CN112085711A - Method for automatically tracking muscle pinnate angle by combining convolutional neural network and Kalman filtering - Google Patents
Method for automatically tracking muscle pinnate angle by combining convolutional neural network and Kalman filtering Download PDFInfo
- Publication number
- CN112085711A CN112085711A CN202010865743.0A CN202010865743A CN112085711A CN 112085711 A CN112085711 A CN 112085711A CN 202010865743 A CN202010865743 A CN 202010865743A CN 112085711 A CN112085711 A CN 112085711A
- Authority
- CN
- China
- Prior art keywords
- muscle
- angle
- neural network
- image
- pinnate
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 210000003205 muscle Anatomy 0.000 title claims abstract description 46
- 238000001914 filtration Methods 0.000 title claims abstract description 25
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 14
- 210000001087 myotubule Anatomy 0.000 claims abstract description 63
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 26
- 238000013528 artificial neural network Methods 0.000 claims abstract description 17
- 210000003746 feather Anatomy 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 229910052704 radon Inorganic materials 0.000 claims description 17
- SYUHGPGVQRZVTB-UHFFFAOYSA-N radon atom Chemical compound [Rn] SYUHGPGVQRZVTB-UHFFFAOYSA-N 0.000 claims description 17
- 238000001514 detection method Methods 0.000 claims description 12
- 238000003708 edge detection Methods 0.000 claims description 10
- 238000005259 measurement Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 10
- 239000000835 fiber Substances 0.000 claims description 5
- 239000012528 membrane Substances 0.000 claims description 5
- 210000000518 sarcolemma Anatomy 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 abstract description 5
- 238000002604 ultrasonography Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 9
- 238000013135 deep learning Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000003672 processing method Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000026676 system process Effects 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000003387 muscular Effects 0.000 description 1
- 201000000585 muscular atrophy Diseases 0.000 description 1
- 230000004092 musculoskeletal function Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 210000002027 skeletal muscle Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for automatically tracking a muscle pinnate angle by combining a convolutional neural network and Kalman filtering, which specifically comprises the following steps: (1) preprocessing an ultrasonic image; (2) detecting a myolemma; (3) obtaining a muscle fiber direction observation value; (4) correcting the direction of muscle fiber; (5) and calculating the feather angle. The invention measures the current direction of the muscle fiber by utilizing a deep convolution neural network, and realizes the tracking of the pinnate angle by combining with a Kalman filter. The method improves the robustness of the pinnate angle calculation algorithm, expands the application field of the automatic labeling pinnate angle algorithm, and provides a method for automatically tracking the pinnate angle for an ultrasonic image sequence with poor quality.
Description
Technical Field
The invention relates to the technical field of muscle pinnate angle detection, in particular to a method for automatically tracking a muscle pinnate angle by combining a convolutional neural network and Kalman filtering.
Background
The pinnate angle of the muscle bundle is an important parameter related to the musculoskeletal function, and changes when the muscle is stretched or contracted. The detection of the pinnate angle of the muscle can detect the pathological changes of the muscle at an early stage. An ultrasound image for the specific quantitative measurement of morphological parameters of muscle tissue is defined as a Sonography (SMG), from which muscular structural parameters such as muscle cross-sectional area, muscle thickness, pinnate angle and muscle fiber length can be obtained.
The deep learning is a hotspot in the field of machine learning in recent years, and by virtue of the ultrahigh prediction accuracy in recognition application, the deep learning obtains great attention in the field of image processing, can improve the performance of the existing image processing system and create a new application field, and is also very suitable for analysis and detection of medical images. The method has strong automatic feature extraction and complex model construction capabilities, can avoid complicated manual feature labeling, effectively utilizes unsupervised data, has excellent generalization capability, and can be applied to different medical fields. Particularly, for medical images with relatively large noise, the medical images are converted into the visual problem of deep learning, so that the performance of the medical analysis and detection system can be greatly improved, and therefore, the medical analysis and detection system has attracted extensive attention of researchers in the medical field.
