US-20260127223-A1 - EFFICIENT CHANGE DETECTION BASED ON DYNAMIC INFORMATION RETRIEVAL

US20260127223A1US 20260127223 A1US20260127223 A1US 20260127223A1US-20260127223-A1

Abstract

One or more systems, devices, computer program products and/or computer-implemented methods of use provided herein relate to efficient change detection based on dynamic information retrieval by employing foundation models. For example, according to an embodiment, a system is provided. The system can comprise a memory that can store computer executable components. The system can further comprise a processor that can execute the computer executable components stored in the memory, where the computer executable components can comprise a data access component that can accesses multi-dimensional query data and features of interest (FOI) data related to the multi-dimensional query data. The computer executable components can further comprise an artificial intelligence (AI) component that can identify, based on the FOI data, one or more key frames within the multi-dimensional query data.

Inventors

Amitay Lev
Tamir Yehuda Shazman
Nati Daniel
Hannah Michele Ornstein
Doron Shaked
Angeles Perez-Agosto
Gopal Biligeri Avinash

Assignees

GE Precision Healthcare LLC

Dates

Publication Date: 20260507
Application Date: 20241106

Claims (20)

1 . A system, comprising: a memory that stores computer executable components; and a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise: a data access component that accesses multi-dimensional query data and features of interest (FOI) data related to the multi-dimensional query data; a data parsing component that parses the multi-dimensional query data into a sequence of two-dimensional (2D) images; an artificial intelligence (AI) component that identifies, based on the FOI data and the sequence of 2D images, one or more key frames within the sequence of 2D images by detecting differences between a first frame comprised in the sequence of 2D images and respective subsequent frames comprised in the sequence of 2D images, wherein the first frame is a query image provided for each of the respective subsequent frames, and wherein the detecting comprises: generating, by the AI component, via a dynamic image retrieval (DIR) decoder, output image embeddings of the first frame and output image embeddings of the respective subsequent frames; and detecting, by the AI component, via a change point detection and level analysis module, a subsequent frame comprised in the respective subsequent frames as a key frame if a similarity value between the first frame and the subsequent frame is less than a similarity threshold, wherein the similarity threshold represents a degree of similarity between the output image embeddings of the first frame and the output image embeddings of the respective subsequent frames, wherein the detecting further comprises: adjusting, by the AI component, via the change point detection and level analysis module, the similarity threshold based on a range of similarity values associated with respective combinations of frames, wherein the similarity value is comprised within the range of similarity values, wherein each combination of frames of the respective combinations of frames comprises the first frame and a different subsequent frame of the respective subsequent frames; and generating, by the AI component, a set of key frames comprising respective key frames corresponding to the respective combinations of frames.
2 . (canceled)
3 . The system of claim 1 , wherein the detecting the one or more key frames further comprises: extracting, by a redundant key frames component, the one or more key frames from the set of key frames, wherein the extracting the one or more key frames comprises: filtering, by the redundant key frames component, the set of key frames; and eliminating, by the redundant key frames component, redundant key frames that comprise redundant information.
4 . The system of claim 1 , wherein the multi-dimensional query data comprises a sequence of medical images having a format selected from a group consisting of 2D data, three-dimensional (3D) data and four-dimensional (4D) data.
5 . The system of claim 1 , wherein the AI component is a change detection-based foundation model that further generates, in an automated manner, annotations based on the one or more key frames, and wherein the data parsing component parses the multi-dimensional query data into the sequence of 2D images while maintaining a data sequence of the multi-dimensional query data.
6 . The system of claim 1 , wherein the FOI data comprises local FOI data, wherein the local FOI data has a format selected from a group consisting of textual data, mask data, points and bounding boxes.
7 . The system of claim 1 , wherein the FOI data comprises global FOI data, wherein the global FOI data comprises images marked as positive images or negative images.
8 . The system of claim 1 , wherein the FOI data is provided as a prompt by an entity via a graphical user interface (GUI), wherein the FOI data directs the AI component to focus areas within respective images comprised within the sequence of 2D images, and wherein identifying the one or more key frames based on the FOI data reduces an amount of time spent by the entity in inspecting large temporal data.
9 . A computer-implemented method, comprising: accessing, by a system operatively coupled to a processor, multi-dimensional query data, FOI data related to the multi-dimensional query data and a cloud database; parsing, by the system, the multi-dimensional query data into a sequence of 2D images; identifying, by the system, from the cloud database, based on the FOI data and the sequence of 2D images, a set of images that are similar to the multi-dimensional query data by detecting differences between a first frame comprised in a rank filtered database and respective subsequent frames comprised in the rank filtered database, wherein the rank filtered database is derived from the multi-dimensional query data, and wherein the detecting comprises: generating, by the system, via a dynamic image retrieval (DIR) decoder, output image embeddings of the first frame and output image embeddings of the respective subsequent frames; and detecting, by the system, via a change point detection and level analysis module, a subsequent frame comprised in the respective subsequent frames as a key frame if a similarity value between the first frame and the subsequent frame is less than a similarity threshold, wherein the similarity threshold represents a degree of similarity between the output image embeddings of the first frame and the output image embeddings of the respective subsequent frames, wherein the detecting further comprises: adjusting, by the system, via the change point detection and level analysis module, the similarity threshold based on a range of similarity values associated with respective combinations of frames, wherein the similarity value is comprised within the range of similarity values, wherein each combination of frames of the respective combinations of frames comprises the first frame and a different subsequent frame of the respective subsequent frames; and generating, by the system, a set of key frames comprising respective key frames corresponding to the respective combinations of frames.
10 . The computer-implemented method of claim 9 , wherein the cloud database comprises first images and embeddings of the first images with varying levels of similarity to the multi-dimensional query data.
11 . The computer-implemented method of claim 10 , wherein the identifying the set of images comprises: generating, by the system, based on the first images, the rank filtered database, wherein the rank filtered database comprises second images having a first level of similarity to the multi-dimensional query data.
12 . (canceled)
13 . The computer-implemented method of claim 9 , wherein the detecting further comprises: extracting, by the system, the set of images from the set of key frames, wherein the extracting comprises: filtering, by the system, the set of key frames; and retaining, by the system, key frames having a defined level of similarity to the multi-dimensional query data.
14 . The computer-implemented method of claim 9 , wherein the multi-dimensional query data comprises a sequence of medical images having a format selected from a group consisting of 2D data, 3D data and 4D data.
15 . The computer-implemented method of claim 9 , further comprising: the parsing, by the system, the multi-dimensional query data into the sequence of 2D images while maintaining a data sequence of the multi-dimensional query data.
16 . The computer-implemented method of claim 9 , wherein the FOI data comprises local FOI data, wherein the local FOI data has a format selected from a group consisting of textual data, mask data, points and bounding boxes.
17 . The computer-implemented method of claim 9 , wherein the FOI data comprises global FOI data, wherein the global FOI data comprises images marked as positive images or negative images.
18 . The computer-implemented method of claim 9 , wherein the set of images are identified by an AI component, wherein the DIR decoder is comprised in the AI component, and wherein the DIR decoder is dynamically updated during identification of the set of images.
19 . A computer program product comprising a non-transitory computer readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to: access multi-dimensional query data, FOI data related to the multi-dimensional query data and a cloud database; parse the multi-dimensional query data into a sequence of 2D images; identify, based on the FOI data and the sequence of 2D images, one or more key frames within the sequence of 2D images by detecting differences between a window comprised in a rank filtered database and respective subsequent frames comprised in the rank filtered database, wherein the detecting comprises: generating, by the processor, via a DIR decoder, output image embeddings of the window and output image embeddings of the respective subsequent frames; and detecting, by the processor, via a change point detection and level analysis module, a subsequent frame comprised in the respective subsequent frames as a key frame if a similarity value between the window and the subsequent frame is less than a similarity threshold, wherein the similarity threshold represents a degree of similarity between the output image embeddings of the window and the output image embeddings of the respective subsequent frames, wherein the detecting further comprises: adjusting, by the processor, via the change point detection and level analysis module, the similarity threshold based on a range of similarity values associated with respective combinations of frames, wherein the similarity value is comprised within the range of similarity values, wherein each combination of frames of the respective combinations of frames comprises the window and a different subsequent frame of the respective subsequent frames; generating, by the processor, a set of key frames comprising respective key frames corresponding to the respective combinations of frames; and extracting by the processor, the one or more key frames from the set of key frames.
20 . (canceled)

Description

BACKGROUND The subject disclosure relates to artificial intelligence (AI) and, more specifically, to efficient change detection based on dynamic information retrieval by employing foundation models. SUMMARY The following presents a summary to provide a basic understanding of one or more embodiments described herein. This summary is not intended to identify key or critical elements, delineate scope of particular embodiments or scope of claims. Its sole purpose is to present concepts in a simplified form as a prelude to the more detailed description that is presented later. In one or more embodiments described herein, systems, computer-implemented methods, apparatus and/or computer program products that enable efficient change detection based on dynamic information retrieval by employing foundation models are discussed. According to an embodiment, a system is provided. The system can comprise a memory that can store computer executable components. The system can further comprise a processor that can execute the computer executable components stored in the memory, where the computer executable components can comprise a data access component that can accesses multi-dimensional query data and features of interest (FOI) data related to the multi-dimensional query data. The computer executable components can further comprise an AI component that can identify, based on the FOI data, one or more key frames within the multi-dimensional query data. According to another embodiment, a computer-implemented method is provided. The computer-implemented method can comprise accessing, by a system operatively coupled to a processor, multi-dimensional query data, FOI data related to the multi-dimensional query data and a cloud database. The computer-implemented method can further comprise identifying, by the system, from the cloud database, based on the FOI data, a set of images that are similar to the multi-dimensional query data. According to yet another embodiment, a computer program product is provided. The computer program product can comprise a non-transitory computer readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to access multi-dimensional query data, FOI data related to the multi-dimensional query data and a cloud database. The program instructions can be further executable by the processor to cause the processor to identify, based on the FOI data, one or more key frames within the multi-dimensional query data. BRIEF DESCRIPTION OF THE DRAWINGS One or more embodiments are described below in the Detailed Description section with reference to the following drawings: FIG. 1 illustrates a block diagram of an example, non-limiting system that can employ dynamic information retrieval and foundation models to identify key frames in a data sequence or video or retrieve images from a cloud database, in accordance with one or more embodiments described herein. FIG. 2 illustrates another block diagram of an example, non-limiting system that can employ dynamic information retrieval and foundation models to identify key frames in a data sequence or video or retrieve images from a cloud database, in accordance with one or more embodiments described herein. FIG. 3 illustrates a flow diagram of an example, non-limiting method that can employ dynamic information retrieval and foundation models to identify key frames in a data sequence or video or retrieve images from a cloud database, in accordance with one or more embodiments described herein. FIG. 4 illustrates a block diagram of an example, non-limiting architecture of an AI component that can employ dynamic information retrieval and foundation models to identify key frames in a data sequence or video or retrieve images from a cloud database, in accordance with one or more embodiments described herein. FIGS. 5 and 6 illustrate diagrams of an example, non-limiting method that can employ the architecture of the AI component illustrated in FIG. 4 to efficiently detect change in data, in accordance with one or more embodiments described herein. FIG. 7 illustrates a diagram of an example, non-limiting graph of cosine similarities, in accordance with one or more embodiments described herein. FIG. 8 illustrates a diagram of an example, non-limiting scenario wherein key frames can be detected in medical image data, in accordance with one or more embodiments described herein. FIG. 9 illustrates a flow diagram of an example, non-limiting method wherein medical images can be retrieved from the cloud, in accordance with one or more embodiments described herein. FIG. 10 illustrates another flow diagram of an example, non-limiting method wherein medical images can be retrieved from the cloud, in accordance with one or more embodiments described herein. FIG. 11 illustrates flow diagrams of example, non-limiting methods that can employ a foundation model to detect key frames in multi-dimensional query data,