Siemens Medical Solutions USA, Inc.

Siemens Healthcare Gmbh

Siemens Healthineers Ag

Yoshihisa Shinagawa

Gerardo Hermosillo Valadez

Bing Jian

Halid Yerebakan

Yiqiang Zhan

Xiang Sean Zhou

Parmeet Singh Bhatia

Simon Allen-Raffl

Mahesh Ranganath

Matthias Wolf

Shu Liao

Fitsum Aklilu Reda

Vandana Mohan

Halid Ziya Yerebakan

Luca Bogoni

Marcos Salganicoff

Sarang Lakare

Jinbo Bi

Dijia Wu

Zhigang Peng

Le Lu

Anna Jerebko

Yu Zhao

Zhennan Yan

Ke Zeng

Yuping Lin

Shishir Dube

Katharina Schmidler Burk

Shaika Chowdhury

Michael Rusitska

A computer implemented method and apparatus for determining a location at which a given feature is represented in medical imaging data is disclosed. A first descriptor for a first location in first medical imaging data is obtained. The first location is the location within the first medical imaging data at which the given feature is represented. A second descriptor for each of a plurality of candidate second locations in second medical imaging data is obtained. A similarity metric indicating a degree of similarity with the first descriptor is calculated for each of the plurality of candidate second locations. A candidate second location is selected from among the plurality of candidate second locations based on the calculated similarity metrics. The location at which the given feature is represented in the second medical imaging data is determined based on the selected candidate second location.

Determining a location at which a given feature is represented in medical imaging data

There is disclosed a method and apparatus for annotating a first portion of medical imaging data with one or more words corresponding to a respective one or more features represented in the first portion of medical imaging data. A similarity metric indicating a degree of similarity between the first portion and each of a plurality of second portions of reference medical imaging data is determined, at least one of the plurality of second portions being annotated with one or more first words corresponding to a respective one or more features represented in the second portion. A second portion is selected based on the similarity metrics, and the first portion is annotated with the one or more first words with which the second portion, selected for the first portion, is annotated.

Method and apparatus for annotating a portion of medical imaging data with one or more words

A framework for processing medical images. The framework may include receiving a target medical image, a reference medical image and at least one marker associated with a location in the reference medical image. A corresponding location of the at least one marker is determined in the target medical image. The target medical image is overlaid with the at least one marker at the determined corresponding location to provide an overlaid image. Display data is generated to cause a display device to display the overlaid image.

Method for processing medical images

A framework for segmenting a medical text report into sections is disclosed. For each sentence of the report, a first sentence representation is determined by inputting a word-level context representation for each sentence sequentially into a neural network. A second sentence representation is determined by inputting an aggregated representation for each sentence sequentially into another neural network. For each sentence, a third sentence representation is determined based on a combination of the first and second sentence representations, and a section classification for the sentence is determined by inputting the third sentence representation into a section classifier. Each sentence is assigned the section classification determined for the sentence.

Method and apparatus for segmenting a medical text report into sections

Method and apparatus for image retrieval. An image descriptor of a query image indicating a medical abnormality and non-image patient data may be received. For each of a plurality of candidate images stored in a database, an image descriptor of the candidate image and data about a medical abnormality known to be indicated by the candidate image may further be received. A similarity metric between the image descriptors of the query image and the candidate image may be determined for each candidate image. A first probability of the query image medical abnormality being the candidate image medical abnormality given the non-image patient data associated with the query image may be determined for each candidate image. A score may then be determined for each candidate image based on the determined similarity metric and the determined first probability. One or more of the candidate images may be retrieved from the database in accordance with the determined scores.

Image retrieval

A computer-implemented method of encoding a word for use in a method of text analysis comprises receiving input text to be analysed, the input text comprising a first word which is not represented in a vocabulary set stored on a storage. The vocabulary set comprises a plurality of words and an associated word embedding vector for each word in the set. The method comprises identifying the first word as a word which is not represented in the vocabulary set and determining one or more sub-words within the first word with which to encode the first word. Each of the one or more sub-words corresponds with a word represented in the vocabulary set and having an embedding vector in the vocabulary set. The method comprises determining an encoding for the first word based on the one or more sub-words.

Encoding textual information for text analysis

A method and for quantifying a three-dimensional medical image volume are provided. An embodiment of the method includes: providing a two-dimensional representation image based on the medical image volume; defining a region of interest in the two-dimensional representation image; generating a feature signature for the region of interest; defining a plurality of two-dimensional image patches in the medical image volume; calculating, for each of the image patches, a degree of similarity between the region of interest and the respective image patch on the basis of the feature signature; visualizing the degrees of similarities.

Spatial distribution of pathological image patterns in 3D image data

A visualization framework based on document representation learning is described herein. The framework may first convert a free text document into word vectors using learning word embeddings. Document representations may then be determined in a fixed-dimensional semantic representation space by passing the word vectors through a trained machine learning model, wherein more related documents lie closer than less related documents in the representation space. A clustering algorithm may be applied to the document representations for a given patient to generate clusters. The framework then generates a visualization based on these clusters.

Visualization framework based on document representation learning

A framework for automated measurement. In accordance with one aspect, the framework detects a centerline point of a structure of interest in an image. A centerline of the structure of interest may be traced based on the detected centerline point. A trained segmentation learning structure may be used to generate one or more contours of the structure of interest along the centerline. One or more measurements may then be extracted from the one or more contours.

Automated measurement based on deep learning

A framework for automatic retrieval of medical images. In accordance with one aspect, the framework detects patches in a query image volume that contain at least a portion of an anatomical region of interest by using a first trained classifier. The framework determines disease probabilities by applying a second trained classifier to the detected patches, and selects, from the patches, a sub-set of informative patches with disease probabilities above a pre-determined threshold value. For a given patch from the sub-set of informative patches, the framework retrieves, from a database, patches that are most similar to the given image. Image volumes associated with the retrieved patches are then retrieved from the database. A report based on the retrieved image volumes may then be generated and presented.

Automatic medical image retrieval

Disclosed herein is a framework for segmenting articulated structures. In accordance with one aspect, the framework receives a target image, a reference image, statistical shape models, local appearance models and a learned landmark detector. The framework may automatically detect first centerline landmarks along centerlines of articulated structures in the target image using the learned landmark detector. The framework may then determine a non-rigid transformation function that registers second centerline landmarks along centerlines of articulated structures in the reference image with the first centerline landmarks. Mean shapes of the statistical shape models may then be deformed to the target image space by applying the non-rigid transformation function on the mean shapes. The framework may further search for candidate points in the mean shapes using the local appearance models. The mean shapes may be fitted to the candidate points to generate a segmentation mask.

Automatic segmentation of articulated structures

Described herein is a technology for facilitating classification of an object. In one implementation, at least one quotient appearance manifold mapping is constructed from a sample image to untangle appearance fiber bundles. A feature characteristic of the sample image may then be extracted based on the quotient appearance manifold mapping. A classifier may further be trained based on the extracted feature, and adapted for associating the object with an object class.

Quotient appearance manifold mapping for image classification

Systems and methods are provided for image segmentation. In accordance with some implementations, a current segmentation mask associated with an object of interest is iteratively refined. Any image element associated with a previously generated fence is excluded from the current segmentation mask. The fence may be generated around one or more image elements that violate a shape constraint.

Shape-based image segmentation

Described herein is a technology for facilitating visualization of a tubular structure. Digitized image data of the tubular structure is received and processed to determine a centerline. A first transformation operation is performed on a first set of coordinates representing the tubular structure to generate a transformed tubular structure with a straight centerline. A second transformation operation is then performed locally on a second set of coordinates representing at least one fold of the transformed tubular structure to generate a transformed fold, which is perpendicular to the centerline.

Image display of a tubular structure

Described herein is a framework for automatically classifying a structure in digital image data are described herein. In one implementation, a first set of features is extracted from digital image data, and used to learn a discriminative model. The discriminative model may be associated with at least one conditional probability of a class label given an image data observation Based on the conditional probability, at least one likelihood measure of the structure co-occurring with another structure in the same sub-volume of the digital image data is determined. A second set of features may then be extracted from the likelihood measure.

Growing community of inventors

Downingtown, PA, United States of America

Yoshihisa Shinagawa