Kalman filtering (Kalman filtering) is an algorithm that uses a linear system state equation to optimally estimate the state of a system by inputting and outputting observation data through the system. A typical example of kalman filtering is to predict the coordinate position and velocity of an object from a finite set of observations of the position of the object, including noise. Its silhouette can be found in many engineering applications (radar, computer vision). The feather angle transformation of the muscle is continuous and regularly repeatable, so if kalman filter tracking is applied to the tracking system of the feather angle, it is suitable.
The clinician usually determines these muscle structure parameters by means of manual labeling, which is time-consuming, labor-intensive, subjective and may affect the accuracy of the parameters. The prior art is mainly based on image processing methods, and includes the following steps:
(1) a method for tracking the direction of muscle fibers in a muscle ultrasound image based on hough transform was proposed in 2007. This study proposed an improved hough transform, a voting-based hough transform (RVHT), aimed at automatically estimating the direction of straight line patterns, such as muscle fibers and muscle-bone interfaces in ultrasound images. Firstly, the original ultrasonic image needs to be subjected to edge detection, the data volume is greatly reduced by the edge detection, irrelevant information is removed, and important structural attributes of the muscle ultrasonic image, namely the texture of the muscle, are reserved. Secondly, the edge mapping of the ultrasound image is converted into hough space, and a global peak corresponding to the position of the straight line in the image space is found. Then, all the feature points near the detection line are removed from the edge map, and the same process of searching for another straight line is performed using the updated edge map. The iteration is repeated until a predetermined termination condition is met. Thus, the method will extract the straight lines in the edge map one by one until the maximum in Hough space is below a specified threshold.
(2) The muscle fibers in the muscle ultrasound image were tracked by Radon Transform (RT). The method comprises the steps of firstly carrying out edge detection on an original ultrasonic image, reducing data volume and reserving important structural attributes of the muscle ultrasonic image. Edge detection converts the original grayscale image into a binary black-and-white edge image, and uses a local radon transform algorithm for the detected edge image. The radon transform is an integral transform that transforms a two-dimensional plane function f into a linear function Rf defined on a two-dimensional space, and the value of Rf is the value of the function f integrating the line Rf. For the detection of muscle bundles, some a priori knowledge is used for localization. For example, the muscle fibers must lie between a superficial and a deep sarcolemma line, which generally have a high echogenic intensity and are relatively easy to detect. The periosteal thread can first be extracted over the entire image. The cluster of bundles can then be found in the area between the myofascial lines. This area can be further divided into smaller parts until the muscle bundles are correctly detected.
(3) And (3) adding a Bayesian Kalman filtering algorithm to track the direction of the muscle fiber on the basis of the step (2). First, a Bayesian Kalman Filter (BKF) tracking strategy is used to predict the direction of muscle fibers from the current frame to the next frame. And the BKF represents the system state by adopting Gaussian Mixture (GM), and performs optimal state estimation according to the observed value and the given state dynamic. In addition, a new direct density simplification algorithm is adopted, and the use of a resampling technology in the traditional Particle Filtering (PF) is avoided. Finally, a set of local radon transforms is used to extract the fiber orientation from the enhanced image, providing observations for the BKF corrections.
In the prior art, firstly, the angles of a myolemal line and a myofiber bundle are calculated for an ultrasonic image by using voting Hough transform, and although the Hough transform has a good effect in locating the myolemal, the myofiber cannot be well located; secondly, calculating the angle of a muscle membrane and a muscle fiber bundle by using local radon transform of the ultrasonic image, wherein the radon transform performance is obviously reduced when the muscle fibers are unclear and the noise is large; the method adds a Kalman filter to improve the performance on the basis of the original method, but the method still has larger limitation because the method also depends on the measurement value of the traditional image processing method, namely the detection of local radon transform. The muscle fiber has the characteristics of intermittence and promptness, and the robustness of the image processing-based method is not strong.
