Medical image-based tumor detection and diagnostic device
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11568538-B2
Publication Date
2023-01-31
Expiration Date
Abstract
A device and a method for detecting a tumor using a medical image and diagnosing a shape and a property of the detected tumor are disclosed. An exemplary medical image-based tumor detection and diagnostic device includes: an input unit configured to obtain a medical image related to a patient; a preprocessing unit configured to preprocess the obtained medical image to observe a tumor region; an analysis unit configured to divide the preprocessed image into a plurality of regions by applying a deep neural network-based deep learning technique; and a measurement unit configured to group the plurality of divided regions by performing clustering on the plurality of divided regions. The measurement unit extracts a group feature value in respect to each of the grouped regions and derives diagnosis information related to the tumor based on the extracted group feature value.
Core Innovation
A medical image-based tumor detection and diagnostic device obtains a medical image related to a patient, preprocesses the obtained medical image to observe a tumor region, and divides a preprocessed image into a plurality of regions by applying a deep neural network-based deep learning technique. The device includes an analysis unit for region division and a measurement unit for grouping divided regions by performing clustering on the plurality of divided regions.
The medical images are processed as three-dimensional images, and preprocessing includes resampling based on voxel resolution to correct resolution, normalizing a resampled image, and adjusting a size of the normalized image to apply the deep neural network-based deep learning technique. The image with an adjusted size is divided into a plurality of regions so that an output is equal in magnitude to an input of the deep neural network.
The measurement unit extracts a group feature value with respect to each grouped region and derives diagnosis information related to a tumor based on the extracted group feature value. The device further obtains body information of the patient, derives the diagnosis information using both the group feature value and the body information, and includes reverse mapping to match grouped results back into an original image space and stores matrix information for grouped regions and associated feature/prognosis results.
Claims Coverage
The document identifies one independent claim (clm-00001). It includes seven core inventive features: obtaining a patient medical image; preprocessing to observe a tumor region; 3D deep neural network-based region division; clustering-based voxel-to-object grouping and group feature extraction; diagnosis information derivation from group feature values; diagnosis information derivation using group features and body information; and voxel-resolution resampling, normalization, and resizing for deep neural network input. Dependent claims further refine neural network architecture, reverse mapping/storage, and tumor/cancer types.
Patient medical image acquisition and tumor-region preprocessing
An input unit configured to obtain a medical image related to a patient, and a preprocessing unit configured to preprocess an obtained medical image to observe a tumor region.
3D deep neural network-based region division
An analysis unit configured to divide a preprocessed image into a plurality of regions by applying a deep neural network-based deep learning technique, wherein an image inputted to the deep neural network and an image related to the plurality of divided regions are three-dimensional images.
Clustering-based voxel-to-object grouping and group feature extraction
A measurement unit configured to group a plurality of divided regions by performing a clustering on the plurality of divided regions, wherein the measurement unit performs the clustering to process the tumor from a voxel level to an object level and extracts the group feature value in respect to each of the grouped regions.
Diagnosis information derivation from group feature values
The measurement unit derives diagnosis information related to a tumor based on an extracted group feature value, wherein the measurement unit quantifies the extracted group feature value and the diagnosis information includes information on a location, a region, a diameter, and a volume of the tumor.
Diagnosis information derivation using group features and body information
The input unit further obtains body information of the patient, and the measurement unit derives the diagnosis information using both the group feature value and the body information.
Voxel-resolution resampling, normalization, and resizing for deep neural network input
The analysis unit resamples the preprocessed image based on voxel resolution to correct resolution, normalizes a resampled image, adjusts a size of a normalized image to apply the deep neural network-based deep learning technique, and divides an image with an adjusted size into a plurality of regions to implement an output equal in magnitude to an input of the deep neural network.
Reverse mapping of grouped results to original image space and storage
A storage unit that reverse-processes grouped matrix information from a measurement unit to map it back into the original medical image space from an input unit and then stores the matrix information.
Overall claim coverage centers on obtaining and preprocessing a patient-related medical image into a tumor-observable 3D representation, dividing it into regions using a deep neural network, clustering the regions from voxel level to object level to extract group feature values, and deriving tumor diagnosis information using both image-derived group feature values and patient body information. Dependent concepts add reverse processing and storage back to original image space, while also enumerating tumor and cancer types.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?