hampered by the primitive, pixel or voxel level representations from those region-based segmentation [1]. Deformable models, on the other hand, provide an explicit representation of the boundary and the shape of the object. They combine several desirable features such as inherent connectivity and smoothness,which counteract noise and boundary irregularities, as well as the ability to incorporate knowledge about the object of interest [2, 3, 4]. However, parametric deformablemodels have two main limitations. First, in situations where the initial model and desired object boundary differ greatly in size and shape, the model must be reparameterized dynamically to faithfully recover the object boundary. The second limitation

Wavelets in Medical Image Processing: Denoising, Segmentation, and Registration

Wavelet transforms and other multi-scale analysis functions have been used for compact signal and image representations in de-noising, compression and feature detection processing problems for about twenty years. Numerous research works have proven that space-frequency and spacescale expansions with this family of analysis functions provided a very efficient framework for signal or image data. The wavelet transform itself offers great design flexibility. Basis selection, spatial-frequency tiling, and various wavelet threshold strategies can be optimized for best adaptation to a processing application, data characteristics and feature of interest. Fast implementation of wavelet transforms using a filter-bank framework enable real time processing capability. Instead of trying to replace standard image processing techniques, wavelet transforms offer an efficient representation of the signal, finely tuned to its intrinsic properties. By combining such representations with simple processing techniques in the transform domain, multi-scale analysis can accomplish remarkable performance and efficiency for many image processing problems.

Segmentation and Evaluation of Adipose Tissue from Whole Body MRI Scans

Accurate quantification of total body and the distribution of regional adipose tissue using manual segmentation is a challenging problem due to the high variation between manual delineations. Manual segmentation also requires highly trained experts with knowledge of anatomy. We present a hybrid segmentation method that provides robust delineation results for adipose tissue from whole body MRI scans. A formal evaluation of accuracy of the segmentation method is performed. This semi-automatic segmentation algorithm reduces significantly the time required for quantification of adipose tissue, and the accuracy measurements show that the results are close to the ground truth obtained from manual segmentations.

De-noising SPECT/PET Images Using Cross-Scale Regularization

De-noising of SPECT and PET images is a challenging task due to the inherent low signal-to-noise ratio of acquired data. Wavelet based multi-scale denoising methods typically apply thresholding operators on sub-band coefficients to eliminate noise components in spatial-frequency space prior to reconstruction. In the case of high noise levels, detailed scales of sub-band images are usually dominated by noise which cannot be easily removed using traditional thresholding schemes. To address this issue, a cross-scale regularization scheme is introduced, which takes into account cross-scale coherence of structured signals. Preliminary results show promising performance in denoising clinical SPECT and PET images for liver and brain studies. Wavelet thresholding was also compared to denoising with a brushlet expansion. The proposed regularization scheme eliminates the need for threshold parameter settings, making the denoising process less tedious and suitable for clinical practice.

Improving statistics for hybrid segmentation of high-resolution multichannel images

High-resolution multichannel textures are difficult to characterize with simple statistics and the high level of detail makes the selection of a particular contour using classical gradient-based methods not effective. We have developed a hybrid method that combines fuzzy connectedness and Voronoi diagram classification for the segmentation of color and multichannel objects. The multi-step classification process relies on homogeneity measures derived from moment statistics and histogram information. These color features have been optimized to best combine individual channel information in the classification process. The segmentation initialization requires only a set of interior and exterior seed points, minimizing user intervention and the influence of the initialization on the overall quality of the results. The method was tested on volumes from the Visible Human and on brain multi-protocol MRI data sets. The hybrid segmentation produced robust, rapid and finely detailed contours with good visual accuracy. The addition of quantized statistics and color histogram distances as classification features improved the robustness of the method with regards to initialization when compared to our original implementation.

Evaluation of ischemic stroke hybrid segmentation in a rat model of temporary middle cerebral artery occlusion using ground truth from histologic and MR data

A segmentation method that quantifies cerebral infarct using rat data with ischemic stroke is evaluated using ground truth from histologic and MR data. To demonstrate alternative approach to rapid quantification of cerebral infarct volumes using histologic stained slices that requires scarifying animal life, a study with MR acquire volumetric rat data is proposed where ground truth is obtained by manual delineations by experts and automated segmentation is assessed for accuracy. A framework for evaluation of segmentation is used that provides more detailed accuracy measurements than mere cerebral infarct volume. Our preliminary experiment shows that ground truth derived from MRI data is at least as good as the one obtained from the histologic slices for evaluating segmentation algorithms for accuracy. Therefore we can develop and evaluate automated segmentation methods for rapid quantification of stroke without the necessitating animal sacrifice.

Flow-resolution Enhancement in Electrophoretic NMR Using De-noising and Linear Prediction

. Abstract: Detection of electrophoretic motion of ionic species using multi-dimensional Electrophoretic NMR (nD-ENMR) has demonstrated the potential to distinguish signals from two molecules in a solution mixture without their physical separation (1). Therefore, this technique may be applied for simultaneous structure determination of proteins and protein conformations, even during their biochemical interactions. Indeed, this has been achieved by introducing an additional dimension of electrophoretic mobility to the conventional multi-dimensional NMR by applying an external DC electric field. Consequently, the protein spectra are differently modulated by their electrophoretic mobilities in the electrophoretic flow dimension. Unfortunately, spectral resolution in the flow dimension has been limited by severe signal truncations due to the limited DC electric field available before onset of heating-induced convection. Linear prediction (2), which have been widely used for high-resolution spectral estimation from finite Fourier samples, have already been proposed to extend the truncated ENMR flow oscillation curves(3). However, we found that the spectral quality of linear prediction deteriorates as the spectral S/N decreases. To alleviate this problem, we have denoised the ENMR data using low pass filters prior to linear prediction. This technique has lead to improved resolution in the electrophoretic flow dimension. The approach was applied to analyze a 2D ENMR data matrix obtained from a mixture solution of two proteins ubiquitin and bovine serum albumin (BSA) in D2O.