Abstract: Described herein is a technology for facilitating coordinated description in image analysis. In one implementation, the technology includes receiving image data including at least first and second descriptors (204) describing portions of the image data. The first and second descriptors are coordinated by determining at least one conditional probability of observing the first descriptor in the image data given an occurrence of the second descriptor (206). A classifier may then be trained based on the conditional probability (208).

Abstract: Described herein is a method and system for facilitating automatic classification of regions-of-interest (ROIs). Contrast enhanced image data may be received (202) and processed to generate at least one texture value of at least one ROI in the image data (204). The ROI may be automatically classified as either a mass or a non-mass like enhancement (NMLE) based on the texture value (e.g., bumpiness) (208).

Abstract: A method for automatic detection of lesions within a medical image include acquiring medical image data. Regions of suspicion are automatically identified within the medical image data. It is automatically determined whether each identified region of suspicion is of a benign state, is of a suspicious state that requires a biopsy, or is of an indeterminate state that requires subsequent imaging after a particular length of time. When an identified region of suspicion is determined to be of an indeterminate state, the determination is automatically reconsidered in light of a calibration factor that biases the automatic determination towards either a benign state or a suspicious state. The calibration factor may be based on data collected from follow-up examinations that reveal whether a lesion previously characterized as indeterminate was actually a benign or malignant lesion or on additional diagnostic information including prior image data or non-image data.

Abstract: Described herein is a framework for constructing a hierarchical classifier for facilitating classification of digitized data. In one implementation, a divergence measure of a node of the hierarchical classifier is determined. Data at the node is divided into at least two child nodes based on a splitting criterion to form at least a portion of the hierarchical classifier. The splitting criterion is selected based on the divergence measure. If the divergence measure is less than a predetermined threshold value, the splitting criterion comprises a divergence-based splitting criterion which maximizes subsequent divergence after a split. Otherwise, the splitting criterion comprises an information-based splitting criterion which seeks to minimize subsequent misclassification error after the split.

Abstract: Described herein is a method and system for facilitating segmentation of images. A difference image is received and processed to extract at least one histogram (402). A noise component is determined by fitting a symmetric Gaussian distribution to the extracted histogram such that the negative portion of the Gaussian distribution coincides with the negative portion of the histogram (403). The noise component is then subtracted from the histogram to generate a probability distribution function (404), which may be converted to a cumulative distribution function (406) and applied to the difference image to generate a probabilistic representation of contrast enhancement (408).

Abstract: A method of detecting blood vessel shadows in an anterior posterior x-ray radiograph comprising the steps of: generating candidate sub areas of the radiograph showing changes in contrast above a threshold level; supressing rib shadow edges; eliminating lung tissue shadow edges, and categorizing and eliminating nodule shadows.

Abstract: A method for computer aided detection of pulmonary emboli includes acquiring medical image data. A pulmonary embolism candidate comprising a cluster of voxels is identified. It is determined whether the candidate is a true pulmonary embolism or a false positive based on a spatial distribution of intensity values for the voxels of the cluster of voxels. The pulmonary embolism candidate is presented to a user when the candidate is determined to be a true pulmonary embolism.

Abstract: A method for diagnosing malignancy of suspect regions in a 2D imaging projection of a body organ of a subject comprising the steps of: normalizing the image projection to map regions thereof onto a standard shape; extracting location of regions mapped on the standard shape and using normalized position of a suspect region to assess likelihood that it is malignant.

Abstract: A method of removing obscuring features of relatively standard size and shape from an image comprising the obscuring features, features of interest and background, comprising the steps of applying a geometric transformation to generate a transformed image with the obscuring features aligned with an axis of image; applying an attenuation algorithm to the transformed image to suppress features having a dimension in direction of axis of range appropriate to dimension of the obscuring features in similar alignment, and applying a reverse geometrical transformation to produce a modified image with obscuring features attenuated.

Abstract: A method for obtaining a tissue volume, includes inputting a dataset including a plurality of voxels; initializing a tissue probability volume for the plurality of voxels to a pre-determined value; updating, by one of increasing or decreasing the tissue probability volume of each of the plurality of voxels, based on corresponding intensity values of each of the plurality of voxels; and generating the tissue volume by combining the updated tissue probability volume and the inputted dataset.

Abstract: A method of detecting malacia in airways includes providing a plurality of 3-dimensional (3D) digital lung images acquired over an inhalation/exhalation cycle from a same subject, each said image comprising a set of intensities on a 3-dimensional grid of points, registering a successive pair of images, wherein a registration mapping of the point grid of one image is calculated, locating airways in each of said pair of images, and collecting those points between the airways in each of said pair of images, wherein a volume of said collected points is a measure of an extent of malacia in said airways.

