Abstract: A method for training a computer system for automatic detection of regions of interest includes receiving patient records. For each of the received patient records a text field and a medical image are identified from within the patient record and the medical image is automatically segmented to identify a structure of interest. The text field is searched for one or more keywords indicative of a particular abnormality associated with the structure of interest. The medical image is added to a grouping representing the particular abnormality when the text field indicates that the patient has the particular abnormality and the medical image is added to a grouping representing the absence of the particular abnormality when the text field does not indicate that the patient has the particular abnormality. The groupings of medical images are used to automatically train a computer system for the subsequent detection of the particular abnormality.

Abstract: Disclosed is robust click-point linking, defined as estimating a single point-wise correspondence between data domains given a user-specified point in one domain or as an interactive localized registration of a monomodal data pair. To link visually dissimilar local regions, Geometric Configuration Context (GCC) is introduced. GCC represents the spatial likelihood of the point corresponding to the click-point in the other domain. A set of scale-invariant saliency features are pre-computed for both data. GCC is modeled by a Gaussian mixture whose component mean and width are determined as a function of the neighboring saliency features and their correspondences. This allows correspondence of dissimilar parts using only geometrical relations without comparing the local appearances. GCC models are derived for three transformation classes: pure translation, scaling and translation, and similarity transformation.

Abstract: A method for segmenting digitized images includes providing a training set comprising a plurality of digitized whole-body images, providing labels on anatomical landmarks in each image of said training set, aligning each said training set image, generating positive and negative training examples for each landmark by cropping the aligned training volumes into one or more cropping windows of different spatial scales, and using said positive and negative examples to train a detector for each landmark at one or more spatial scales ranging from a coarse resolution to a fine resolution, wherein the spatial relationship between a cropping windows of a coarse resolution detector and a fine resolution detector is recorded.

Abstract: A method of detecting breast masses and calcifications in digitized images, includes providing a plurality of 2-dimensional (2D) digital X-ray projectional breast images acquired from different viewing angles, extracting candidate lesions and 2D features from said 2D projectional images, computing spicularity characteristics of said candidate lesions, including location, periodicity, and amplitude, applying learning algorithms to said candidate lesions to predict a probability of malignancy of said lesion, receiving from said learning algorithm a probability map of detections for each breast image, said detections comprising associating pixels with a probability of being associated with a malignancy, creating a synthetic 2D slice for each X-ray image wherein malignant regions are indicated by ellipses on a non-malignant background, and constructing a synthetic 3-dimensional (3D) image volume from said 2D synthetic slices.

Abstract: Systems and methods for performing a medical imaging study include 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 bolus tracking includes acquiring one or more baseline images. One or more trigger regions are automatically established within the baseline images. A bolus is administered. The automatically established trigger regions are monitored for bolus arrival at the one or more trigger regions. Bolus arrival at a volume of interest is forecasted based on the bolus arrival at the one or more trigger regions. A diagnostic scan of the volume of interest is acquired at the forecasted time.

Abstract: A system for knowledge-based image computer aided detection includes a text interpretation system receiving an electronic patient record and outputting an assertion relevant for the electronic patient record, an annotation/detection system detects anatomical and functional structures in input images and interacts with the text interpretation system to receive the assertion and outputting annotated images based on the input images, and an imaging decision support system taking the annotated images and outputting classifications of annotated structures in the annotated images.

Abstract: A method of automatic change quantification for medical decision support includes: automatically detecting a structure in a set of medical images; characterizing the detected structure including modeling of deformation characteristics of the detected structure; matching images based on the characterization of the detected structure, wherein a size measure of the detected structure is constrained according to the deformation characteristics; and quantifying a change in the detected structure.

Abstract: A method for determining a volume of interest in data includes determining fixed-bandwidth estimations of a plurality of analysis bandwidths, wherein the estimation of the fixed-bandwidth comprises, providing an estimate of a mode location of the volume of interest in the data, and determining a covariance of the volume of interest using a local Hessian matrix. The method further includes determining the volume of interest as a most stable fixed-bandwidth estimation across each of the plurality of analysis bandwidths.

Abstract: A system and method are provided for prior-constrained mean shift analysis of a data array, the system including a processor, an input adapter in signal communication with the processor for receiving at least one data array, and a prior constraints unit in signal communication with the processor for performing a prior-constrained mean shift analysis on the at least one data array; and the method including receiving initialization data, selecting an initial point relative to the initialization data, Gaussian fitting with a prior-constrained mean shift responsive to the initial point to parse a structure, and setting the parsed structure as a prior constraint.

