General Electric Company

Roshni Bhagalia

Bipul Das

Debashish Pal

Michael John Gerdes

Ali Can

Xiaodong Tao

Musodiq O Bello

Bhushan Dayaram Patil

Fei Zhao

Xiaojie Huang

Mirabela Rusu

Qi Song

Suvadip Mukherjee

Srikrishnan V

Ziyue Xu

Peter Lamb

Francis Edward Pavan-Woolfe

As set forth herein low energy signal data and high energy signal data can be acquired. A first material decomposed (MD) image of a first material basis and a second material decomposed (MD) image of a second material basis can be obtained using the low energy signal data and the high energy signal data. At least one of the first or second MD image can be input into a guide filter for output of at least one noise reduced and cross-contamination reduced image. A computed tomography (CT) imaging system can be provided that includes an X-ray source and a detector having a plurality of detector elements that detect X-ray beams emitted from the X-ray source. Low energy signal data and high energy signal data can be acquired using the detector.

Guided filter for multiple level energy computed tomography (CT)

Embodiments described herein provide a hybrid technique which incorporates learned pulmonary nodule features in a model based energy minimization segmentation using graph cuts. Features are extracted from training samples using a convolutional neural network, and the segmentation cost function is augmented via the deep learned energy. The system and method improves segmentation performance and more robust initialization.

Automated segmentation using deep learned priors

The present approach relates, in some aspects, to a multi-level and a multi-channel frame work for segmentation using model-based or 'shallow' classification (i.e. learning processes such as linear regression, clustering, support vector machines, and so forth) followed by deep learning. This framework starts with a very low resolution version of the multi-channel data and constructs an shallow classifier with simple features to generate a coarser level tissue mask that in turn is used to crop patches from the high-resolution volume. The cropped volume is then processed using the trained convolution network to perform a deep learning based segmentation within the slices.

Material segmentation in image volumes

A method of creating a diagnostic evaluation for usual interstitial pneumonia is provided, including obtaining a first plurality of series of HRCT lung slices indicating the presence of UIP, obtaining an identification of UIP and non-UIP voxels, extracting textural and localization features from the UIP and non-UIP voxels, selecting features that are more accurate in differentiating UIP voxels from non-UIP voxels than other features are, eliminating features highly correlated with a more accurate feature, and constructing a predictive model by performing a second classifier to provide a probability that a voxel signifies the presence of UIP. Also provided is a method of identifying UIP in a subject's lung by applying a diagnostic evaluation for UIP that was created with the foregoing method.

Textural analysis of diffused disease in the lung

A hierarchical multi-object active appearance model (AAM) framework is disclosed for processing image data, such as localizer or scout image data. In accordance with this approach, a hierarchical arrangement of models (e.g., a model pyramid) maybe employed where a global or parent model that encodes relationships across multiple co-located structures is used to obtain an initial, coarse fit. Subsequent processing by child sub-models add more detail and flexibility to the overall fit.

Automatic estimation of anatomical extents

A computer-implemented method of post-processing medical image data is provided. The method includes receiving tracked image data representative of multiple blood vessels, generating a binary tree structure for the multiple blood vessels based on a parent-child relationship between branches of the multiple blood vessels, generating a likelihood model for determining a validity of the branches of the multiple blood vessels, and generating a likelihood score for each branch based on the respective branch's compatibility with the likelihood model. The method also includes generating a reconstructed tree for the multiple blood vessels. Compatible branches are included in the reconstructed tree, while invalid branches are not included in the reconstructed tree.

Methods for generating a reconstructed tree structure from medical image data

Techniques for removing image autoflourescence from fluorescently stained biological images are provided herein. The techniques utilize non-negative matrix factorization that may constrain mixing coefficients to be non-negative. The probability of convergence to local minima is reduced by using smoothness constraints. The non-negative matrix factorization algorithm provides the advantage of removing both dark current and autofluorescence.

Growing community of inventors

Niskayuna, NY, United States of America

Roshni Bhagalia