Abstract: A method and system for on-line learning of landmark detection models for end-user specific diagnostic image reading is disclosed. A selection of a landmark to be detected in a 3D medical image is received. A current landmark detection result for the selected landmark in the 3D medical image is determined by automatically detecting the selected landmark in the 3D medical image using a stored landmark detection model corresponding to the selected landmark or by receiving a manual annotation of the selected landmark in the 3D medical image. The stored landmark detection model corresponding to the selected landmark is then updated based on the current landmark detection result for the selected landmark in the 3D medical image. The landmark selected in the 3D medical image can be a set of landmarks defining a custom view of the 3D medical image.

Abstract: A method and software system for flaw identification, grouping and sizing for fatigue life assessment for rotors used in turbines and generators. The method includes providing ultrasonic data of a plurality of rotor slices and providing volume reconstruction of the ultrasonic data. The method also includes providing in-slice identification, grouping and sizing of flaw indications in the rotor based on the volume reconstruction. Further, the method includes providing inter-slice identification, grouping and sizing of the flaw indications based on the in-slice flaw indications and providing flaw location and size information. The method can be used in both phased-array and A-scan inspections.

Abstract: The present invention provides a method and system for vascular landmark detection in CT volumes. A CT volume is received and an initial position of a plurality of vascular landmarks is detected. The initial position of each of the plurality of vascular landmarks is then adjusted in order to position each vascular landmark inside a vessel lumen. A new position of each of the plurality of vascular landmarks representing the adjusted initial positions is output.

Abstract: A method of performing image retrieval includes training a random forest RF classifier based on low-level features of training images and a high-level feature, using similarity values generated by the RF classifier to determine a subset of the training images that are most similar to one another, and classifying input images for the high-level feature using the RF classifier and the determined subset of images.

Abstract: Methods and Systems for training a learning based classifier and object detection in medical images is disclosed. In order to train a learning based classifier, positive training samples and negative training samples are generated based on annotated training images. Features for the positive training samples and the negative training samples are extracted. The features include an extended Haar feature set including tip features and corner features. A discriminative classifier is trained based on the extracted features.

Abstract: An image reconstruction method includes receiving volume data comprising a plurality of sampling points, determining a first conditioning of the sampling points suppressing low amplitudes and conserving maximum amplitudes, determining a second conditioning of the sampling points wherein an influence of a sampling point depends on its distance to a grid point in a sampling grid, determining a kernel comprising a plurality of weighting functions for the first conditioning and the second conditioning to determine an energy spread of each of the plurality of sampling points without determining a shape or size of the kernel, and outputting a reconstructed volume according to the energy spread of each of the plurality of sampling points.

Abstract: A method for predicting fatigue crack growth in materials includes providing a prior distribution obtained using response measures from one or more target components using a fatigue crack growth model as a constraint function, receiving new crack length measurements, providing a posterior distribution obtained using the new crack length measurements, and sampling the posterior distribution to obtain crack length measurement predictions.

Abstract: A method dynamically reconstructing a stress and strain field of a turbine blade includes providing a set of response measurements from at least one location on a turbine blade, band-pass filtering the set of response measurements based on an upper frequency limit and a lower frequency limit, determining an upper envelope and a lower envelope of the set of response measurements from local minima and local maxima of the set of response measurements, calculating a candidate intrinsic mode function (IMF) from the upper envelope and the lower envelope of the set of response measurements, providing an N×N mode shape matrix for the turbine blade, where N is the number of degrees of freedom of the turbine blade, when the candidate IMF is an actual IMF, and calculating a response for another location on the turbine blade from the actual IMF and mode shapes in the mode shape matrix.

Abstract: A method and system for automatic detection and volumetric quantification of bone lesions in 3D medical images, such as 3D computed tomography (CT) volumes, is disclosed. Regions of interest corresponding to bone regions are detected in a 3D medical image. Bone lesions are detected in the regions of interest using a cascade of trained detectors. The cascade of trained detectors automatically detects lesion centers and then estimates lesion size in all three spatial axes. A hierarchical multi-scale approach is used to detect bone lesions using a cascade of detectors on multiple levels of a resolution pyramid of the 3D medical image.

Abstract: A method and system for automatic bone segmentation and landmark detection for joint replacement surgery is disclosed. A 3D medical image of at least a target joint region of a patient is received. A plurality bone structures are automatically segmented in the target joint region of the 3D medical image and a plurality of landmarks associated with a joint replacement surgery are automatically detected in the target joint region of the 3D medical image. The boundaries of segmented bone structures can then be interactively refined based on user inputs.