Abstract: A method for performing pharmacokinetic analysis in magnetic resonance (MR) images includes administering a dose of contrast agent (CA) into a subject. A sequence of medical images is acquired of the subject at set temporal intervals. The time-based behavior of concentrations of CA is described within the subject for each voxel of each medical image of the sequence of medical images based on a reference voxel using a compartmental model for pharmacokinetic analysis that is based on a set of compartmental model parameters. The compartmental model is solved for each of the compartmental model parameters. The solution for the compartmental model parameters is used to estimate one or more parameters of physiological significance.

Abstract: A method for decomposing digital medical images includes providing a digital medical image, segmenting the image into one or more biological structures, extracting one or more segmented biological structures from the image by extracting all voxels within a spatial extent of each of the biological structures to construct one or more new component volumes of the biological structures. For each of the one or more new component volumes, generate a sequence of 2-dimensional projective views by moving a projection viewpoint around each the biological structure in the one or more new component images, and generate a 2-dimensional projective view from each viewpoint, and display a cine loop of the sequence of projective views where the biological structures appear to be rotating in the display.

Abstract: A method for segmenting vertebrae in digitized images includes providing a plurality of digitized whole-body images, detecting and segmenting a spinal cord using 3D polynomial spinal model in each of the plurality of images, finding a height of each vertebrae in each image from intensity projections along the spinal cord, and building a parametric model of a vertebrae from the plurality of images. The method further includes providing a new digitized whole-body image including a spinal cord, fitting an ellipse to each vertebrae of the spinal cord to find the major and minor axes, and applying constraints to the major and minor axes in the new image based on the parametric model to segment the vertebrae.

Abstract: A method for differentiating pulmonary nodules in digitized medical images includes identifying an object of interest from a digital image of the lungs, computing a first distance map of each point of the object of interest, determining a seed point from the first distance map, starting from the seed point, growing a first region by adding successive adjacent layers of points until a background point is reached, and partitioning the first region into a nodule region and a non-nodule region.

Abstract: A method and system for facilitating computer-aided detection (CAD) in which, in one implementation, image data is received (302) and iterations of an iterative segmentation process is performed on the image data. Each iterative segmentation process may include ascertaining whether a segment is normal (304), removing the segment from the image if ascertained to be normal (308) and transforming the shape of the segment (310). The iterative segmentation process may be stopped if a stop condition is met (312).

Abstract: A method for performing a medical imaging study includes acquiring a preliminary scan. A set of local feature candidates is automatically detected from the preliminary scan. The accuracy of each local feature candidate is assessed using multiple combinations of the other local feature candidates and removing a local feature candidate that is assessed to have the lowest accuracy. The assessing and removing steps are repeated until only a predetermined number of local feature candidates remain. A region of interest (ROI) is located from within the preliminary scan based on the remaining predetermined number of local feature candidates. A medical imaging study is performed based on the location of the ROI within the preliminary scan.

Abstract: A method for characterizing a shape of an object surface includes acquiring image data including the object. The image data is analyzed at a locus of points that are at a predetermined distance from a point of interest proximate to the object surface to determine which of the locus of points represents a foreground and which of the locus of points represents a background. The shape of the object surface is characterized based on the characterization of the locus of points.

Abstract: A method for detecting and localizing multiple anatomical landmarks in medical images, including: receiving an input requesting identification of a plurality of anatomical landmarks in a medical image; applying a multi-landmark detector to the medical image to identify a plurality of candidate locations for each of the anatomical landmarks; for each of the anatomical landmarks, applying a landmark-specific detector to each of its candidate locations, wherein the landmark-specific detector assigns a score to each of the candidate locations, and wherein candidate locations having a score below a predetermined threshold are removed; applying spatial statistics to groups of the remaining candidate locations to determine, for each of the anatomical landmarks, the candidate location that most accurately identifies the anatomical landmark; and for each of the anatomical landmarks, outputting the candidate location that most accurately identifies the anatomical landmark.

Abstract: A landmark location system for locating landmarks in volumes includes a medical image database including volumes of medical images, a learning unit that trains a multi-class classifier to locate a landmark point in each volume from extracted features of the volumes near a sample point offset from the landmark point and discrete displacements of the sample point to the landmark point, and a landmark locator that locates the landmark point in an input volume using the trained multi-class classifiers.