CN111862259A

CN111862259A - Medical perfusion image processing method and medical imaging device

Info

Publication number: CN111862259A
Application number: CN202010729909.6A
Authority: CN
Inventors: 王梅云; 宗金光
Original assignee: Shanghai United Imaging Healthcare Co Ltd
Current assignee: Shanghai United Imaging Healthcare Co Ltd
Priority date: 2020-07-27
Filing date: 2020-07-27
Publication date: 2020-10-30
Anticipated expiration: 2040-07-27
Also published as: CN111862259B

Abstract

The embodiment of the invention discloses a medical perfusion image processing method and medical imaging equipment. The method comprises the following steps: acquiring a medical perfusion image set to be processed, and respectively performing spatial feature extraction and temporal feature extraction on the medical perfusion image set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image set; inputting the medical perfusion image set, the gray-scale spatial distribution characteristics and the contrast agent-time distribution characteristics into an artery input function model to obtain a model output result; identifying a target artery input point in the set of medical perfusion images according to the model output result. By the technical scheme, the artery input points can be identified from the medical perfusion images more quickly and accurately.

Description

Medical perfusion image processing method and medical imaging device

Technical Field

The embodiment of the invention relates to an image processing technology, in particular to a medical perfusion image processing method and medical imaging equipment.

Background

Perfusion is a dynamic enhanced scanning examination method commonly used in modern medicine and is mainly used for clinical diagnosis of acute cerebral ischemic diseases or tumors at present. For cerebral ischemic diseases, perfusion imaging may show lesions 30 minutes after the earliest ischemia. In addition, perfusion can be used for evaluating the degree of ischemia, and the hemodynamic change in unit tissues is quantitatively analyzed, so that a reference is provided for establishing a correct thrombolysis method.

In the process of processing medical perfusion images, it is necessary to accurately obtain the position of the artery (i.e. the artery input point) where the contrast agent enters from the vein at the earliest Time, and further obtain the contrast agent concentration-Time variation curve (TDC) of the artery input point, and the accuracy of the TDC of the artery input point directly affects the accuracy of perfusion parameters, such as Blood Flow (BF), Blood Volume (BV) and Mean Transit Time (MTT), and finally affects the diagnosis of the patient by the doctor.

Currently, the methods for obtaining artery input points from medical perfusion images mainly include: 1. and (4) manually selecting. The arterial input points are manually selected from the medical perfusion image by an experienced physician with a very high accuracy. 2. And (6) automatically selecting. A plurality of initial artery input points are determined from a medical perfusion image by utilizing an algorithm generated based on TDC characteristics of the artery input points, and then are corrected by a doctor.

However, the prior art arterial input point determination method has the following problems: the manual selection mode has high precision, but the determination process is complicated and time-consuming. In the automatic selection mode, only the characteristics of the artery input points TDC are considered, so that a plurality of initial artery input points output by the algorithm are distributed at various positions on the medical perfusion image, and the determination accuracy of the artery input points is poor.

Disclosure of Invention

Embodiments of the present invention provide a medical perfusion image processing method and a medical imaging device, so as to achieve faster and more accurate identification of an artery input point from a medical perfusion image.

In a first aspect, an embodiment of the present invention provides a medical perfusion image processing method, including:

acquiring a medical perfusion image set to be processed, and respectively performing spatial feature extraction and temporal feature extraction on the medical perfusion image set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image set;

inputting the medical perfusion image set, the gray-scale spatial distribution characteristics and the contrast agent-time distribution characteristics into an artery input function model to obtain a model output result;

identifying a target artery input point in the set of medical perfusion images according to the model output result.

In a second aspect, an embodiment of the present invention further provides a medical perfusion image processing method, where the method includes:

acquiring a medical perfusion image set to be processed;

acquiring contrast agent-time distribution characteristics of the set of medical perfusion images; performing spatial feature extraction on the medical perfusion image set to generate gray level spatial distribution features corresponding to the medical perfusion image set;

inputting the medical perfusion image set, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic into an artery input function model to obtain a model output result;

identifying a target artery input point in the medical perfusion image set according to the model output result;

a value of a perfusion parameter of a region of interest containing an input point of the target artery is acquired.

In a third aspect, an embodiment of the present invention further provides a medical imaging apparatus, including:

one or more processors;

a storage device for storing one or more programs,

when executed by the one or more processors, cause the one or more processors to implement a method of medical perfusion image processing as provided by any of the embodiments of the invention.

The method comprises the steps of obtaining a medical perfusion image set to be processed, and respectively carrying out spatial feature extraction and time feature extraction on the medical perfusion image set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image set; inputting the medical perfusion image set, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic into an artery input function model to obtain a model output result; a target artery input point in the set of medical perfusion images is identified from the model output result. When the artery input points in the medical perfusion image are identified, the spatial distribution characteristics of the arteries in the image and the time distribution characteristics of the artery input points TDC are considered, the problem that the artery input points output by the model are not concentrated in a small area to cause positioning errors of the artery input points is solved, the problem of excessive human intervention is also solved, and the accuracy and the efficiency of automatic identification of the artery input points are improved.

Drawings

FIG. 1 is a flow chart of a medical perfusion image processing method according to a first embodiment of the present invention;

FIG. 2 is a schematic diagram of a probability image of artery input points and candidate artery input points according to a first embodiment of the present invention;

FIG. 3 is a flowchart of a medical perfusion image processing method according to a second embodiment of the present invention;

FIG. 4 is a graph showing a contrast agent concentration-time curve at an arterial input point according to a second embodiment of the present invention;

FIG. 5 is a diagram illustrating a function curve of a residual function according to a second embodiment of the present invention;

fig. 6 is a schematic diagram of a model architecture of an artery input function model in a medical perfusion image processing method according to a third embodiment of the present invention;

FIG. 7 is a flowchart of a process for processing a set of medical perfusion images using an arterial input function model according to a third embodiment of the present invention;

FIG. 8 is a flowchart of a method for training an arterial input function model according to a fourth embodiment of the present invention;

fig. 9 is a schematic structural diagram of a medical perfusion image processing device in the fifth embodiment of the invention;

FIG. 10 is a schematic structural diagram of an arterial input function model training apparatus according to a sixth embodiment of the present invention;

fig. 11 is a schematic structural diagram of a medical imaging apparatus in a seventh embodiment of the invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.