Abstract: A method for synchronizing patient data between at least two independent applications in a distributed environment includes capturing screen information from a display window of a first application client that is displaying a medical image of a patient, analyzing the screen information captured from the first application client display to extract patient identifying information, and synchronizing a display of information of the patient on a second application system display screen with the first application display window using the extracted patient identification information.

Abstract: A method for displaying pre-rendered medical images on a workstation includes receiving three-dimensional medical image data. A region of suspicion is automatically identified within the three-dimensional medical image data. A rendering workstation is used to pre-render the three-dimensional medical image data into a sequence of two-dimensional images in which the identified region of suspicion is featured from a vantage point that is automatically selected to maximize diagnostic value of the two-dimensional images for determining whether the region of suspicion is an actual abnormality. The sequence of pre-rendered two-dimensional images is then stored in a PACS, where it can then be displayed on a viewing workstation.

Abstract: A method for training a computer system for automatic detection of regions of interest includes receiving patient records. For each of the received patient records a text field and a medical image are identified from within the patient record and the medical image is automatically segmented to identify a structure of interest. The text field is searched for one or more keywords indicative of a particular abnormality associated with the structure of interest. The medical image is added to a grouping representing the particular abnormality when the text field indicates that the patient has the particular abnormality and the medical image is added to a grouping representing the absence of the particular abnormality when the text field does not indicate that the patient has the particular abnormality. The groupings of medical images are used to automatically train a computer system for the subsequent detection of the particular abnormality.

Abstract: We propose using different classifiers based on the spatial location of the object. The intuitive idea behind this approach is that several classifiers may learn local concepts better than a “universal” classifier that covers the whole feature space. The use of local classifiers ensures that the objects of a particular class have a higher degree of resemblance within that particular class. The use of local classifiers also results in memory, storage and performance improvements, especially when the classifier is kernel-based. As used herein, the term “kernel-based classifier” refers to a classifier where a mapping function (i.e., the kernel) has been used to map the original training data to a higher dimensional space where the classification task may be easier.

Abstract: A method for determining a structure in volumetric data includes determining an anisotropic scale-space for a local region around a given spatial local maximum, determining L-normalized scale-space derivatives in the anisotropic scale-space, and determining the presence of noise in the volumetric data and upon determining noise in the volumetric data, determining the structure by a most-stable-over-scales determination, and upon determining noise below a desirable level, determining the structure by one of the most-stable-over-scales determination and a maximum-over-scales determination.

Abstract: A system and method for monitoring disease progression or response to therapy using multi-modal visualization are provided. The method comprises: selecting a first image dataset of a first timepoint; loading the first image dataset of the first timepoint; selecting a second image dataset of a second timepoint; loading the second image dataset of the second timepoint; registering the first image dataset of the first timepoint and the second image dataset of the second timepoint; and displaying the first image dataset of the first timepoint and the second image dataset of the second timepoint.

Abstract: CAD (computer-aided decision) support systems, methods and tools are provided for automated decision support for screening, evaluating, and/or diagnosing medial conditions. For example, CAD support systems and tools implement methods for automatically processing patient data for a subject patient using various interpretation methods, and integrally rending and presenting the interpretation results to a user (e.g., physician, radiologist, etc.) in a manner that enables fast and efficient screening, evaluation, and/or diagnosis of potential medical conditions of the subject patient.

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: CAD (computer-aided decision) support systems, methods and tools for medical imaging are provided, which use machine learning classification for automated detection and marking of regions of interest in medical images. Machine learning methods are used for adapting/optimizing a CAD process by seamlessly incorporating physician knowledge into the CAD process using training data that is obtained during routine use of the CAD system.

Abstract: CAD (computer-aided detection) systems, methods and tools are provided for automatically inserting “false” marks (e.g., incorrect marks, misleading marks, etc.) in medical images to ensure an unbiased CAD-assisted review of the marked medical images by physicians, clinicians, radiologists, etc. For example, a method for automatic detection of medical conditions in medical images includes the steps of receiving image data, processing the image data to detect potential medical conditions in the image data, adding a mark in the image data that indicates a detected medical condition, adding a false mark in the image data; and outputting marked image data comprising one or more marks that indicate a detected medical condition, or one or more false marks, or both. The individual performing a CAD-assisted review of the “marked” image data is aware that one or more “false” marks may be included in displayed images, which prevents blind reliance on the CAD results.