Device and method for detecting cerebral microbleeds using magnetic resonance images
Inventors
Kim, Dong Hyun • CHOI, Sang Hyeok • AL-MASNI, Mohammed • NOH, Young • KIM, Eung Yeop
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11481902-B2
Publication Date
2022-10-25
Expiration Date
Abstract
A device and method for detecting cerebral microbleeds use magnetic resonance images. The disclosed device includes a preprocessing unit that normalizes an SWI image and a phase image, respectively, of the magnetic resonance images, and performs phase image conversion for inverting a code of the normalized phase image, a YOLO neural network module that receives a two-channel image in which the preprocessed SWI image and phase image are concatenated and detects a plurality of candidate regions for the cerebral microbleeds, and a cerebral microbleeds determination neural network module that receives patch images of candidate regions of the SWI image and phase image based on the candidate regions and determines whether the patch images of each candidate region are an image with a symptom of the cerebral microbleeds through a neural network operation.
Core Innovation
The invention detects cerebral microbleeds using magnetic resonance images by preprocessing a sensitivity-weighted imaging (SWI) image and a phase image. The preprocessing unit normalizes the SWI image and the phase image, and performs phase image conversion for inverting a code of the normalized phase image. The preprocessed SWI image and the converted phase image are used as inputs for subsequent neural network operations.
The invention performs a YOLO neural network operation using a YOLO neural network module. The YOLO neural network module receives a two-channel image in which the preprocessed SWI image and phase image are concatenated, and detects a plurality of candidate regions for the cerebral microbleeds. The detected plurality of candidate regions and probability information associated with bounding boxes are used to guide further determination.
For each candidate region, the invention performs a cerebral microbleeds determination neural network operation using a cerebral microbleeds determination neural network module. The determination neural network module receives patch images of the candidate regions of the SWI image and phase image based on the plurality of candidate regions. Through a neural network operation, the determination neural network module determines whether the patch images of each candidate region are an image with a symptom of the cerebral microbleeds.
Claims Coverage
The document provides two independent claims (device and method), each built from the same three core inventive features: preprocessing with normalization and phase code inversion; YOLO candidate-region detection using a two-channel concatenated SWI+phase input; and neural-network determination using patch images from the candidate regions.
Preprocessing with SWI normalization and phase code inversion
A preprocessing unit that normalizes a sensitivity-weighted imaging (SWI) image and a phase image of the magnetic resonance images, respectively, and performs phase image conversion for inverting a code of the normalized phase image.
YOLO candidate-region detection from concatenated two-channel SWI and phase
A You Only Look Once (YOLO) neural network module that receives a two-channel image in which the preprocessed SWI image and phase image are concatenated and detects a plurality of candidate regions for the cerebral microbleeds.
Neural determination of microbleed symptom from candidate-region patch images
A cerebral microbleeds determination neural network module that receives patch images of the candidate regions of the SWI image and phase image based on the plurality of candidate regions and determines whether the patch images of each candidate region are an image with a symptom of the cerebral microbleeds through a neural network operation.
Across the independent device and method claims, detection is performed by preprocessing that normalizes SWI and phase and inverts the code of the normalized phase, YOLO detection of candidate regions from a concatenated two-channel SWI+phase input, and determination of each candidate-region patch image as having a symptom of cerebral microbleeds.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?