Example one

The medical perfusion image processing method provided by the embodiment can be applied to the situation of automatically identifying the artery input point from the medical perfusion image. The method may be performed by a medical perfusion image processing apparatus, which may be implemented by software and/or hardware, and the apparatus may be integrated into an electronic device with image processing function, such as an image processing workstation of a medical device, such as a laptop, a desktop, a server or a magnetic resonance imaging device, a computed tomography imaging device, and the like. Referring to fig. 1, the method of the present embodiment specifically includes:

s110, a medical perfusion image set to be processed is obtained, and spatial feature extraction and time feature extraction are respectively carried out on the medical perfusion image set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image set.

The medical perfusion image set refers to a plurality of images obtained by scanning the same scanning part at different times after the perfusion of the contrast agent, and the images form time series images of the flow of the contrast agent along with blood. The gray scale spatial distribution characteristic refers to a distribution characteristic of pixel values in a spatial dimension, which is used for representing the appearance of anatomical structures (such as artery structures, vein structures, tissue structures and the like) of a scanning part on an image. The contrast agent-time (contrast agent injection time) distribution characteristic refers to a characteristic that a contrast agent concentration appears on an image over time, which may correspond to a TDC curve. Taking a CT image as an example, the contrast agent-time distribution characteristic may be a contrast agent density-time curve, the abscissa of the curve is time, the ordinate is an increased CT value after the contrast agent is injected, and the increased CT value reflects the concentration change of the contrast agent in the organ, thereby indirectly reflecting the change of the perfusion volume in the tissue organ. Of course, the medical perfusion image set may be a magnetic resonance image dataset, a computed tomography image dataset, a positron emission tomography image dataset, or the like.

In the process of automatically locating the artery input points, if only the contrast agent-time distribution characteristics, namely, the characteristics of the artery input point TDC, such as peak height, narrow peak, early peak (the time point when the peak of the curve starts early) and steep rising edge of the curve, are considered, the points in the vein or tissue with similar characteristics of the TDC, are determined as the artery input points, resulting in low algorithm accuracy. Based on this, the embodiment of the present invention considers the contrast agent-time distribution characteristics and the anatomical structure (such as artery, vein, tissue, etc.) of the scanned region at the same time, so that in the process of locating the artery input point, it is determined whether a pixel point is an artery structure first, and then it is determined whether the pixel point is an input point on the basis of the artery structure. In specific implementation, feature extraction is performed twice on the medical perfusion image set. The method comprises the steps of firstly, utilizing the pixel gray value of an image to extract features from the space dimension to obtain the gray spatial distribution features. And the other time, calculating the corresponding concentration value of the contrast agent by using the pixel gray value of the image, and then extracting a TDC curve from the time dimension to obtain the contrast agent-time distribution characteristic.

Illustratively, the spatial feature extraction is performed on the medical perfusion image set, and the generating of the gray scale spatial distribution feature corresponding to the medical perfusion image set comprises: dividing the set of medical perfusion images into a plurality of first medical perfusion image subsets in accordance with a first grid size; and determining the mean and the variance of each first medical perfusion image subset to generate the gray scale spatial distribution characteristics corresponding to the medical perfusion image set. In one example, the first medical perfusion image subsets may include a plurality of pixels, and a mean and a variance of pixel values corresponding to the plurality of pixels in each first medical perfusion image subset are respectively calculated, and the mean and the variance are used to characterize the gray scale spatial distribution. In specific implementation, in order to reduce the amount of data operations, first, each medical perfusion image in the medical perfusion image set is correspondingly divided into sub-images of a first grid size (a custom value, for example, 4 × 4 in a two-dimensional direction), and a series of sub-images with the same image scanning position but different image acquisition times are constructed as a first medical perfusion image subset, so that the same number (4 × 4) of first medical perfusion image subsets as the first grid size can be obtained. Then, the mean and variance of all pixel values in each first medical perfusion image subset are calculated, so as to obtain a 4 × 4 feature matrix, where the matrix elements are feature vectors formed by the mean and variance, and the feature matrix is the gray scale spatial distribution features corresponding to the medical perfusion image set.

Illustratively, the time feature extraction is performed on the medical perfusion image set, and the generation of the contrast agent-time distribution feature corresponding to the medical perfusion image set comprises: dividing the set of medical perfusion images into a plurality of second medical perfusion image subsets in accordance with a second grid size; and extracting a contrast agent concentration-time change curve of each medical perfusion image subset, determining curve parameters corresponding to the corresponding medical perfusion image subset according to each contrast agent concentration-time change curve, and generating contrast agent-time distribution characteristics corresponding to the medical perfusion image set. The curve parameter refers to a parameter of a contrast agent concentration-time variation curve, which is determined according to TDC characteristics of an artery input point, for example, the curve parameter includes a peak value, a peak width, a peak area, a peak starting point, a maximum rising slope, and the like of the curve. In one example, the contrast agent-time distribution characteristic is characterized by a contrast agent concentration-time variation curve parameter. Likewise, in order to reduce the amount of data operations, the set of medical perfusion images is divided in this embodiment into a same number (16 × 16 in two dimensions) of second medical perfusion image subsets as the second grid size. Then, calculating corresponding contrast agent concentration-time change curves of all pixel values in each second medical perfusion image subset, further extracting curve parameters, and obtaining a 16 × 16 feature matrix, wherein elements of the matrix are feature vectors formed by a peak value, a peak width, a peak area, a peak starting point and a maximum rising slope, and the feature matrix is corresponding contrast agent-time distribution features of the medical perfusion image set. In one example, the second mesh size is larger than the first mesh size because the spatially distributed features of the arterial structure can be extracted quickly from the larger scale image, further increasing the operation speed, while the TDC features of the input points need to be extracted from the smaller scale image, ensuring the accuracy of the TDC curve.

