CN116051509A - Brain medical image spectrum processing method based on deep learning - Google Patents
Brain medical image spectrum processing method based on deep learning Download PDFInfo
- Publication number
- CN116051509A CN116051509A CN202310050576.8A CN202310050576A CN116051509A CN 116051509 A CN116051509 A CN 116051509A CN 202310050576 A CN202310050576 A CN 202310050576A CN 116051509 A CN116051509 A CN 116051509A
- Authority
- CN
- China
- Prior art keywords
- brain
- image
- spectrum
- segmentation
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
- A61B5/0075—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
- A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
- A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2576/00—Medical imaging apparatus involving image processing or analysis
- A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
- A61B2576/026—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the brain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Abstract
The invention belongs to the technical field of medical image spectrum processing, and discloses a brain medical image spectrum processing method based on deep learning, wherein the brain medical image spectrum processing system based on deep learning comprises the following steps: the brain spectrum image acquisition module, the central control module, the image feature extraction module, the image segmentation module, the brain diagnosis module, the image storage module and the display module. According to the invention, the low-level feature extractor is adaptively learned from the brain spectrum image data by the image feature extraction module, namely, the brain spectrum image features can be adaptively extracted, the extraction efficiency is higher, and the problems that the self-adaptive extraction of the features cannot be carried out for each brain spectrum image and the extraction time is too long in the prior art are solved; meanwhile, the image segmentation module is used for extracting the characteristics of the brain spectrum image to determine a target segmentation algorithm corresponding to the brain spectrum image, so that the technical problem of poor image segmentation effect in the prior art is solved.
Description
Technical Field
The invention belongs to the technical field of medical image spectrum processing, and particularly relates to a brain medical image spectrum processing method based on deep learning.
Background
The hyperspectral imaging technology is based on an image data technology with a very large number of narrow wave bands, combines the imaging technology with a spectrum technology, detects two-dimensional geometric space and one-dimensional spectrum information of a target, acquires continuous and narrow wave band image data with high spectral resolution, has rapid development, and commonly comprises grating light splitting, acousto-optic tunable filter light splitting, prism light splitting, chip coating and the like, and is mainly applied to the fields of food safety, medical diagnosis, aerospace and the like. Imaging spectrometers can be divided into two types, depending on their structure. An imaging spectrometer is an area array detector and a push-broom scanner, which scans by using a line array detector and completes spectrum scanning by using a dispersive element and the area array detector. Spatial scanning is accomplished with a line array detector and its movement in the track direction. The other is an imaging spectrometer using a linear array detector and a photo-mechanical scanner, which collects spectral information by using a point detector, divides the spectral information into different wave bands after passing through a dispersive element, and completes space scanning by the point scanning mirror swinging in a plane vertical to the track direction and moving along the track direction on different elements of the linear array detector respectively, and completes spectral scanning by using the linear detector. Because of the structural limitations of the two imaging spectrometers, the imaging spectrometers take too long time for extracting the spectral image features of the brain; meanwhile, the image segmentation effect is poor. Through the above analysis, the problems and defects existing in the prior art are as follows:
(1) The existing imaging spectrometer takes too long time to extract the spectral image features of the brain.
(2) The image segmentation effect is poor.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides a brain medical image spectrum processing method based on deep learning.
The invention discloses a brain medical image spectrum processing method based on deep learning, which comprises the following steps:
step one, acquiring brain spectrum images by using medical spectrum equipment through a brain spectrum image acquisition module;
step two, the central control module extracts brain spectrum image characteristics through an image characteristic extraction module;
dividing the brain spectrum image characteristics through an image dividing module; diagnosing the brain according to the brain spectral image by a brain diagnosis module;
storing the brain spectrum image, the characteristics and the diagnosis result through an image storage module; and displaying the brain spectrum image, the characteristics and the diagnosis result through a display module.
Another object of the present invention is to provide a brain medical image spectrum processing system based on deep learning, including:
the brain spectrum image acquisition module is used for acquiring brain spectrum images through medical spectrum equipment;
the image feature extraction module is used for extracting brain spectrum image features;
the image segmentation module is used for segmenting the brain spectrum image characteristics;
the brain diagnosis module is used for diagnosing the brain according to the brain spectrum image;
the image storage module is used for storing brain spectrum images, characteristics and diagnosis results;
the display module is used for displaying brain spectrum images, characteristics and diagnosis results;
the central control module is connected with the brain spectrum image acquisition module, the image feature extraction module, the image segmentation module, the brain diagnosis module, the image storage module and the display module and used for controlling the normal work of each module.
Further, the brain diagnosis module diagnosis method is as follows:
acquiring a brain three-dimensional image with a designated mode, and marking an artificial focus on the brain three-dimensional image; transforming the brain and the corresponding label in the brain three-dimensional image to the center position of the three-dimensional image by a three-dimensional space transformation method of simulated annealing, and enabling the middle axis surface of the brain to coincide with the middle vertical surfaces of the left horizontal axis and the right horizontal axis to carry out symmetry correction;
testing a mainstream deep learning algorithm, a network model, a brain image analysis and detection framework and a data set, and analyzing the advantages and disadvantages of various deep learning algorithms and convolution network models in the aspects of image segmentation, target detection and disease classification and grading;
aiming at the requirements and characteristics of brain image analysis, selecting a certain number of brain images aiming at typical brain diseases as samples, collecting brain images of various types and different angles for the same medical record, and filling multi-level classification labels;
based on a specific type of brain image data set and application requirements for detecting, classifying and grading focuses, respectively training different deep learning convolutional neural networks to obtain preliminary results of focus target segmentation detection, classification and grading for a specific brain image, and then processing a plurality of preliminary results by constructing a weighted Bayesian network to obtain a final analysis diagnosis result.
Further, the brain medical image spectrum processing method based on deep learning comprises the following steps:
further, the image feature extraction module extracts the following steps:
(1) Constructing a brain spectrum image set; using a clustering algorithm to acquire a plurality of clustering centers from the brain spectrum image dataset to be classified as a low-level feature extractor; performing convolution operation on each brain spectrum image in the brain spectrum image dataset by using the plurality of low-level feature extractors, and generating a plurality of convolution brain spectrum images with the same quantity as the plurality of low-level feature extractors for each brain spectrum image;
(2) Thresholding is carried out on the plurality of convolution brain spectrum images respectively to obtain a plurality of sparse brain spectrum images; performing a normalization operation on the plurality of sparse brain spectral images, respectively, the normalization operation comprising: forming a vector by pixel values of the same position of each brain spectrum image in the plurality of sparse brain spectrum images, normalizing the vector, and then respectively returning each component of the vector to the corresponding position of each brain spectrum image to obtain a plurality of normalized sparse brain spectrum images;
(3) Performing low-level feature integration on the plurality of sparse brain spectrum images to obtain a plurality of integrated brain spectrum images; and performing middle-layer feature extraction operation on the plurality of integrated brain spectrum images to obtain middle-layer features.
Further, before the clustering algorithm is used to obtain a plurality of clustering centers from the brain spectrum image dataset to be classified as the low-level feature extractor, the method further includes:
and carrying out normalization and decoupling preprocessing operation on the brain spectrum images in the brain spectrum image data set to obtain the brain spectrum image data set to be classified.
Further, the thresholding the plurality of convolved brain spectral images to obtain a plurality of sparse brain spectral images includes:
and judging each pixel value of each convolution brain spectrum image in the convolution brain spectrum images, if the pixel value is larger than a preset threshold value, reserving the pixel value, otherwise, setting the pixel value to 0, and correspondingly generating a sparse brain spectrum image by the pixel value after thresholding operation of each convolution brain spectrum image to obtain a plurality of sparse brain spectrum images.
Further, performing low-level feature integration on the plurality of sparse brain spectral images to obtain a plurality of integrated brain spectral images, including:
dividing each sparse brain spectrum image in the plurality of sparse brain spectrum images into a plurality of m multiplied by m areas, respectively forming a plurality of pixel values of the areas into m 2-dimensional vectors, forming a plurality of integrated brain spectrum images by the pixel values of the same positions of the vectors, wherein m is an integer greater than or equal to 2, and the number of the integrated brain spectrum images is m2 times that of the sparse brain spectrum images.
Further, the image segmentation module segments the following method:
carrying out enhancement treatment on the brain spectrum image; selecting a target segmentation algorithm corresponding to the brain spectrum image from candidate segmentation algorithms by extracting features of the brain spectrum image,
wherein the candidate segmentation algorithm comprises: a saliency segmentation algorithm and a semantic segmentation algorithm;
performing rough segmentation on the brain spectrum image by using the target segmentation algorithm to obtain a first segmentation result, wherein the rough segmentation is used for preliminarily determining a target object in the brain spectrum image;
and carrying out fine segmentation on the first segmentation result to obtain a target object in the brain spectrum image.