Disclosure of Invention
In order to solve the technical problem, the invention discloses a method for automatically tracking a muscle pinnate angle by combining a convolutional neural network and Kalman filtering.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for automatically tracking a muscle pinnate angle by combining a convolutional neural network and Kalman filtering specifically comprises the following steps:
step (1), ultrasonic image preprocessing;
step (2), detecting the sarcolemma;
step (3), obtaining an observed value of the muscle fiber direction;
step (4), correcting the direction of muscle fibers;
and (5) calculating the pinnate angle.
Further, in the step (2), the myofascial detection specifically comprises the following processes:
(21) carrying out Gaussian filtering operation on the image by adopting a canny edge detection algorithm to smooth the image; selecting an operator to calculate the gradient value and the gradient direction of the image to obtain a brightness difference approximate value;
(22) carrying out local radon transformation on the edge graph;
(23) marking the myolemma by the straight line with the maximum probability of the point location gradient map, and calculating the vector of the myolemma
Further, in step (21), in the canny edge detection algorithm, the following formula is adopted:
wherein, A is original image, GxAnd GyThe images are detected by the transverse and longitudinal edges, respectively, and theta is the gradient direction.
Further, in step (23), the edge map is subjected to local radon transform, using the following formula:
where in the x-y image plane, F (x, y) is the image intensity at location (x, y) as the Dirac function, ρ is the distance from the origin to the line, and θ is the angle between the x-axis and the line perpendicular to the line from the origin.
Further, in the step (3), a reference line is marked randomly, then whether the reference line is parallel to the muscle fiber direction is judged through a neural network, if not, adjustment is carried out until the neural network judges that the reference line is parallel, and the angle is recorded and used as an observed value of Kalman filtering.
Further, in step (4), using a kalman filter tracking system, correcting the muscle fiber direction of the current frame according to the muscle fiber direction of the previous frame and the measurement of the muscle fiber angle in step (3), and predicting the muscle fiber direction of the next frame.
Further, the step (4) specifically includes:
(41) predicting a system for a next state using a process model of the system;
(42) combining the predicted value and the measured value obtained in the last step to obtain the optimized estimated value of the current state, namely the optimized muscle fiber vector
(43) And obtaining the optimal estimated value under the current state, and updating the covariance.
Further, in step (5), a feather angle value is calculated from the vector of the myofascial line and the myofascial bundle line:
The beneficial effect of the invention is that,
(1) the algorithm has strong robustness. And compared with the result obtained by the traditional image processing method, the method using deep learning in the measurement stage is more reliable and can be applied to the ultrasonic image with poor quality. On the basis, the robustness of tracking the feather angle by using a Kalman filtering algorithm is stronger.
(2) The algorithm can be suitable for various front-end ultrasonic image acquisition equipment without adjustment, and when new ultrasonic equipment exists, the algorithm effective to the new equipment can be obtained only by adding the new image into a training set for fine adjustment.
In a word, the direction of the current muscle fiber is measured by utilizing a deep convolution neural network, and the feather-shaped angle tracking is realized by combining with a Kalman filter. The method improves the robustness of the pinnate angle calculation algorithm, expands the application field of the automatic labeling pinnate angle algorithm, and provides a method for automatically tracking the pinnate angle for an ultrasonic image sequence with poor quality.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 shows a muscle fiber orientation measurement process of the ultrasound image;
the system of fig. 3 outputs the results.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention relates to a method for tracking a muscle pinnate angle by an ultrasonic image based on deep learning and Kalman filtering, which measures the pinnate angle of a pinnate muscle tissue from the ultrasonic image of the pinnate muscle by utilizing a deep residual convolution neural network in the deep learning, corrects the value by using a Kalman filtering algorithm to obtain alpha, and the parameter has important guiding significance for function evaluation, disease diagnosis and monitoring, later-stage rehabilitation plan formulation, tissue function prognosis and the like. The invention expands the new application field of the ultrasonic imaging technology and realizes the automatic tracking of the muscle pinnate angle.
A method for automatically tracking muscle pinnate angle by combining convolutional neural network and Kalman filtering comprises the following steps:
step one, ultrasonic image preprocessing
The raw data stored by the ultrasound equipment often contains information other than images, such as time, patient name, equipment model, etc. Therefore, the image sequence needs to be further cropped, denoised, and interfered, and only the ultrasound image is left.
Step two, detecting the myolemma
The edge map of the ultrasonic image is obtained by convolution calculation, and the muscle membrane is positioned by detecting straight line segments in the edge map by using local radon transform.
The detection of the myofascium specifically comprises the following processes:
(1) the original ultrasound image is derived to obtain a gradient map of the original image to clarify the edges of the sarcolemma therein. Firstly, carrying out Gaussian filtering operation on an image by adopting a canny edge detection algorithm to smooth the image; secondly, an operator is selected to calculate the gradient value and the gradient direction of the image, taking the Sobel operator as an example, the operator comprises two groups of 3 x 3 matrixes which are respectively in the transverse direction and the longitudinal direction, and the transverse direction and the longitudinal direction of the matrix are subjected to plane convolution with the image, so that the transverse direction and the longitudinal direction of the brightness difference approximate value can be obtained respectively.
In the canny edge detection algorithm, the following formula is adopted:
wherein, A is original image, GxAnd GyThe images are detected by the transverse and longitudinal edges, respectively, and theta is the gradient direction.
In addition, canny's algorithm further determines the image boundaries using a non-maximum suppression and dual threshold approach.
(2) Local radon transform on edge maps
The radon transform is an integral transform that integrates a function f (x, y) defined on a two-dimensional plane along any straight line on the plane, which is equivalent to performing a CT scan on the function f (x, y). Since the position and angle of the deep and superficial myofascial line are within a certain range, the term is detected as a priori knowledge. And carrying out local radon transformation on the edge graph by adopting the following formula:
where in the x-y image plane, F (x, y) is the image intensity at location (x, y) as the Dirac function, ρ is the distance from the origin to the line, and θ is the angle between the x-axis and the line perpendicular to the line from the origin.
(3) Marking the myolemma by the straight line with the maximum probability of the point location gradient map, and calculating the vector of the myolemma
Step three, obtaining the observed value of the muscle fiber direction
And randomly marking a reference line, judging whether the reference line is parallel to the muscle fiber direction through a neural network, if not, adjusting until the neural network judges that the reference line is parallel, and recording the angle as an observed value of Kalman filtering.
The specific operation steps of the deep learning model for detecting the muscle fiber direction can be divided into two parts of training and detecting:
(1) the training section provides a method of generating a database of myofiber ultrasound images, comprising: randomly taking a point in the middle area of two muscle membranes in an ultrasonic image as a coordinate of an upper left corner, and determining a square image with the size of 224 multiplied by 224 pixels as a muscle fiber image according to each upper left corner coordinate; in the process, the square is randomly rotated relative to the original image to expand the training set.
To improve accuracy, reference line types that float up and down parallel to the muscle fibers are labeled, also as part of the training set.
(2) In the process of processing the image, generally, the upper left corner of the image is used as an origin, a reference line parallel to muscle fibers in the ultrasonic image is marked, the slope k of the reference line is used as a standard, a point is randomly selected at the left side boundary of the image to be used as a reference line starting point A, a reference line end point B on the right side boundary of the image is determined according to the reference line starting point, and an end point is determined according to the method that: according to k and A, if the reference line parallel muscle fiber needs to be generated, the slope of the straight line of the point B and the point A is equal to k; if the reference line is required to be generated to be higher than the muscle fiber direction image, the slope of a straight line where the point B and the point A are located is smaller than k; if the reference line to be generated is lower than the muscle fiber, the slope of the straight line of the point B and the point A is larger than k;
(3) after generating the database, training a deep residual convolutional neural (ResNet) network, normalizing and transforming the muscle fiber ultrasonic images of the training sample and the test sample into pix' for each pixel pix in each image, and then:
the values of the muscle bundle ultrasound image are normalized from an integer of [0,255] to a floating point of [ -1,1] for deep neural network computation.
(4) And inputting the images with different reference lines into the trained network, and if the output result is not parallel, adjusting the slope of the reference lines according to the output result until the output result is parallel.
(5) And when the output result of the neural network is parallel, recording the angle as an observed value of the next Kalman filtering algorithm.
Step four, correcting the muscle fiber direction
And (3) correcting the muscle fiber direction of the current frame according to the muscle fiber direction of the previous frame and the measurement of the muscle fiber angle in the step (3) by using a Kalman filtering tracking system, and predicting the muscle fiber direction of the next frame.
The method specifically comprises the following steps:
(1) predicting a system for a next state using a process model of the system;
assuming that the present system state is k, according to the model of the system, the present state can be predicted based on the last state of the system:
X(k|k-1)=AX(k-1|k-1) (6)
in equation (6), X (k | k-1) is the result of prediction using the previous state, and X (k-1| k-1) is the optimum result of the previous state, i.e., the angle of the muscle fiber direction in the previous frame.
Wherein, thetakFor the prediction of the prior model, i.e. the angle value, U, predicting the current muscle fibre directionkFor the speed of the angular change of the direction of the muscle fiber, A is the state transition matrix, and the setting in this system is shown as (8).
P(k|k-1)=AP(k-1|k-1)A′+Q (9)
The covariance of X (k | k-1) is updated using equation (9), P represents the covariance (covariance), P (k | k-1) is the covariance for X (k | k-1), P (k-1| k-1) is the covariance for X (k-1| k-1), A' represents the transposed matrix for A, and Q is the covariance of the system process.
(2) And combining the predicted value and the measured value obtained in the previous step to obtain an optimized estimated value X (k | k) of the current state (k):
X(k|k)=X(k|k-1)+Kg(k)(Z(k)-HX(k|k-1)) (10)
wherein, Z (k) is the measured value at the time k, namely the measurement of the current frame neural network on the muscle fiber direction angle; h is a parameter of the measurement system.
Where Kg (k) is Kalman Gain (Kalman Gain), where R is the covariance of the measurement noise.
(3) Obtaining the optimal estimation value under the current state and updating the covariance
Obtaining the optimal estimated value X (k | k) in the k state, and keeping the Kalman filter running continuously until the Kalman filter is straightened
By the end of the system process, the covariance of X (k | k) in k state needs to be updated:
P(k|k)=(I-Kg(k)H)P(k|k-1) (13)
and step five, calculating the pinnate angle.
The vector of the muscle membrane line and the muscle fiber bundle line is used for calculating the pinnate angle value:
The invention provides a method and a system for tracking a pinnate angle in a muscle ultrasonic image sequence by utilizing a method combining deep learning and Kalman filtering through evaluating the clinical requirements of the characteristics and functions of muscle tissues by ultrasonic muscle imaging,
the direction of the current muscle fiber is measured by utilizing a deep convolution neural network, and the feather-shaped angle tracking is realized by combining with a Kalman filter. The method improves the robustness of the pinnate angle calculation algorithm, expands the application field of the automatic labeling pinnate angle algorithm, and provides a method for automatically tracking the pinnate angle for an ultrasonic image sequence with poor quality.
The method of the invention has the following brief overview:
(1) the direction of the muscle fibers is obtained using a deep learning method.
And (3) marking a reference line for each ultrasonic muscle fiber image at random, judging whether the direction of the reference line is parallel to the direction of the muscle fiber bundle through a neural network, if not, adjusting until the direction of the reference line is parallel to the direction of the muscle fiber bundle through the neural network, and outputting the angle of the reference line.
(2) The direction of muscle fibers from the current frame to the next frame is predicted using a Kalman Filter (KF) tracking strategy. And (3) observing the current measured value of the muscle fiber direction by using the method (1) for KF correction, and finally obtaining the direction of the corrected muscle fiber reference line.
(3) And obtaining the direction of the deep and shallow myofascial lines by using a local radon transform method.
(4) And (4) acquiring the value of the current feather angle according to the results of the (2) and the (3).
In order to verify the feasibility and effectiveness of the invention, ultrasonic images of gastrocnemius muscle of normal people and patients with muscular atrophy, which are acquired by an ultrasonic diagnostic apparatus, are analyzed, and the pinnate angle alpha of muscle tissue is estimated from the ultrasonic images by adopting the method provided by the invention, as shown in fig. 3.
In addition, in the present invention:
(1) the detection of the myofascial line is changed into other algorithms, such as a radon algorithm, a hough transform and the like.
(2) The invention can use not only the deep residual error neural network, but also the system can obtain better effect under the condition of using other deep convolution neural networks through experiments, such as AlexNet, VGG-Net and the like.
(3) The kalman tracking section may also use a variation of the kalman algorithm to achieve tracking.
It is to be understood that the above description is not intended to limit the present invention, and the present invention is not limited to the above examples, and those skilled in the art may make modifications, alterations, additions or substitutions within the spirit and scope of the present invention.
Claims (8)
1. A method for automatically tracking a muscle pinnate angle by combining a convolutional neural network and Kalman filtering is characterized by comprising the following steps:
step (1), ultrasonic image preprocessing;
step (2), detecting the sarcolemma;
step (3), obtaining an observed value of the muscle fiber direction;
step (4), correcting the direction of muscle fibers;
and (5) calculating the pinnate angle.
2. The method for automatically tracking the muscle pinnate angle by combining the convolutional neural network with the kalman filter as claimed in claim 1, wherein the detection of the sarcolemma in step (2) specifically comprises the following processes:
(21) carrying out Gaussian filtering operation on the image by adopting a canny edge detection algorithm to smooth the image; selecting an operator to calculate the gradient value and the gradient direction of the image to obtain a brightness difference approximate value;
(22) carrying out local radon transformation on the edge graph;
3. The method for automatically tracking muscle pinnate angle by using a convolutional neural network in combination with kalman filtering as claimed in claim 2, wherein in the step (21), the following formula is adopted in the canny edge detection algorithm:
wherein, A is original image, GxAnd GyThe images are detected by the transverse and longitudinal edges, respectively, and theta is the gradient direction.
4. The method for automatically tracking muscle pinnate angle by using the combination of the convolutional neural network and the kalman filter as claimed in claim 2, wherein in the step (23), the local radon transform is performed on the edge map by using the following formula:
where in the x-y image plane, F (x, y) is the image intensity at location (x, y) as the Dirac function, ρ is the distance from the origin to the line, and θ is the angle between the x-axis and the line perpendicular to the line from the origin.
5. The method for automatically tracking the muscle pinnate angle by combining the convolutional neural network and the kalman filter as claimed in claim 1, wherein in the step (3), a reference line is marked randomly, then the neural network is used to judge whether the reference line and the muscle fiber direction are parallel, if not, the adjustment is carried out until the neural network is parallel, and the angle is recorded and used as the observed value of the kalman filter.
6. The method of claim 1, wherein in step (4), a Kalman filter tracking system is used to correct the direction of the muscle fiber of the current frame according to the direction of the muscle fiber of the previous frame and the measurement of the angle of the muscle fiber in step (3) and predict the direction of the muscle fiber of the next frame.
7. The method for automatically tracking the muscle pinnate angle by combining the convolutional neural network with the kalman filter as claimed in claim 6, wherein the step (4) specifically comprises:
(41) predicting a system for a next state using a process model of the system;
(42) combining the predicted value and the measured value obtained in the last step to obtain the optimized estimated value of the current state, namely the optimized muscle fiber vector
(43) And obtaining the optimal estimated value under the current state, and updating the covariance.
8. The method for automatically tracking the muscle feather angle by combining the convolutional neural network with the kalman filter as claimed in claim 1, wherein in the step (5), the feather angle value is calculated by the vector of the muscle membrane line and the muscle fiber bundle line:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010865743.0A CN112085711B (en) | 2020-08-25 | 2020-08-25 | Method for automatically tracking muscle feather angle by combining convolutional neural network and Kalman filtering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010865743.0A CN112085711B (en) | 2020-08-25 | 2020-08-25 | Method for automatically tracking muscle feather angle by combining convolutional neural network and Kalman filtering |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112085711A true CN112085711A (en) | 2020-12-15 |
CN112085711B CN112085711B (en) | 2023-12-22 |
Family
ID=73729006
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010865743.0A Active CN112085711B (en) | 2020-08-25 | 2020-08-25 | Method for automatically tracking muscle feather angle by combining convolutional neural network and Kalman filtering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112085711B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070022068A1 (en) * | 2005-07-01 | 2007-01-25 | Ralph Linsker | Neural networks for prediction and control |
CN110111386A (en) * | 2019-05-10 | 2019-08-09 | 深圳大学 | The method for automatic measurement and system of structural angle in a kind of musculature |
CN110930394A (en) * | 2019-11-29 | 2020-03-27 | 深圳先进技术研究院 | Method and terminal equipment for measuring slope and pinnate angle of muscle fiber bundle line |
-
2020
- 2020-08-25 CN CN202010865743.0A patent/CN112085711B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070022068A1 (en) * | 2005-07-01 | 2007-01-25 | Ralph Linsker | Neural networks for prediction and control |
CN110111386A (en) * | 2019-05-10 | 2019-08-09 | 深圳大学 | The method for automatic measurement and system of structural angle in a kind of musculature |
CN110930394A (en) * | 2019-11-29 | 2020-03-27 | 深圳先进技术研究院 | Method and terminal equipment for measuring slope and pinnate angle of muscle fiber bundle line |
Non-Patent Citations (1)
Title |
---|
曾公任等: ""基于神经网络与卡尔曼滤波的手部实时追踪方法"", 《液晶与显示》, vol. 35, no. 5, pages 464 - 468 * |
Also Published As
Publication number | Publication date |
---|---|
CN112085711B (en) | 2023-12-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11633169B2 (en) | Apparatus for AI-based automatic ultrasound diagnosis of liver steatosis and remote medical diagnosis method using the same | |
CN107480677B (en) | Method and device for identifying interest region in three-dimensional CT image | |
US10004448B2 (en) | Osteoporosis diagnostic support apparatus | |
CN107610162A (en) | A kind of three-dimensional multimode state medical image autoegistration method based on mutual information and image segmentation | |
US11455720B2 (en) | Apparatus for ultrasound diagnosis of liver steatosis using feature points of ultrasound image and remote medical-diagnosis method using the same | |
CN109124662B (en) | Rib center line detection device and method | |
CN112258514B (en) | Segmentation method of pulmonary blood vessels of CT (computed tomography) image | |
CN110782428B (en) | Method and system for constructing clinical brain CT image ROI template | |
CN109801276B (en) | Method and device for calculating heart-chest ratio | |
Zheng et al. | Fast and automatic heart isolation in 3D CT volumes: Optimal shape initialization | |
CN116580068B (en) | Multi-mode medical registration method based on point cloud registration | |
CN111652844B (en) | X-ray defect detection method and system based on digital image region growing | |
CN112215878B (en) | X-ray image registration method based on SURF feature points | |
CN116580033B (en) | Multi-mode medical image registration method based on image block similarity matching | |
CN113012127A (en) | Cardiothoracic ratio measuring method based on chest medical image | |
CN109816665B (en) | Rapid segmentation method and device for optical coherence tomography image | |
CN115661152B (en) | Target development condition analysis method based on model prediction | |
CN112085711A (en) | Method for automatically tracking muscle pinnate angle by combining convolutional neural network and Kalman filtering | |
CN115861230A (en) | Method and system for determining lung region by thoracic cavity CT image segmentation | |
CN113112473B (en) | Automatic diagnosis system for human body dilated cardiomyopathy | |
CN115239751A (en) | Postoperative image needle track extraction method | |
CN111640126B (en) | Artificial intelligent diagnosis auxiliary method based on medical image | |
CN115222878A (en) | Scene reconstruction method applied to lung bronchoscope surgical robot | |
CN114240893A (en) | Method for measuring and calculating spinal Cobb angle in external image | |
CN109712124B (en) | Ultrasonic image mark removing method and device |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |