International Business Machines Corporation

Merative US L.p.

David James Beymer

Tanveer Fathima Syeda-Mahmood

Fei Wang

Ehsan Dehghan Marvast

Mehdi Moradi

Ritwik K Kumar

Yaniv Gur

Yong Zhang

Karen Wolfe Brannon

Ahmed El Harouni

Shafiqul Abedin

Dulce Beatriz Ponceleon

Arnon Amir

Yufan Guo

Sangeeta T Doraiswamy

Anup Pillai

Satyananda Kashyap

Tyler Shore Baldwin

Vandana Mukherjee

Quan Wang

Hakan Bulu

Hayit Greenspan

Deepta Rajan

Mohammadreza Negahdar

Tanveer Fathima Mahmood

Luyao Shi

Patrick McNeillie

Omar U F Choque

Guy Talmor

Dai Shi

Nathaniel Mason Braman

Nripesh Parajuli

Dinesh C Verma

Shumin Zhai

Ting Chen

Pawan Raghunath Chowdhary

Colin Compas

Stephen Paul Farrell

Swanand Ravindra Kadhe

Hongzhi Wang

Girish Narayan

King Chung Ho

Moritz A W Hardt

Systems, methods, and computer program products for obfuscating search queries are described herein. A method comprises reading an input query; reading a set of known terms organized into hierarchical classes; determining whether the input query is included in the set of known terms; determining one or more classifications for the input query in accordance with its inclusion in the set of known terms; generating a prompt in accordance with the one or more classifications and the input query; providing the prompt to a generative language model as input; receiving, from the generative language model, a plurality of candidate queries in accordance with the prompt; determining a score for the plurality of candidate queries; and generating a plurality of decoy queries based on the plurality of candidate queries and the score.

Obfuscating search queries

A mechanism is provided for implement a discrepancy detection mechanism for detecting discrepancies between clinical notes and administrative records. Clinical concepts are extracted from the clinical notes and the administrative records in a patient's electronic medical records (EMRs). The extracted clinical concepts are filtered based on semantic type information to identify concepts that reference diseases or syndromes while also removing negated instances. Utilizing the positive mentions of diseases in clinical notes, the positive mentions of diseases or syndromes in the clinical notes are compared against each positive entry in the administrative records. A discrepancy summary is then generated for diseases or syndromes that failed to translate correctly from clinical notes to the administrative records in the patient's EMRs.

Detecting discrepancies between clinical notes and administrative records

A candidate generator generates a set of candidate three-dimensional image patches from an input volume. A candidate classifier classifies the set of candidate three-dimensional image patches as containing or not containing disease. Classifying the set of candidate three-dimensional image patches comprises generating an attention mask for each given candidate three-dimensional image patch within the set of candidate three-dimensional image patches to form a set of attention masks, applying the set of attention masks to the set of candidate three-dimensional image patches to form a set of masked image patches, and classifying the set of masked image patches as containing or not containing the disease. The candidate classifier applies soft attention and hard attention to the three-dimensional image patches such that distinctive image regions are highlighted proportionally to their contribution to classification while completely removing image regions that may cause confusion.

Disease detection with masked attention

Methods and systems for training computer-aided condition detection systems. One method includes receiving a plurality of images for a plurality of patients, some of the images including an annotation associated with a condition; iteratively applying a first deep learning network to each of the images to produce an attention map, a feature map, and an image-level probability of the condition for each of the images; iteratively applying a second deep learning network to each feature map produced by the first network to produce a plurality of outputs; training the first network based on the attention map produced for each image; and training the second network based on the output produced for each of the patients. The second network includes a plurality of convolution layers and a plurality of convolutional long short-term memory (LSTM) layers. Each of the outputs includes a patient-level probability of the condition for one of the patients.

Automatic condition diagnosis using an attention-guided framework

Methods and systems for training computer-aided condition detection systems. One method includes receiving a plurality of images for a plurality of patients, some of the images including an annotation associated with a condition; iteratively applying a first deep learning network to each of the images to produce a segmentation map, a feature map, and an image-level probability of the condition for each of the images; iteratively applying a second deep learning network to each feature map produced by the first network to produce a plurality of outputs; training the first network based on the segmentation map produced for each image; and training the second network based on the output produced for each of the patients. The second network includes a plurality of convolution layers and a plurality of convolutional long short-term memory (LSTM) layers. Each of the outputs includes a patient-level probability of the condition for one of the patients.

Automatic condition diagnosis using a segmentation-guided framework

Systems and methods for developing a disease detection model. One method includes training the model using an image study and an associated disease label mined from a radiology report. The image study including a sequence of a plurality of two-dimensional slices of a three-dimensional image volume, and the model including a convolutional neural network layer and a convolutional long short-term memory layer. Training the model includes individually extracting a set of features from each of the plurality of two-dimensional slices using the convolutional neural network layer, sequentially processing the features extracted by the convolutional neural network layer for each of the plurality of two-dimensional slices using the convolutional long short-term memory layer, processing output from the convolutional long short-term memory layer for each of the plurality of two-dimensional slices to generate a probability of the disease, and updating the model based on comparing the probability to the label.

Disease detection from weakly annotated volumetric medical images using convolutional long short-term memory and multiple instance learning

An automatic extraction of disease-specific features from Doppler images to help diagnose valvular diseases is provided. The method includes the steps of obtaining a raw Doppler image from a series of images of an echocardiogram, isolating a region of interest from the raw Doppler image, the region of interest including a Doppler image and an ECG signal, and depicting at least one heart cycle, determining a velocity envelope of the Doppler image in the region of interest, extracting the ECG signal to synchronize the ECG signal with the Doppler image over the at least one heart cycle, within the region of interest, calculating a value of a clinical feature based on the extracted ECG signal synchronized with the velocity envelope, and comparing the value of the clinical feature with clinical guidelines associated with the clinical feature to determine a diagnosis of a disease.

Automatic extraction of disease-specific features from doppler images

A generalized distributed framework is provided for parallel search and retrieval of unstructured and structured patient data across zones with hierarchical ranking. In various embodiments, patient data is ingested from a plurality of data sources. A plurality of data models is populated based on the ingested patient data, each data model comprising an abstract data type. The plurality of data models is stored in an index. A search request is processed against the index, the search request comprising one or more attribute of the abstract data type.

Generalized distributed framework for parallel search and retrieval of unstructured and structured patient data across zones with hierarchical ranking

Systems and methods for developing a disease detection model. One method includes training the model using an image study and an associated disease label mined from a radiology report. The image study including a sequence of a plurality of two-dimensional slices of a three-dimensional image volume, and the model including a convolutional neural network layer and a convolutional long short-term memory layer. Training the model includes individually extracting a set of features from each of the plurality of two-dimensional slices using the convolutional neural network layer, sequentially processing the features extracted by the convolutional neural network layer for each of the plurality of two-dimensional slices using the convolutional long short-term memory layer, processing output from the convolutional long short-term memory layer for a sequentially last of the plurality of two-dimensional slices to generate a probability of the disease, and updating the model based on comparing the probability to the label.

Disease detection from weakly annotated volumetric medical images using convolutional long short-term memory

Annotation of large image datasets is provided. In various embodiments, a plurality of medical images is received. At least one collection is formed containing a subset of the plurality of medical images. One or more image from the at least one collection is provided to each of a plurality of remote users. An annotation template is provided to each of the plurality of remote users. Annotations for the one or more image are received from each of the plurality of remote users. The annotations and the plurality of medical images are stored together.

Rapid cross-validated ground truth annotation of large image datasets for image analytics

A mechanism is provided in a data processing system comprising a processor and a memory, the memory comprising instructions that are executed by the processor to specifically configure the processor to implement an intelligent medical image viewing engine. The intelligent medical image viewing engine receives a medical imaging study data structure comprising a plurality of electronic medical images from a medical image database. An image processing component executing within the intelligent medical image viewing engine analyzes the medical imaging study data structure to identify, for each electronic medical image in the plurality of electronic medical images, a corresponding set of image attributes. The intelligent medical image viewing engine receives a user input specifying at least one filter attribute for generating a medical image output and correlates the at least one filter attribute with at least one medical attribute of medical images to be used for selection of electronic medical images from the medical imaging study data structure. An image selection component executing within the intelligent medical image viewing engine selects a subset of the electronic medical images in the plurality of electronic medical images based on the correlation of the at least one filter attribute with the at least one medical attribute. A user interface generation component executing within the intelligent medical image viewing engine generates and outputs a medical image output comprising the subset of electronic medical images based on the selection.

Intelligently organizing displays of medical imaging content for rapid browsing and report creation

Embodiments of the present invention relate to discriminating between normal and abnormal left ventricles in echocardiography. In some embodiments, a first region of a first image of a heart is located by clustering. The first region depicts a chamber of the heart. The chamber has a chamber shape. A boundary of the first region is determined by contour tracing a retaining the traced contour overlapping the first region. A predetermined shape is compared to the boundary. The predetermined shape had been determined without reference to the first image. A first plurality of parameters is determined that, when applied to the predetermined shape, conforms the predetermined shape to the boundary and minimizes at least one error between the predetermined shape and the boundary. An indication of normality or abnormality is determine from the first plurality of parameters.

Discriminating between normal and abnormal left ventricles in echocardiography

Automatic detection of disease from analysis of echocardiographer findings in echocardiogram videos is provided. In various embodiments, a plurality of medical images containing embedded text are read. The plurality of medical images are clustered into a plurality of clusters. Medical images of a first cluster of the plurality of clusters are ranked based on the frequency of measurement names within the medical images of the first cluster. A candidate tabular template is generated indicative of a layout of measurement name/value pairs within the medical images of the first cluster. According to the candidate tabular template, optical character recognition is applied to one of the plurality of medical images of the first cluster to extract candidate measurements. The candidate tabular template and the candidate measurements are presented to a user.

Automatic detection of disease from analysis of echocardiographer findings in echocardiogram videos

Growing community of inventors

San Jose, CA, United States of America

David James Beymer