And S120, inputting the medical perfusion image set, the gray-scale spatial distribution characteristic and the contrast agent-time distribution characteristic into an artery input function model to obtain a model output result.

The arterial input function (arterial input function) model refers to a model capable of automatically locating an arterial input point, and is obtained by training a selected machine learning model. Of course, the artery input function model may also be a function model formed through multiple debugging, and the form of the artery input function model in the embodiment of the present application is not particularly limited.

In the embodiment of the invention, the selected machine learning model is trained in advance so as to be capable of carrying out feature extraction and classification analysis on the medical perfusion image set based on the gray scale space distribution feature and the contrast agent-time distribution feature at the same time. In specific implementation, the medical perfusion image set, the gray scale spatial distribution characteristics and the contrast agent-time distribution characteristics are input into the artery input function model, and a model output result can be obtained through model operation. The output result of the model can be one or more point positions representing the artery input points, and can also be a probability matrix with the size consistent with that of a medical perfusion image, each element in the matrix represents the probability that the corresponding image pixel point belongs to the artery input point, and the probability is higher, so that the probability is higher, and the probability is higher. This probability matrix may be referred to as an arterial input point probability image.

And S130, identifying a target artery input point in the medical perfusion image set according to the model output result.

If the model output result is a point location, then the point location is determined to be the target artery input point. If the output result of the model is a plurality of point positions or probability matrixes, the point positions or the probability matrixes need to be further subjected to point screening or point fusion and other processing, and the target artery input points can be obtained.

Exemplarily, S130 includes: and if the output result of the model is the probability image of the artery input point, determining a plurality of candidate artery input points according to the probability image of the artery input point and a preset probability threshold, and screening and/or fusing the candidate artery output points to generate a target artery input point in the medical perfusion image set. The preset probability threshold is a preset probability value, and is set according to the model precision requirement. Specifically, the output result of the model may be the probability image of the artery input point shown in fig. 2, and the whiter the color of the pixel point in the image is, the larger the probability value is. At this time, the points (i.e. candidate artery input points) with probability values greater than the preset probability threshold in the artery input point probability image need to be screened out. In one example, if there is one candidate artery input point, the candidate artery input point is the target artery input point. In another example, the candidate artery input points are multiple, such as multiple white points in fig. 2. Considering that the spatial features of the artery and the temporal features of the input points are considered simultaneously in the model, these candidate arterial input points are concentrated in the same small region at the artery, as the candidate arterial input points in candidate blocks S1, S2, and S3. Wherein each candidate artery input point in the candidate box S1 is a true artery point, and each candidate artery input point in the candidate boxes S2 and S3 is an interference point. Based on the method, one point can be secondarily screened out from all candidate artery input points to serve as a target artery input point; or, performing multi-point fusion on all candidate artery input points to obtain a fusion point as the target artery input point. The advantage of this arrangement is that a more accurate target artery input point can be obtained from the artery input point probability image, and the accuracy of artery input point identification is further improved.

According to the technical scheme of the embodiment, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic corresponding to the medical perfusion image set are generated by acquiring the medical perfusion image set to be processed and respectively performing spatial characteristic extraction and time characteristic extraction on the medical perfusion image set; inputting the medical perfusion image set, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic into an artery input function model to obtain a model output result; a target artery input point in the set of medical perfusion images is identified from the model output result. When the artery input points in the medical perfusion image are identified, the spatial distribution characteristics of the arteries in the image and the time distribution characteristics of the artery input points TDC are considered, the problem that the artery input points output by the model are not concentrated in a small area to cause positioning errors of the artery input points is solved, the problem of excessive human intervention is also solved, and the accuracy and the efficiency of automatic identification of the artery input points are improved.

Example two

The present embodiment provides a medical perfusion image processing method, which may be performed by a magnetic resonance system, a computed tomography system, or a medical post-processing workstation, and referring to fig. 3, the method may include the following specific steps:

s210, acquiring a medical perfusion image set to be processed.

The set of medical perfusion images includes images of one or more slices. The medical image to be processed may be obtained, for example, by perfusion imaging of the liver blood flow, for example, by a slice-dynamic scan of a selected slice after a bolus injection of contrast agent. Of course, the region of interest corresponding to the medical image to be processed may also be an organ or tissue such as a brain, a heart, a lung, a kidney, and the like.

The medical perfusion image set may be obtained by the following process: a bolus of contrast agent is applied to the vascular system of the patient and the region of interest is repeatedly imaged at a number of different points in time over a period encompassing the time of transmission of the contrast agent through the tissue of the region of interest, for example 3-30 repeated scans with a few seconds difference between successive scans.

S220, obtaining contrast agent-time distribution characteristics of the medical perfusion image set, extracting spatial characteristics of the medical perfusion image set, and generating gray scale spatial distribution characteristics corresponding to the medical perfusion image set.

It is understood that the step of taking the contrast agent-temporal distribution characteristic of the medical perfusion image set may be performed synchronously with or sequentially with the gray-scale spatial distribution characteristic corresponding to the medical perfusion image set, and the timing of performing the two steps is not particularly limited.

And S230, inputting the medical perfusion image set, the gray-scale spatial distribution characteristic and the contrast agent-time distribution characteristic into an artery input function model to obtain a model output result.

And S240, identifying a target artery input point in the medical perfusion image set according to the model output result.

And S250, acquiring a perfusion parameter value of the region of interest containing the target artery input point.

Wherein the perfusion parameter value may be one or more of TDC, perfusion parameter blood flow, blood volume, or mean transit time.