Abstract: Multiple object segmentation is performed for three-dimensional computed tomography. The adjacent objects are individually segmented. Overlapping regions or locations designated as belonging to both objects may be identified. Confidence maps for the individual segmentations are used to label the locations of the overlap as belonging to one or the other object, not both. This re-segmentation is applied for the overlapping local, and not other locations. Confidence maps in re-segmentation and application just to overlap locations may be used independently of each other or in combination.

Abstract: A method and system for fully automatic segmentation the prostate in magnetic resonance (MR) image data is disclosed. Intensity normalization is performed on an MR image of a patient to adjust for global contrast changes between the MR image and other MR scans and to adjust for intensity variation within the MR image due to an endorectal coil used to acquire the MR image. An initial prostate segmentation in the MR image is obtained by aligning a learned statistical shape model of the prostate to the MR image using marginal space learning (MSL). The initial prostate segmentation is refined using one or more trained boundary classifiers.

Abstract: A method and system for detecting multiple objects in an image is disclosed. A plurality of objects in an image is sequentially detected in an order specified by a trained hierarchical detection network. In the training of the hierarchical detection network, the order for object detection is automatically determined. The detection of each object in the image is performed by obtaining a plurality of sample poses for the object from a proposal distribution, weighting each of the plurality of sample poses based on an importance ratio, and estimating a posterior distribution for the object based on the weighted sample poses.

Abstract: An apparatus and method for training a landmark detector receives training data which includes a plurality of positive training bags, each including a plurality of positively annotated instances, and a plurality of negative training bags, each including at least one negatively annotated instance. Classification function is initialized by training a first weak classifier based on the positive training bags and the negative training bags. All training instances are evaluated using the classification function. For each of a plurality of remaining classifiers, a cost value gradient is calculated based on spatial context information of each instance in each positive bag evaluated by the classification function. A gradient value associated with each of the remaining weak classifiers is calculated based on the cost value gradients, and a weak classifier is selected which has a lowest associated gradient value and given a weighting parameter and added to the classification function.

Abstract: Ribs are automatically ordered and paired. After ordering ribs on each side, magnetic and spring functions are used to solve for rib pairing. The magnetic function is used to constrain possible pairs across sides, and the spring function is used to maintain the order on each side while accounting for missing or fused ribs.

Abstract: A method for probabilistically predicting fatigue life in materials includes sampling a random variable for an actual equivalent initial flaw size (EIFS), generating random variables for parameters (ln C, m) of a fatigue crack growth equation ? a ? N = C ? ( ? ? ? K ) m from a multivariate distribution, and solving the fatigue crack growth equation using these random variables. The reported EIFS data is obtained by ultrasonically scanning a target object, recording echo signals from the target object, and converting echo signal amplitudes to equivalent reflector sizes using previously recorded values from a scanned calibration block. The equivalent reflector sizes comprise the reported EIFS data.

Abstract: A method and apparatus for three-dimensional visualization and analysis for automatic non-destructive examination of a solid Rotor using ultrasonic phased array is disclosed. Data is acquired by scanning a solid rotor with a phased array ultrasound transducer producing a plurality of two dimensional ultrasound scans. Each of a plurality of sample points of a plurality of two dimensional ultrasound scans are associated with a corresponding 3D image point of a regular grid. A kernel function for each of the plurality of sample points defining a size and shape of a kernel located at the corresponding image point is determined. A weight is assigned to each kernel which, in one embodiment, is based on the sample point value. A value for each image point of the regular 3D grid is determined based on kernels overlapping each image point. A three-dimensional volume representing the solid rotor is then visualized.

Abstract: A method and system for providing detecting and classifying coronary stenoses in 3D CT image data is disclosed. Centerlines of coronary vessels are extracted from the CT image data. Non-vessel regions are detected and removed from the coronary vessel centerlines. The cross-section area of the lumen is estimated based on the coronary vessel centerlines using a trained regression function. Stenosis candidates are detected in the coronary vessels based on the estimated lumen cross-section area, and the significant stenosis candidates are automatically classified as calcified, non-calcified, or mixed.

Abstract: A method and system for segmenting multiple organs in medical image data is disclosed. A plurality of landmarks of a plurality of organs are detected in a medical image using an integrated local and global context detector. A global posterior integrates evidence of a plurality of image patches to generate location predictions for the landmarks. For each landmark, a trained discriminative classifier for that landmark evaluates the location predictions for that landmark based on local context. A segmentation of each of the plurality of organs is then generated based on the detected landmarks.

Abstract: In a general methodology for insulation defect identification in a generator core, a Chattock coil is used to measure magnetic potential difference between teeth. Physical knowledge and empirical knowledge is combined in a model to predict insulation damage location and severity. Measurements are taken at multiple excitation frequencies to solve for multiple characteristics of the defect.