Further, the selecting a target segmentation algorithm corresponding to the brain spectrum image from candidate segmentation algorithms by extracting features of the brain spectrum image includes:
inputting the brain spectrum image into a backbone network in a segmentation model, and assigning the brain spectrum image to a corresponding node by the backbone network through feature extraction of the brain spectrum image, wherein the node comprises: a first node corresponding to each category of which the number of the sample brain spectrum images exceeds a preset value, and a second node corresponding to all categories of which the sample brain spectrum images are smaller than or equal to the preset value;
if the brain spectrum image is distributed to the first node, determining the target segmentation algorithm as the semantic segmentation algorithm;
and if the brain spectrum image is distributed to the second node, determining the target segmentation algorithm as the significance segmentation algorithm.
Further, the segmentation model also includes a segmentation network, the segmentation network comprising: the semantic segmentation network and the saliency segmentation network perform rough segmentation on the brain spectrum image by using the target segmentation algorithm to obtain a first segmentation result, wherein the method comprises the following steps:
the first feature layer output by the backbone network is input to a segmentation network corresponding to a target segmentation algorithm;
performing rough segmentation on the brain spectrum image through the segmentation network;
further, in the case where the target segmentation algorithm is a saliency segmentation algorithm, performing coarse segmentation on the brain spectral image through the segmentation network includes:
stacking a plurality of second characteristic layers according to a preset layer jump rule to obtain a plurality of single-channel prediction masks, wherein the second characteristic layers are obtained by carrying out convolution and deconvolution on the first characteristic layers output by the backbone network;
acquiring first linear average values of the plurality of single-channel prediction masks;
determining the first linear mean as a first segmentation result corresponding to the brain spectral image;
extracting n single-channel prediction masks from the plurality of single-channel prediction masks, wherein n is a positive integer less than the number of the single-channel prediction masks;
acquiring second linear average values of the n single-channel prediction masks;
and determining the second linear mean as a first segmentation result corresponding to the brain spectrum image.
In combination with the above technical solution and the technical problems to be solved, please analyze the following aspects to provide the following advantages and positive effects:
first, aiming at the technical problems in the prior art and the difficulty in solving the problems, the technical problems solved by the technical proposal of the invention are analyzed in detail and deeply by tightly combining the technical proposal to be protected, the results and data in the research and development process, and the like, and some technical effects brought after the problems are solved have creative technical effects. The specific description is as follows:
the method comprises the steps that a plurality of clustering centers are obtained from a brain spectrum image dataset to be classified through an image feature extraction module by using a clustering algorithm and serve as a low-level feature extractor; performing convolution operation on each brain spectrum image in the brain spectrum image dataset by using the plurality of low-level feature extractors, and respectively generating a plurality of convolution brain spectrum images with the same quantity as the plurality of low-level feature extractors; thresholding is carried out on the plurality of convolution brain spectrum images respectively to obtain a plurality of sparse brain spectrum images; performing low-level feature integration on the plurality of sparse brain spectrum images to obtain a plurality of integrated brain spectrum images; the middle-layer feature extraction operation is carried out on the plurality of integrated brain spectrum images to obtain middle-layer features, so that the low-layer feature extractor can be self-adaptively learned from brain spectrum image data, namely, the brain spectrum image features can be self-adaptively extracted, the extraction efficiency is higher, and the problems that the self-adaptive extraction of the features cannot be carried out for each brain spectrum image and the extraction time is too long in the prior art are solved; meanwhile, the image segmentation module is used for extracting the characteristics of the brain spectrum image to determine a target segmentation algorithm corresponding to the brain spectrum image, so that the technical problem of poor image segmentation effect in the prior art is solved.
Secondly, the technical scheme is regarded as a whole or from the perspective of products, and the technical scheme to be protected has the following technical effects and advantages:
the invention effectively solves the problems that the self-adaptive extraction of the characteristics of each brain spectrum image can not be realized and the extraction time is too long in the prior art, has higher extraction efficiency, short time consumption and higher accuracy, simultaneously solves the technical problem of poor image segmentation effect in the prior art, improves the accuracy of image segmentation, and obtains the final analysis diagnosis result by processing the analysis result of the brain image through the brain diagnosis module, thereby greatly improving the accuracy of the diagnosis result.