In an example, the perfusion parameter value is a TDC of the region of interest. First, a contrast agent concentration-time variation curve TDC at an artery input point (simply referred to as an arterial contrast agent curve) is generated from a medical perfusion image set as shown in fig. 4, in which the horizontal axis represents a time axis in seconds (S), the vertical axis represents a contrast agent concentration in arbitrary relative values, and the vertical axis represents CT relative values in the example of a CT perfusion image. Then, according to equation (1), TDC of the region of interest is determined from the arterial contrast agent curve TDC:

wherein, the TDC_voiA TDC representing a region of interest containing a target artery input point; TDC_arteryRepresents the arterial contrast agent curve TDC; r (t) represents a residual function, the curve of which is shown in FIG. 5, the horizontal axis represents the time axis and has the unit of S;

representing a convolution function.

In an example, the perfusion parameter value may be a perfusion parameter blood flow, blood volume, or average transit time, which may be obtained by equations (2) - (4) as follows, respectively:

BV_voi＝MTT_voi*BF_voi(4)

wherein, BF_voiIndicating bagPerfusion parameter blood flow of a region of interest containing a target artery input point; rho_voiRepresenting blood flow in a region of interest; max represents the maximum operation; MTT_voiRepresenting the mean transit time of the contrast agent in the region of interest; BV (BV)_voiRepresenting the blood volume of the region of interest.

According to the technical scheme of the embodiment, the perfusion parameter value of the region of interest including the target artery input point is obtained, so that the perfusion parameter value of the region of interest is obtained by using the more accurate target artery input point, and the accuracy of the perfusion parameter value is improved.

EXAMPLE III

In this embodiment, based on the first embodiment, the "arterial input function model" is further optimized. The same or corresponding terms as those in the above embodiments are not explained in detail herein.

Referring to fig. 6, the model architecture of the artery input function model in the medical perfusion image processing method provided by this embodiment includes: an identifying network unit 610 and a synthesizing unit 620, the identifying network unit 610 includes a space identifying subnet 611 and a time identifying subnet 612; the spatial recognition subnetwork 611 is configured to recognize the medical perfusion image set in the spatial dimension, the temporal recognition subnetwork 612 is configured to recognize the medical perfusion image in the temporal dimension, and the synthesis unit 620 is configured to obtain a model output result.

The spatial identification subnetwork 611 is used for performing artery identification on the medical perfusion image set from the gray scale spatial distribution dimension according to the gray scale spatial distribution characteristics to obtain an artery probability image corresponding to the medical perfusion image set;

a time identification sub-network 612, configured to perform input point identification on the medical perfusion image set from a time distribution dimension of the contrast agent according to the contrast agent-time distribution characteristic, and obtain an input point probability image corresponding to the medical perfusion image set;

and a synthesizing unit 620, configured to fuse the artery probability image and the input point probability image, and generate an artery input point probability image corresponding to the medical perfusion image set.

The spatial recognition subnetwork 611 and the temporal recognition subnetwork 612 in the recognition network unit 610 are machine learning models with higher attention to image spatial features and machine learning models with higher attention to image temporal features, respectively, for example, the spatial recognition subnetwork 611 is a radial basis function neural network model RBF, and the temporal recognition subnetwork is a dual-mode linear neural network model BLE. The space recognition subnet 611 and the time recognition subnet 612 are independent in the image classification recognition process, and the execution order of the two is not limited.

The operation process of the artery input function model is as follows: the arterial input function model receives input data for a set of medical perfusion images, a gray scale spatial distribution feature, and a contrast agent-time distribution feature. The spatial recognition subnetwork 611 obtains the gray scale spatial distribution characteristics and the medical perfusion image set, and performs artery structure recognition on the medical perfusion image set according to the gray scale spatial distribution characteristics to obtain an artery probability image. The probability value of the pixel points identified as the artery structure in the artery probability image is larger, and the probability value of the pixel points of the non-artery structure is smaller. The time identifier sub-network 612 obtains the contrast agent-time distribution characteristics and the medical perfusion image set, and identifies the input points of the medical perfusion image set according to the contrast agent-time distribution characteristics to obtain the input point probability image. The probability value of the pixel points identified as input points in the input point probability image is larger, and the probability value of the pixel points of non-input points is smaller. The image size of both the artery probability image and the input point probability image is consistent with the image size of the medical perfusion image.

The output results of the spatial recognition subnetwork 611 and the temporal recognition subnetwork 612 are input to the synthesis unit 620. The synthesizing unit 620 fuses the artery probability image and the input point probability image pixel by pixel, and obtains the artery input point probability image of each pixel point through the fusion of the two probabilities. In the probability fusion process, the artery probability and the input point probability are not in a complete parallel relation, and the priority of the artery probability is higher than that of the input point probability. It can be understood that: it is only meaningful to consider whether a pixel belongs to an input point, if it belongs to an arterial structure. If the artery probability value of a pixel point is small, the probability of the input point is large and meaningless, and the corresponding fused artery input point probability is small.

Illustratively, the arterial input function model further comprises a classification network element 630;

the classification network unit 630 is configured to classify the medical perfusion image set according to the gray scale spatial distribution feature and the contrast agent-time distribution feature, to obtain each spatio-temporal classification image subset, and to input each spatio-temporal classification image subset to the identification network unit.

Correspondingly, the space identification sub-network 611 is configured to perform artery identification on each spatio-temporal classification image subset from the gray scale space distribution dimension, and obtain an artery probability sub-image corresponding to each spatio-temporal classification image subset;

a time identification sub-network 612, configured to perform input point identification on each spatio-temporal classification image subset from a time distribution dimension of the contrast agent, to obtain an input point probability sub-image corresponding to each spatio-temporal classification image subset;

and a synthesizing unit 620, configured to fuse the artery probability sub-images and the input point probability sub-images corresponding to the same spatio-temporal classification image subset, generate artery input point probability sub-images corresponding to the spatio-temporal classification image subsets, and generate artery input point probability images from the artery input point probability sub-images.

The classification network element 630 is arranged before the identification network element 610. The classification network unit 630 receives all the model input data, and classifies the medical perfusion image set according to the gray-scale spatial distribution characteristics and the contrast agent-time distribution characteristics, to obtain pixel points belonging to the same spatial category and the same temporal category, and to form each spatiotemporal classification image subset. For example, the medical perfusion image set is classified according to the gray scale spatial distribution characteristics to obtain 10 spatial categories, and each spatial category includes a part of pixel points in the medical perfusion image set. Classifying the medical perfusion image set according to the contrast agent-time distribution characteristics to obtain 100 time categories, wherein each time category comprises part of pixel points in the medical perfusion image set. And dividing each pixel point which belongs to a certain space category and a certain time category into a space-time classification image subset corresponding to the space category and the time category. The advantage of such an arrangement is that the whole medical perfusion image set is divided into more refined small classifications, which not only reduces the data computation amount of the recognition network unit 610 and improves the model computation speed, but also enables the recognition network unit 610 to recognize the artery and the input point more finely, further improving the accuracy of model recognition.

On the basis of adding the classification network unit 630 to the model, the processing objects of the space recognition sub-network 611 and the time recognition sub-network 612 are changed from the whole medical perfusion image set into each space-time classification image subset, and the output results are correspondingly changed into each artery probability sub-image and each input point probability sub-image. The synthesis unit 620 fuses the artery probability sub-images and the input point probability sub-images corresponding to each spatio-temporal classification image subset to obtain the artery input point probability sub-images, and then splices the artery input point probability sub-images to form a complete artery input point probability image.

Illustratively, the synthesis unit 620 is specifically configured to: and multiplying the artery probability sub-images corresponding to the same space-time classification image subset with the corresponding pixel values of the input point probability sub-images to generate the artery input point probability sub-images corresponding to the corresponding space-time classification image subset. Based on the design that the artery probability is considered preferentially in the probability fusion process, when the artery probability sub-images and the input point probability sub-images corresponding to the same space-time classification image subset are fused, the synthesis unit 620 multiplies the probability values of corresponding pixel points in the two probability sub-images to obtain the artery input point probability of each pixel point. The advantage of this setting is that the artery probability and the input point probability can be fused more conveniently, and the generation speed of the artery input point probability image is further improved, so that the identification efficiency of the target artery input point is further improved.

Referring to fig. 7, based on the model structure of the artery input function model, the processing procedure for the medical perfusion image set is as follows: first, a medical perfusion image set is input to a feature generator, and a contrast agent-time distribution feature and a spatial distribution feature of the medical perfusion image set are extracted. Then, the medical perfusion image set and the output result of the feature generator are all input into the classification network unit, and a plurality of space-time classification image subsets are obtained. And then, inputting each spatio-temporal classification image subset into the corresponding RBF and BLE respectively to obtain a corresponding artery probability sub-image and an input point probability sub-image. And finally, inputting the input result of each group of RBF and BLE into a synthesis unit, and obtaining a model output result.

According to the technical scheme, an identification network unit comprising a space identification sub-network and a time identification sub-network is designed in an artery input function model, and the space identification sub-network is used for carrying out artery identification on a medical perfusion image set from a gray scale space distribution dimension according to a gray scale space distribution characteristic to obtain an artery probability image corresponding to the medical perfusion image set; the time identification sub-network is used for identifying input points of the medical perfusion image set from the time distribution dimension of the contrast agent according to the contrast agent-time distribution characteristics to obtain input point probability images corresponding to the medical perfusion image set. The probability that each pixel belongs to the artery is identified from the angle of the anatomical structure of the medical perfusion image by using the space identification sub-network which pays more attention to the image space distribution characteristics in the artery input function model, and the probability that each pixel belongs to the input point is identified from the time distribution dimension of the artery input point TDC by using the time identification sub-network which pays more attention to the time distribution characteristics, so that the identification accuracy of the artery and the input point in the medical perfusion image set is further improved, and the identification accuracy of the artery input point is further improved. The synthetic unit for fusing the artery probability image and the input point probability image and generating the artery input point probability image corresponding to the medical perfusion image set is designed in the artery input function model, so that the artery probability and the input point probability of each pixel point are fused in a more reasonable mode, the artery input point probability with higher accuracy of the corresponding pixel point is obtained, and the identification probability of the artery input point in the medical perfusion image set is further improved.

Example four

The method for training the artery input function model provided by the embodiment can be applied to training the artery input function model constructed based on the machine learning model so as to realize the processing of the medical perfusion image. The method may be performed by an arterial input function model training apparatus, which may be implemented by software and/or hardware, and may be integrated into an electronic device with an image processing function, such as a laptop, a desktop, a server, or the like. Referring to fig. 8, the method for training an artery input function model of the present embodiment specifically includes:

s310, a medical perfusion image sample set is obtained, and spatial feature extraction and time feature extraction are respectively carried out on the medical perfusion image sample set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image sample set.

Wherein the medical perfusion image sample set refers to a medical perfusion image set collected for model training.

As described in S110, the gray-scale spatial distribution characteristic and the contrast agent-time distribution characteristic corresponding to the medical perfusion image sample set are obtained.

Illustratively, the spatial feature extraction is performed on the medical perfusion image sample set, and the generating of the gray scale spatial distribution feature corresponding to the medical perfusion image sample set includes: dividing a set of medical perfusion image samples into a plurality of first medical perfusion image subsets in accordance with a first grid size; and determining the mean and the variance of each first medical perfusion image subset, and generating the gray scale spatial distribution characteristics corresponding to the medical perfusion image sample set.

Illustratively, the time feature extraction of the medical perfusion image sample set, and the generation of the contrast agent-time distribution feature corresponding to the medical perfusion image sample set comprises: dividing the set of medical perfusion image samples into a plurality of second medical perfusion image subsets in accordance with a second grid size; and extracting a contrast agent concentration-time change curve of each medical perfusion image subset, determining curve parameters corresponding to the corresponding medical perfusion image subset according to each contrast agent concentration-time change curve, and generating contrast agent-time distribution characteristics corresponding to the medical perfusion image sample set.

And S320, inputting the medical perfusion image sample set, the gray scale space distribution characteristic and the contrast agent-time distribution characteristic into a machine learning model to obtain a model output result.

The artery input function model in the present embodiment belongs to a machine learning model, and a model architecture of the machine learning model is described below. The model training process is to continuously iteratively solve the optimal model parameters. In specific implementation, initial model parameters, such as default values of the model parameters, are set for the machine learning model to be trained. Then, the medical perfusion image sample set, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic are input into a machine learning model to be trained, and a model output result, such as an artery input point probability image, is obtained through the operation of the machine learning model.

S330, if the model training does not reach the convergence condition, performing iterative training on the machine learning model according to the gold standard corresponding to the medical perfusion image sample set and the model output result until the model training reaches the convergence condition, and generating the artery input function model.

The convergence condition is a preset indicator of the end of model training, and may be, for example, a preset acceptable error of the model to identify the artery input point, or a preset number of iterations. The gold standard refers to an artery input point probability image obtained by an experienced physician manually identifying a medical perfusion image sample set. Since the identification network unit of the artery input function model needs to perform artery identification and input point identification respectively, the golden standard may also include an artery probability image and an input point probability image.

Each time the operation of S320 is performed, the model error can be calculated by using the artery input point probability image output by the model and the corresponding gold standard. And then performing error back-propagation of the model error to update the model parameters. And then judging whether the model training reaches the convergence condition. If the convergence condition is not reached, returning to execute S320, and performing loop iterative training on the machine learning model until the convergence condition is met. If the convergence condition is reached, the arterial input function model is obtained using the last updated model parameters and the machine learning model.

Exemplarily, the artery input function model comprises a recognition network unit and a synthesis unit, wherein the recognition network unit comprises a space recognition sub-network and a time recognition sub-network;

the space identification sub-network is used for carrying out artery identification on the medical perfusion image sample set from the gray scale space distribution dimension according to the gray scale space distribution characteristics to obtain an artery probability image corresponding to the medical perfusion image sample set;

the time identification sub-network is used for identifying input points of the medical perfusion image sample set from the time distribution dimension of the contrast agent according to the contrast agent-time distribution characteristics to obtain input point probability images corresponding to the medical perfusion image sample set;

the synthesis unit is used for fusing the artery probability image and the input point probability image to generate an artery input point probability image corresponding to the medical perfusion image sample set.

Further, the artery input function model further comprises a classification network unit;

the classification network unit is used for classifying the medical perfusion image sample set according to the gray scale space distribution characteristic and the contrast agent-time distribution characteristic to obtain each space-time classification image subset;

correspondingly, the space identification sub-network is used for carrying out artery identification on each space-time classification image subset from the gray scale space distribution dimension to obtain an artery probability sub-image corresponding to each space-time classification image subset;

the time identification sub-network is used for identifying input points of each spatio-temporal classification image subset from the time distribution dimension of the contrast agent to obtain input point probability sub-images corresponding to each spatio-temporal classification image subset;

the synthesis unit is used for fusing the artery probability sub-images and the input point probability sub-images corresponding to the same space-time classification image subset, generating artery input point probability sub-images corresponding to the corresponding space-time classification image subset, and generating artery input point probability images from the artery input point probability sub-images.

According to the technical scheme of the embodiment, a medical perfusion image sample set is obtained, and spatial feature extraction and time feature extraction are respectively carried out on the medical perfusion image sample set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image sample set; inputting the medical perfusion image sample set, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic into a machine learning model to obtain a model output result; and if the model training does not reach the convergence condition, performing iterative training on the machine learning model according to the gold standard corresponding to the medical perfusion image sample set and the model output result until the model training reaches the convergence condition, and generating an artery input function model. The training of the artery input function model constructed based on the machine learning model is realized, when the artery input function model identifies the artery input point in the medical perfusion image, the spatial distribution characteristic of the artery in the image and the time distribution characteristic of the artery input point TDC are considered at the same time, the problem that the artery input point output by the model is not concentrated in a certain small area to cause positioning error of the artery input point is avoided, the problem of excessive human intervention is also avoided, and the accuracy and the efficiency of automatic identification of the artery input point are improved.

EXAMPLE five

The present embodiment provides a medical perfusion image processing apparatus, referring to fig. 9, the apparatus specifically includes:

the feature extraction module 910 is configured to obtain a medical perfusion image set to be processed, and perform spatial feature extraction and temporal feature extraction on the medical perfusion image set respectively to generate a gray scale spatial distribution feature and a contrast agent-temporal distribution feature corresponding to the medical perfusion image set;

a model output result obtaining module 920, configured to input the medical perfusion image set, the gray-scale spatial distribution characteristic, and the contrast agent-time distribution characteristic into the artery input function model, so as to obtain a model output result;

a target artery input point identification module 930 configured to identify a target artery input point in the set of medical perfusion images according to the model output result.

Optionally, the artery input function model comprises an identification network unit and a synthesis unit, wherein the identification network unit comprises a space identification sub-network and a time identification sub-network;

the space identification sub-network is used for carrying out artery identification on the medical perfusion image set from the gray scale space distribution dimension according to the gray scale space distribution characteristics to obtain an artery probability image corresponding to the medical perfusion image set;

the time identification sub-network is used for identifying input points of the medical perfusion image set from the time distribution dimension of the contrast agent according to the contrast agent-time distribution characteristics to obtain input point probability images corresponding to the medical perfusion image set;

the synthesis unit is used for fusing the artery probability image and the input point probability image to generate an artery input point probability image corresponding to the medical perfusion image set.

the classification network unit is used for classifying the medical perfusion image set according to the gray scale space distribution characteristic and the contrast agent-time distribution characteristic to obtain each space-time classification image subset;

Optionally, the synthesis unit is specifically configured to:

and multiplying the artery probability sub-images corresponding to the same space-time classification image subset with the corresponding pixel values of the input point probability sub-images to generate the artery input point probability sub-images corresponding to the corresponding space-time classification image subset.

Optionally, the feature extraction module 910 is specifically configured to:

dividing the set of medical perfusion images into a plurality of first medical perfusion image subsets in accordance with a first grid size;

and determining the mean and the variance of each first medical perfusion image subset to generate the gray scale spatial distribution characteristics corresponding to the medical perfusion image set.

Optionally, the feature extraction module 910 is further specifically configured to:

dividing the set of medical perfusion images into a plurality of second medical perfusion image subsets in accordance with a second grid size;

and extracting a contrast agent concentration-time change curve of each medical perfusion image subset, determining curve parameters corresponding to the corresponding medical perfusion image subset according to each contrast agent concentration-time change curve, and generating contrast agent-time distribution characteristics corresponding to the medical perfusion image set.

Optionally, the target artery input point identifying module 930 is specifically configured to:

and if the output result of the model is the probability image of the artery input point, determining a plurality of candidate artery input points according to the probability image of the artery input point and a preset probability threshold, and screening and/or fusing the candidate artery output points to generate a target artery input point in the medical perfusion image set.

By the artery input point identification device in the fifth embodiment of the invention, the space distribution characteristics of the artery in the image and the time distribution characteristics of the artery input point TDC are considered when the artery input point in the medical perfusion image is identified, so that the problem of positioning error of the artery input point due to the fact that the position of the artery input point output by the model is not concentrated in a small area is avoided, the problem of excessive human intervention is also avoided, and the accuracy and the efficiency of automatic identification of the artery input point are improved.

In another embodiment, the medical perfusion image processing apparatus further comprises a perfusion parameter value acquisition module for:

after identifying a target artery input point in the set of medical perfusion images from the model output result, a perfusion parameter value of a region of interest containing the target artery input point is obtained.

The artery input point identification device provided by the embodiment of the invention can execute the artery input point identification method provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method.

EXAMPLE six

The present embodiment provides an artery input function model training apparatus, referring to fig. 10, the apparatus specifically includes:

the feature extraction module 1010 is configured to obtain a medical perfusion image sample set, and perform spatial feature extraction and temporal feature extraction on the medical perfusion image sample set respectively to generate a gray scale spatial distribution feature and a contrast agent-temporal distribution feature corresponding to the medical perfusion image sample set;

a model output result obtaining module 1020, configured to input the medical perfusion image sample set, the gray-scale spatial distribution characteristic, and the contrast agent-time distribution characteristic into a machine learning model, so as to obtain a model output result;

and the iterative training module 1030 is configured to, if it is determined that the model training does not reach the convergence condition, perform iterative training on the machine learning model according to the gold standard corresponding to the medical perfusion image sample set and the model output result until it is determined that the model training reaches the convergence condition, and generate an arterial input function model.

Optionally, the feature extraction module 1010 is specifically configured to:

dividing a set of medical perfusion image samples into a plurality of first medical perfusion image subsets in accordance with a first grid size;

and determining the mean and the variance of each first medical perfusion image subset, and generating the gray scale spatial distribution characteristics corresponding to the medical perfusion image sample set.

Optionally, the feature extraction module 1010 is further specifically configured to:

dividing the set of medical perfusion image samples into a plurality of second medical perfusion image subsets in accordance with a second grid size;

and extracting a contrast agent concentration-time change curve of each medical perfusion image subset, determining curve parameters corresponding to the corresponding medical perfusion image subset according to each contrast agent concentration-time change curve, and generating contrast agent-time distribution characteristics corresponding to the medical perfusion image sample set.

By the artery input function model training device, training of an artery input function model constructed based on a machine learning model is achieved, when an artery input point in a medical perfusion image is identified by the artery input function model, spatial distribution characteristics of an artery in the image and time distribution characteristics of a TDC (time to live) of the artery input point are considered at the same time, the problem that the artery input point output by the model is not concentrated in a certain small area to cause positioning errors of the artery input point is avoided, the problem of excessive human intervention is also avoided, and the accuracy and the efficiency of automatic identification of the artery input point are improved.

The training device for the artery input function model provided by the embodiment of the invention can execute the training method for the artery input function model provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method.

It should be noted that, in the embodiment of the apparatus, the included units and modules are merely divided according to functional logic, but are not limited to the above division as long as the corresponding functions can be implemented; in addition, specific names of the functional units are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present invention.

EXAMPLE seven

Referring to fig. 11, the present embodiment provides a medical imaging apparatus 1100, where the medical imaging apparatus 1100 may be a medical imaging apparatus such as a magnetic resonance imaging system, a computed tomography imaging system, or a medical image post-processing workstation, and includes: one or more processors 1120; the storage 1110 is used to store one or more programs, and when the one or more programs are executed by the one or more processors 1120, the one or more processors 1120 are enabled to implement the medical perfusion image processing method provided by the embodiment of the invention, including:

acquiring a medical perfusion image set to be processed, and respectively performing spatial feature extraction and time feature extraction on the medical perfusion image set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image set;

a target artery input point in the set of medical perfusion images is identified from the model output result.

Of course, it will be understood by those skilled in the art that the processor 1120 may also implement the technical solution of the medical perfusion image processing method provided by any embodiment of the present invention.

The medical imaging device 1100 shown in fig. 11 is only an example and should not impose any limitation on the functionality and scope of use of embodiments of the present invention. As shown in fig. 11, the medical imaging apparatus 1100 includes a processor 1120, a storage device 1110, an input device 1130, and an output device 1140; the number of the processors 1120 in the medical imaging apparatus may be one or more, and one processor 1120 is taken as an example in fig. 11; the processor 1120, the storage 1110, the input 1130, and the output 1140 of the medical imaging apparatus may be connected by a bus or other means, for example, the bus 1150 in fig. 11.

The storage device 1110 is a computer-readable storage medium for storing software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to the medical perfusion image processing method in the embodiment of the present invention.

The storage 1110 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal, and the like. Further, the storage 1110 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some examples, the storage 1110 may further include memory located remotely from the processor 1120, which may be connected to the medical imaging device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The input device 1130 may be used to receive input numeric or character information and generate key signal inputs related to user settings and function controls of the medical imaging apparatus. The output device 1140 may include a display device such as a display screen.

Embodiments of the present invention also provide another medical imaging apparatus, including: one or more processors; a storage device, configured to store one or more programs, which when executed by one or more processors, cause the one or more processors to implement the method for training an arterial input function model according to an embodiment of the present invention, including:

acquiring a medical perfusion image sample set, and respectively performing spatial feature extraction and time feature extraction on the medical perfusion image sample set to generate a gray scale spatial distribution feature and a contrast agent-time distribution feature corresponding to the medical perfusion image sample set;

inputting the medical perfusion image sample set, the gray scale spatial distribution characteristic and the contrast agent-time distribution characteristic into a machine learning model to obtain a model output result;

and if the model training does not reach the convergence condition, performing iterative training on the machine learning model according to the gold standard corresponding to the medical perfusion image sample set and the model output result until the model training reaches the convergence condition, and generating an artery input function model.

Of course, those skilled in the art will understand that the processor may also implement the technical solution of the training method for the artery input function model provided in any embodiment of the present invention. The hardware structure and function of the medical imaging device can be explained with reference to the content of the sixth embodiment.

Example eight

The present embodiments provide a storage medium containing computer executable instructions which, when executed by a computer processor, are operable to perform a method of medical perfusion image processing, the method comprising:

Of course, the embodiment of the present invention provides a storage medium containing computer-executable instructions, and the computer-executable instructions are not limited to the above method operations, and can also perform related operations in the medical perfusion image processing method provided by any embodiment of the present invention.

From the above description of the embodiments, it is obvious for those skilled in the art that the present invention can be implemented by software and necessary general hardware, and certainly, can also be implemented by hardware, but the former is a better embodiment in many cases. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, and the computer software product may be stored in a computer-readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, and includes several instructions to enable an electronic device (which may be a personal computer, a server, or a network device) to execute the artery input point identification method provided in the embodiments of the present invention.

Embodiments of the present invention also provide another computer-readable storage medium, where computer-executable instructions, when executed by a computer processor, perform a method for arterial input function model training, the method comprising:

Of course, the storage medium containing the computer-executable instructions provided by the embodiments of the present invention is not limited to the above method operations, and may also perform related operations in the method for training the arterial input function model provided by any embodiments of the present invention. The description of the storage medium is explained with reference to the seventh embodiment.

It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A medical perfusion image processing method, comprising:

2. The method of claim 1, wherein the arterial input function model comprises an identification network element and a synthesis element, the identification network element comprising a spatial identification subnetwork and a temporal identification subnetwork;

the spatial recognition subnetwork is configured to recognize the set of medical perfusion images in a spatial dimension, the temporal recognition subnetwork is configured to recognize the medical perfusion images in a temporal dimension, and the synthesis unit is configured to obtain the model output result.

3. The method according to claim 2, wherein the artery input function model further comprises a classification network unit for classifying the set of medical perfusion images according to the gray-scale spatial distribution features and the contrast agent-temporal distribution features to obtain spatio-temporal classification image subsets for inputting the spatio-temporal classification image subsets into the recognition network unit.

4. The method according to claim 3, wherein the spatial recognition sub-network is configured to perform artery recognition on each of the subsets of spatiotemporal classification images from a gray scale spatial distribution dimension to obtain an artery probability sub-image corresponding to each of the subsets of spatiotemporal classification images;

the time identification sub-network is used for carrying out input point identification on each spatio-temporal classification image subset from the time distribution dimension of the contrast agent to obtain an input point probability sub-image corresponding to each spatio-temporal classification image subset;

the synthesis unit is used for fusing the artery probability sub-images and the input point probability sub-images corresponding to the same space-time classification image subset, generating the artery input point probability sub-images corresponding to the corresponding space-time classification image subset, and generating the artery input point probability images from the artery input point probability sub-images.

5. The method according to claim 4, characterized in that the synthesis unit is specifically configured to:

6. The method of claim 1, wherein performing spatial feature extraction on the medical perfusion image set, and generating a gray-scale spatial distribution feature corresponding to the medical perfusion image set comprises:

determining the mean and variance of each first medical perfusion image subset, and generating the corresponding gray scale spatial distribution characteristics of the medical perfusion image set.

7. The method of claim 1, wherein performing temporal feature extraction on the set of medical perfusion images, generating contrast agent-temporal distribution features corresponding to the set of medical perfusion images comprises:

extracting a contrast agent concentration-time change curve of each medical perfusion image subset, determining a curve parameter corresponding to the corresponding medical perfusion image subset according to each contrast agent concentration-time change curve, and generating a contrast agent-time distribution characteristic corresponding to the medical perfusion image set.

8. The method of claim 1, wherein identifying a target artery input point in the set of medical perfusion images from the model output result comprises:

and if the model output result is an artery input point probability image, determining a plurality of candidate artery input points according to the artery input point probability image and a preset probability threshold, and screening and/or fusing the candidate artery output points to generate a target artery input point in the medical perfusion image set.

9. A medical perfusion image processing method, comprising:

acquiring a medical perfusion image set to be processed;

10. A medical imaging device, characterized in that the medical imaging device comprises:

one or more processors;

a storage device for storing one or more programs,

when executed by the one or more processors, cause the one or more processors to implement the medical perfusion image processing method of any one of claims 1-9.