Multi-functional computer-aided gastroscopy system optimized with integrated AI solutions and method
Inventors
He, Xuejian • Wang, Lu • LEUNG, Ping Shun
Assignees
Hong Kong Applied Science and Technology Research Institute ASTRI
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12062169-B2
Publication Date
2024-08-13
Expiration Date
Abstract
A multi-functional, computer-aided gastroscopy system optimized with integrated AI solutions is disclosed. The system makes use of multiple deep-learning neural models to achieve low latency and high-performance requirements for multiple tasks. The optimization is made at three levels: architectural, modular and functional level. At architectural level, the models are designed in such a way that it is able to accomplish HP infection classification and detection of some lesions for one inference in order to reduce computation costs. At modular level, as a sub-model of HP infection classification, the site recognition model is optimized with temporal information. It not only improves the performance of HP infection classification, but also plays important roles for lesion detection and procedure status determination. At functional level, the inference latency is minimized by configuration and resource aware optimization. Also at functional level, the preprocessing is speeded up by image resizing parallelization and unified preprocessing.
Core Innovation
The invention is a computer-aided gastroscopy system that analyzes a gastric image sequence obtained from a gastroscopy instrument. The system includes an AI image processing system comprising at least three cooperative modules at the architecture level, and the cooperative modules perform image quality assessment, lesion detection, cancer identification, HP classification, and site recognition.
At least one module comprises one or more neural models that extract different but related information from the gastric image sequence. The extracted information is shared with other modules so that the overall AI image processing system uses inter-module information sharing rather than independent processing. The neural model that performs HP-plus-site processing fuses HP infection features and site information extracted from other neural models together to boost the classification accuracy of the computer-aided gastroscopy system.
The disclosed approach further specifies an architectural and resource-aware optimization of the AI pipeline, including unified preprocessing and optimization for inference latency. The system uses neural model pruning and quantization, dynamic batching in offline processing mode, and latency control in online processing mode through prioritized model loading and bounded task queues for parallel coprocessor execution.
Claims Coverage
Independent claim clm-00001 is directed to an architecture-level, cooperative multi-module AI image processing system for computer-aided gastroscopy. The claim defines three principal inventive features: cooperative neural feature extraction with information sharing across modules, fused HP infection features plus site information modeling, and resource-aware execution for offline and online latency control.
Cooperative multi-module gastroscopy analysis architecture
An AI image processing system analyzing a gastric image sequence obtained from a gastroscopy instrument, comprising at least three architecture-level modules cooperatively performing image quality assessment, lesion detection, cancer identification, HP classification, and site recognition.
Neural model feature extraction with cross-module sharing
A neural model extracting different but related information from the gastric image sequence and sharing the extracted information with other modules.
Fusing HP infection features with site information to boost classification accuracy
At least one neural model fusing HP infection features and site information extracted from other neural models together to boost the classification accuracy of the computer-aided gastroscopy system.
Unified preprocessing and threshold-triggered parallelized resizing
A unified preprocessing approach for generating an output tensor used by multiple models, including a conditional parallelized resizing operation that pads odd dimensions with zeros, splits into four quadrants, resizes quadrants in parallel, and stitches them into a uniformly resized image when image height or width exceeds a threshold.
Latency control via pruning/quantization, offline dynamic batching, and online prioritized coprocessor scheduling
Execution constraints including neural model pruning and quantization, dynamic batching in an offline processing mode based on available coprocessor computation capability and resources, and an online processing mode latency control procedure using prioritized loading of optimized neural models and fixed-length task queues for parallel coprocessor execution when a coprocessor becomes idle.
The claim set covers a cooperative multi-module gastroscopy AI pipeline with neural models that share extracted information and an HP-plus-site fused neural modeling component that boosts classification accuracy, combined with preprocessing and latency-oriented resource-aware execution in offline and online modes.
Stated Advantages
Boosts classification accuracy of the computer-aided gastroscopy system.
Improves HP classification sensitivity when adding the LSTM-based temporal modeling component.
Reduces site-transition false positives.
Reduces latency through resource-aware optimization including unified preprocessing, model pruning/quantization, dynamic batching, and online latency control.
Documented Applications
Computer-aided gastroscopy analysis of a gastric image sequence obtained from a gastroscopy instrument, including image quality assessment, lesion detection, cancer identification, HP classification, and site recognition.
Interested in licensing this patent?