Drawings
Fig. 1 is a flowchart of a brain medical image spectrum processing method based on deep learning according to an embodiment of the present invention.
Fig. 2 is a block diagram of a brain medical image spectrum processing system based on deep learning according to an embodiment of the present invention.
Fig. 3 is a flowchart of an extraction method of an image feature extraction module according to an embodiment of the present invention.
In fig. 2: 1. the brain spectrum image acquisition module; 2. a central control module; 3. the image feature extraction module; 4. an image segmentation module; 5. a brain diagnostic module; 6. an image storage module; 7. and a display module.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
1. The embodiments are explained. In order to fully understand how the invention may be embodied by those skilled in the art, this section is an illustrative embodiment in which the claims are presented for purposes of illustration.
As shown in fig. 1, the brain medical image spectrum processing method based on deep learning provided by the invention comprises the following steps:
s101, acquiring brain spectrum images by using medical spectrum equipment through a brain spectrum image acquisition module;
s102, a central control module extracts brain spectrum image features through an image feature extraction module;
s103, segmenting the brain spectrum image features through an image segmentation module; diagnosing the brain according to the brain spectral image by a brain diagnosis module;
s104, storing brain spectrum images, characteristics and diagnosis results through an image storage module; and displaying the brain spectrum image, the characteristics and the diagnosis result through a display module.
As shown in fig. 2, the brain medical image spectrum processing system based on deep learning provided by the embodiment of the invention includes: the brain spectrum image acquisition module 1, the central control module 2, the image feature extraction module 3, the image segmentation module 4, the brain diagnosis module 5, the image storage module 6 and the display module 7.
The brain spectrum image acquisition module 1 is connected with the central control module 2 and is used for acquiring brain spectrum images through medical spectrum equipment;
the central control module 2 is connected with the brain spectrum image acquisition module 1, the image feature extraction module 3, the image segmentation module 4, the brain diagnosis module 5, the image storage module 6 and the display module 7 and used for controlling the normal work of each module;
the image feature extraction module 3 is connected with the central control module 2 and is used for extracting brain spectrum image features;
the image segmentation module 4 is connected with the central control module 2 and is used for segmenting the spectral image characteristics of the brain;
a brain diagnosis module 5 connected with the central control module 2 for diagnosing the brain according to brain spectral images;
the brain diagnosis module 5 diagnosis method provided by the embodiment of the invention is as follows:
acquiring a brain three-dimensional image with a designated mode, and marking an artificial focus on the brain three-dimensional image; transforming the brain and the corresponding label in the brain three-dimensional image to the center position of the three-dimensional image by a three-dimensional space transformation method of simulated annealing, and enabling the middle axis surface of the brain to coincide with the middle vertical surfaces of the left horizontal axis and the right horizontal axis to carry out symmetry correction;
testing a mainstream deep learning algorithm, a network model, a brain image analysis and detection framework and a data set, and analyzing the advantages and disadvantages of various deep learning algorithms and convolution network models in the aspects of image segmentation, target detection and disease classification and grading;
aiming at the requirements and characteristics of brain image analysis, selecting a certain number of brain images aiming at typical brain diseases as samples, collecting brain images of various types and different angles for the same medical record, and filling multi-level classification labels;
based on a specific type of brain image data set and application requirements for detecting, classifying and grading focuses, respectively training different deep learning convolutional neural networks to obtain preliminary results of focus target segmentation detection, classification and grading for a specific brain image, and then processing a plurality of preliminary results by constructing a weighted Bayesian network to obtain a final analysis diagnosis result;
the image storage module 6 is connected with the central control module 2 and is used for storing brain spectrum images, characteristics and diagnosis results;
and the display module 7 is connected with the central control module 2 and is used for displaying brain spectrum images, characteristics and diagnosis results.
As shown in fig. 3, the image feature extraction module 3 provided by the present invention has the following extraction method:
s201, constructing a brain spectrum image set; using a clustering algorithm to acquire a plurality of clustering centers from the brain spectrum image dataset to be classified as a low-level feature extractor; performing convolution operation on each brain spectrum image in the brain spectrum image dataset by using the plurality of low-level feature extractors, and generating a plurality of convolution brain spectrum images with the same quantity as the plurality of low-level feature extractors for each brain spectrum image;
s202, thresholding is carried out on the plurality of convolution brain spectrum images respectively to obtain a plurality of sparse brain spectrum images; performing a normalization operation on the plurality of sparse brain spectral images, respectively, the normalization operation comprising: forming a vector by pixel values of the same position of each brain spectrum image in the plurality of sparse brain spectrum images, normalizing the vector, and then respectively returning each component of the vector to the corresponding position of each brain spectrum image to obtain a plurality of normalized sparse brain spectrum images;
s203, performing low-level feature integration on the plurality of sparse brain spectrum images to obtain a plurality of integrated brain spectrum images; and performing middle-layer feature extraction operation on the plurality of integrated brain spectrum images to obtain middle-layer features.
Before the clustering algorithm is used to obtain a plurality of clustering centers from the brain spectrum image dataset to be classified as the low-level feature extractor, the method further comprises:
and carrying out normalization and decoupling preprocessing operation on the brain spectrum images in the brain spectrum image data set to obtain the brain spectrum image data set to be classified.
The thresholding the plurality of convolved brain spectral images to obtain a plurality of sparse brain spectral images includes:
and judging each pixel value of each convolution brain spectrum image in the convolution brain spectrum images, if the pixel value is larger than a preset threshold value, reserving the pixel value, otherwise, setting the pixel value to 0, and correspondingly generating a sparse brain spectrum image by the pixel value after thresholding operation of each convolution brain spectrum image to obtain a plurality of sparse brain spectrum images.
Performing low-level feature integration on the plurality of sparse brain spectral images to obtain a plurality of integrated brain spectral images, including:
dividing each sparse brain spectrum image in the plurality of sparse brain spectrum images into a plurality of m multiplied by m areas, respectively forming a plurality of pixel values of the areas into m 2-dimensional vectors, forming a plurality of integrated brain spectrum images by the pixel values of the same positions of the vectors, wherein m is an integer greater than or equal to 2, and the number of the integrated brain spectrum images is m2 times that of the sparse brain spectrum images.
The image segmentation module segmentation method comprises the following steps:
carrying out enhancement treatment on the brain spectrum image; selecting a target segmentation algorithm corresponding to the brain spectrum image from candidate segmentation algorithms by extracting features of the brain spectrum image,
wherein the candidate segmentation algorithm comprises: a saliency segmentation algorithm and a semantic segmentation algorithm;
performing rough segmentation on the brain spectrum image by using the target segmentation algorithm to obtain a first segmentation result, wherein the rough segmentation is used for preliminarily determining a target object in the brain spectrum image;
and carrying out fine segmentation on the first segmentation result to obtain a target object in the brain spectrum image.
The selecting a target segmentation algorithm corresponding to the brain spectrum image from candidate segmentation algorithms by extracting features of the brain spectrum image comprises the following steps:
inputting the brain spectrum image into a backbone network in a segmentation model, and assigning the brain spectrum image to a corresponding node by the backbone network through feature extraction of the brain spectrum image, wherein the node comprises: a first node corresponding to each category of which the number of the sample brain spectrum images exceeds a preset value, and a second node corresponding to all categories of which the sample brain spectrum images are smaller than or equal to the preset value;
if the brain spectrum image is distributed to the first node, determining the target segmentation algorithm as the semantic segmentation algorithm;
and if the brain spectrum image is distributed to the second node, determining the target segmentation algorithm as the significance segmentation algorithm.
The segmentation model further includes a segmentation network, the segmentation network including: the semantic segmentation network and the saliency segmentation network perform rough segmentation on the brain spectrum image by using the target segmentation algorithm to obtain a first segmentation result, wherein the method comprises the following steps:
the first feature layer output by the backbone network is input to a segmentation network corresponding to a target segmentation algorithm;
performing rough segmentation on the brain spectrum image through the segmentation network;
when the target segmentation algorithm is a saliency segmentation algorithm, performing rough segmentation on the brain spectrum image through the segmentation network, including:
stacking a plurality of second characteristic layers according to a preset layer jump rule to obtain a plurality of single-channel prediction masks, wherein the second characteristic layers are obtained by carrying out convolution and deconvolution on the first characteristic layers output by the backbone network;
acquiring first linear average values of the plurality of single-channel prediction masks;
determining the first linear mean as a first segmentation result corresponding to the brain spectral image;
extracting n single-channel prediction masks from the plurality of single-channel prediction masks, wherein n is a positive integer less than the number of the single-channel prediction masks;
acquiring second linear average values of the n single-channel prediction masks;
and determining the second linear mean as a first segmentation result corresponding to the brain spectrum image.
2. Application example. In order to prove the inventive and technical value of the technical solution of the present invention, this section is an application example on specific products or related technologies of the claim technical solution.
The method comprises the steps that a plurality of clustering centers are obtained from a brain spectrum image dataset to be classified through an image feature extraction module by using a clustering algorithm and serve as a low-level feature extractor; performing convolution operation on each brain spectrum image in the brain spectrum image dataset by using the plurality of low-level feature extractors, and respectively generating a plurality of convolution brain spectrum images with the same quantity as the plurality of low-level feature extractors; thresholding is carried out on the plurality of convolution brain spectrum images respectively to obtain a plurality of sparse brain spectrum images; performing low-level feature integration on the plurality of sparse brain spectrum images to obtain a plurality of integrated brain spectrum images; the middle-layer feature extraction operation is carried out on the plurality of integrated brain spectrum images to obtain middle-layer features, so that the low-layer feature extractor can be self-adaptively learned from brain spectrum image data, namely, the brain spectrum image features can be self-adaptively extracted, the extraction efficiency is higher, and the problems that the self-adaptive extraction of the features cannot be carried out for each brain spectrum image and the extraction time is too long in the prior art are solved; meanwhile, the image segmentation module is used for extracting the characteristics of the brain spectrum image to determine a target segmentation algorithm corresponding to the brain spectrum image, so that the technical problem of poor image segmentation effect in the prior art is solved.
It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.
The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.
Claims (10)
1. The brain medical image spectrum processing method based on deep learning is characterized by comprising the following steps of:
step one, acquiring brain spectrum images by using medical spectrum equipment through a brain spectrum image acquisition module;
step two, the central control module extracts brain spectrum image characteristics through an image characteristic extraction module;
dividing the brain spectrum image characteristics through an image dividing module; diagnosing the brain according to the brain spectral image by a brain diagnosis module;
storing the brain spectrum image, the characteristics and the diagnosis result through an image storage module; and displaying the brain spectrum image, the characteristics and the diagnosis result through a display module.
A brain medical image spectral processing system based on deep learning, characterized in that the brain medical image spectral processing system based on deep learning comprises:
the brain spectrum image acquisition module is used for acquiring brain spectrum images through medical spectrum equipment;
the image feature extraction module is used for extracting brain spectrum image features;
the image segmentation module is used for segmenting the brain spectrum image characteristics;
the brain diagnosis module is used for diagnosing the brain according to the brain spectrum image;
the image storage module is used for storing brain spectrum images, characteristics and diagnosis results;
the display module is used for displaying brain spectrum images, characteristics and diagnosis results;
the central control module is connected with the brain spectrum image acquisition module, the image feature extraction module, the image segmentation module, the brain diagnosis module, the image storage module and the display module and used for controlling the normal work of each module.
2. The brain medical image spectrum processing method based on deep learning according to claim 1, wherein the brain diagnosis module diagnosis method is as follows: acquiring a brain three-dimensional image with a designated mode, and marking an artificial focus on the brain three-dimensional image; transforming the brain and the corresponding label in the brain three-dimensional image to the center position of the three-dimensional image by a three-dimensional space transformation method of simulated annealing, and enabling the middle axis surface of the brain to coincide with the middle vertical surfaces of the left horizontal axis and the right horizontal axis to carry out symmetry correction;
testing a mainstream deep learning algorithm, a network model, a brain image analysis and detection framework and a data set, and analyzing the advantages and disadvantages of various deep learning algorithms and convolution network models in the aspects of image segmentation, target detection and disease classification and grading;
aiming at the requirements and characteristics of brain image analysis, selecting a certain number of brain images aiming at typical brain diseases as samples, collecting brain images of various types and different angles for the same medical record, and filling multi-level classification labels;
based on a specific type of brain image data set and application requirements for detecting, classifying and grading focuses, respectively training different deep learning convolutional neural networks to obtain preliminary results of focus target segmentation detection, classification and grading for a specific brain image, and then processing a plurality of preliminary results by constructing a weighted Bayesian network to obtain a final analysis diagnosis result.
3. The brain medical image spectrum processing method based on deep learning as claimed in claim 1, wherein the image feature extraction module extracts the following:
(1) Constructing a brain spectrum image set; using a clustering algorithm to acquire a plurality of clustering centers from the brain spectrum image dataset to be classified as a low-level feature extractor; performing convolution operation on each brain spectrum image in the brain spectrum image dataset by using the plurality of low-level feature extractors, and generating a plurality of convolution brain spectrum images with the same quantity as the plurality of low-level feature extractors for each brain spectrum image;
(2) Thresholding is carried out on the plurality of convolution brain spectrum images respectively to obtain a plurality of sparse brain spectrum images; performing a normalization operation on the plurality of sparse brain spectral images, respectively, the normalization operation comprising: forming a vector by pixel values of the same position of each brain spectrum image in the plurality of sparse brain spectrum images, normalizing the vector, and then respectively returning each component of the vector to the corresponding position of each brain spectrum image to obtain a plurality of normalized sparse brain spectrum images;
(3) Performing low-level feature integration on the plurality of sparse brain spectrum images to obtain a plurality of integrated brain spectrum images; and performing middle-layer feature extraction operation on the plurality of integrated brain spectrum images to obtain middle-layer features.
4. A brain medical image spectral processing method based on deep learning according to claim 3, wherein said method further comprises, before said using a clustering algorithm to obtain a plurality of cluster centers from a brain spectral image dataset to be classified as low-level feature extractors:
and carrying out normalization and decoupling preprocessing operation on the brain spectrum images in the brain spectrum image data set to obtain the brain spectrum image data set to be classified.
5. The brain medical image spectrum processing method based on deep learning according to claim 3, wherein said thresholding said plurality of convolved brain spectral images to obtain a plurality of sparse brain spectral images, respectively, comprises:
and judging each pixel value of each convolution brain spectrum image in the convolution brain spectrum images, if the pixel value is larger than a preset threshold value, reserving the pixel value, otherwise, setting the pixel value to 0, and correspondingly generating a sparse brain spectrum image by the pixel value after thresholding operation of each convolution brain spectrum image to obtain a plurality of sparse brain spectrum images.
6. The brain medical image spectrum processing method based on deep learning according to claim 3, wherein performing low-level feature integration on the plurality of sparse brain spectral images to obtain a plurality of integrated brain spectral images comprises:
dividing each sparse brain spectrum image in the plurality of sparse brain spectrum images into a plurality of m multiplied by m areas, respectively forming a plurality of pixel values of the areas into m 2-dimensional vectors, forming a plurality of integrated brain spectrum images by the pixel values of the same positions of the vectors, wherein m is an integer greater than or equal to 2, and the number of the integrated brain spectrum images is m2 times that of the sparse brain spectrum images.
7. The brain medical image spectrum processing method based on deep learning as claimed in claim 1, wherein the image segmentation module segments the method as follows:
1) Carrying out enhancement treatment on the brain spectrum image; selecting a target segmentation algorithm corresponding to the brain spectrum image from candidate segmentation algorithms by extracting features of the brain spectrum image,
wherein the candidate segmentation algorithm comprises: a saliency segmentation algorithm and a semantic segmentation algorithm;
performing rough segmentation on the brain spectrum image by using the target segmentation algorithm to obtain a first segmentation result, wherein the rough segmentation is used for preliminarily determining a target object in the brain spectrum image;
and carrying out fine segmentation on the first segmentation result to obtain a target object in the brain spectrum image.
8. The brain medical image spectrum processing method based on deep learning according to claim 7, wherein said selecting a target segmentation algorithm corresponding to said brain spectral image from candidate segmentation algorithms by feature extraction of the brain spectral image comprises:
inputting the brain spectrum image into a backbone network in a segmentation model, and assigning the brain spectrum image to a corresponding node by the backbone network through feature extraction of the brain spectrum image, wherein the node comprises: a first node corresponding to each category of which the number of the sample brain spectrum images exceeds a preset value, and a second node corresponding to all categories of which the sample brain spectrum images are smaller than or equal to the preset value;
if the brain spectrum image is distributed to the first node, determining the target segmentation algorithm as the semantic segmentation algorithm;
and if the brain spectrum image is distributed to the second node, determining the target segmentation algorithm as the significance segmentation algorithm.
9. The brain medical image spectral processing method based on deep learning according to claim 7, wherein said segmentation model further includes a segmentation network including: the semantic segmentation network and the saliency segmentation network perform rough segmentation on the brain spectrum image by using the target segmentation algorithm to obtain a first segmentation result, wherein the method comprises the following steps:
the first feature layer output by the backbone network is input to a segmentation network corresponding to a target segmentation algorithm;
and performing rough segmentation on the brain spectrum image through the segmentation network.
10. The brain medical image spectrum processing method based on deep learning according to claim 7, wherein, in the case where the object segmentation algorithm is a saliency segmentation algorithm, performing coarse segmentation on the brain spectral image through the segmentation network includes:
stacking a plurality of second characteristic layers according to a preset layer jump rule to obtain a plurality of single-channel prediction masks, wherein the second characteristic layers are obtained by carrying out convolution and deconvolution on the first characteristic layers output by the backbone network;
acquiring first linear average values of the plurality of single-channel prediction masks;
determining the first linear mean as a first segmentation result corresponding to the brain spectral image;
extracting n single-channel prediction masks from the plurality of single-channel prediction masks, wherein n is a positive integer less than the number of the single-channel prediction masks;
acquiring second linear average values of the n single-channel prediction masks;
and determining the second linear mean as a first segmentation result corresponding to the brain spectrum image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310050576.8A CN116051509A (en) | 2023-02-01 | 2023-02-01 | Brain medical image spectrum processing method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310050576.8A CN116051509A (en) | 2023-02-01 | 2023-02-01 | Brain medical image spectrum processing method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116051509A true CN116051509A (en) | 2023-05-02 |
Family
ID=86123464
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310050576.8A Pending CN116051509A (en) | 2023-02-01 | 2023-02-01 | Brain medical image spectrum processing method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116051509A (en) |
-
2023
- 2023-02-01 CN CN202310050576.8A patent/CN116051509A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Cheng et al. | RetinaNet with difference channel attention and adaptively spatial feature fusion for steel surface defect detection | |
Zhang et al. | Concrete crack detection using context‐aware deep semantic segmentation network | |
Hong et al. | Multimodal GANs: Toward crossmodal hyperspectral–multispectral image segmentation | |
CN106683048B (en) | Image super-resolution method and device | |
US20200012838A1 (en) | Method and system for automatic chromosome classification | |
US10121245B2 (en) | Identification of inflammation in tissue images | |
He et al. | Deep regression neural network for industrial surface defect detection | |
Yogesh et al. | Computer vision based analysis and detection of defects in fruits causes due to nutrients deficiency | |
US9170211B2 (en) | Design-based inspection using repeating structures | |
AU2020272936B2 (en) | Methods and systems for crack detection using a fully convolutional network | |
Jiang et al. | Detection of molecular particles in live cells via machine learning | |
JP7427080B2 (en) | Weakly supervised multitask learning for cell detection and segmentation | |
Ngugi et al. | A new approach to learning and recognizing leaf diseases from individual lesions using convolutional neural networks | |
CN117015796A (en) | Method for processing tissue images and system for processing tissue images | |
US20220245801A1 (en) | Deep learning-based root cause analysis of process cycle images | |
CN115661069A (en) | Hyperspectral anomaly detection method and computer device | |
Gupta et al. | Deep learning for morphological identification of extended radio galaxies using weak labels | |
Nauman et al. | Identification of Anomalous Behavioral Patterns in Crowd Scenes. | |
CN116051509A (en) | Brain medical image spectrum processing method based on deep learning | |
CN112348011B (en) | Vehicle damage assessment method and device and storage medium | |
Kazmi et al. | A Deep Learning-Based Framework for Visual Inspection of Plastic Bottles | |
Zhang et al. | IDDM: An incremental dual-network detection model for in-situ inspection of large-scale complex product | |
Mishra et al. | Weakly supervised crack segmentation using crack attention networks on concrete structures | |
Thiyagarajan | Performance Comparison of Hybrid CNN-SVM and CNN-XGBoost models in Concrete Crack Detection | |
Umanandhini et al. | Batch Normalization based Convolutional Block YOLOv3 Real Time Object Detection of Moving Images with Backdrop Adjustment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |