CN114176514B - Vein blood vessel identification positioning method and system based on near infrared imaging - Google Patents
Vein blood vessel identification positioning method and system based on near infrared imaging Download PDFInfo
- Publication number
- CN114176514B CN114176514B CN202111353446.9A CN202111353446A CN114176514B CN 114176514 B CN114176514 B CN 114176514B CN 202111353446 A CN202111353446 A CN 202111353446A CN 114176514 B CN114176514 B CN 114176514B
- Authority
- CN
- China
- Prior art keywords
- blood vessel
- infrared imaging
- infrared
- wavelength
- vein
- 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.)
- Active
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/48—Other medical applications
- A61B5/4887—Locating particular structures in or on the body
- A61B5/489—Blood vessels
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
- A61B5/0062—Arrangements for scanning
- A61B5/0064—Body surface scanning
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
- A61B5/0075—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Vascular Medicine (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
- Image Input (AREA)
Abstract
The invention relates to a vein blood vessel identification positioning method and system based on near infrared imaging, wherein the method comprises the steps of respectively shooting infrared imaging pictures at a first infrared wavelength and a second infrared wavelength through a near infrared imaging camera, wherein the infrared imaging pictures are respectively a first infrared imaging picture and a second infrared imaging picture; preprocessing a first infrared imaging picture, and enhancing the characteristics of vein blood vessels to obtain an enhanced image; inputting the enhanced image into a first vein blood vessel segmentation network model, and carrying out blood vessel distribution recognition to obtain a segmentation image; inputting the segmented image into a second vein blood vessel segmentation network model, and labeling blood vessels suitable for puncture to obtain labeled blood vessels; extracting the position of the connected domain to obtain a target puncture point and an optimal puncture angle; and calculating the depth information of the marked blood vessel according to the light intensity change of the marked blood vessel in the first infrared imaging picture and the second infrared imaging picture. Compared with the prior art, the method has the advantages of high identification precision, good robustness and the like.
Description
Technical Field
The invention relates to the technical field of vein blood vessel identification, in particular to a vein blood vessel identification positioning method and system based on near infrared imaging.
Background
Venipuncture is a common medical practice, and is currently in the technical stage of manual puncture. Because of the limitation of human eyes, the manual puncture is often performed for obese people with higher BMI and infants with finer blood vessels for multiple times due to the error of puncture positions; in addition, the puncture angle and the puncture pose are difficult to continuously and stably judge accurately. At the same time, hospitals are also expensive to pay for developing a technically mature nurse.
With the rapid development of smart medicine, the market is increasingly concerned about venipuncture blood collection/injection robots. For the existing vein automatic identification technology, the vein network of a 2D plane is generally identified by virtue of near infrared, and the identification of the depth of a blood vessel is carried out by combining ultrasound. However, in the prior art, two kinds of ultrasonic and infrared sensor data are adopted, the two kinds of independent sensor data are used for describing the split of the space information of the blood vessel, the recognition accuracy of the position of the blood vessel is easily reduced when fusion processing is carried out subsequently, and the error is most obvious in the recognition of the space angle of the blood vessel.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a vein blood vessel identification and positioning method and system based on near infrared imaging.
The aim of the invention can be achieved by the following technical scheme:
a vein blood vessel identification positioning method based on near infrared imaging, comprising:
s1, respectively shooting infrared imaging pictures at a first infrared wavelength and a second infrared wavelength through a near infrared imaging camera, wherein the infrared imaging pictures are respectively a first infrared imaging picture and a second infrared imaging picture;
s2, preprocessing a first infrared imaging picture, and enhancing the characteristics of vein blood vessels to obtain an enhanced image;
s3, inputting the enhanced image into a first vein blood vessel segmentation network model, and carrying out blood vessel distribution recognition to obtain a segmentation image;
s4, inputting the segmented image into a second vein blood vessel segmentation network model, and labeling blood vessels suitable for puncture to obtain labeled blood vessels; meanwhile, extracting the positions of the communicating domains from the marked blood vessel to obtain the central points of the communicating domains in the marked blood vessel, rotating a plurality of straight lines passing through the communicating domains at different angles by taking each central point as the center, selecting the straight line which is most intersected with the communicating domains from the straight lines, taking the central point corresponding to the selected straight line as a target puncture point, and taking the slope of the straight line as the optimal puncture angle;
and S5, calculating the depth of the body surface fat layer of the marked blood vessel according to the light intensity change of the marked blood vessel in the first infrared imaging picture and the second infrared imaging picture, namely the depth information of the marked blood vessel.
Further, the first vein blood vessel segmentation network model and the second vein blood vessel segmentation network model both adopt a TransUNet neural network model.
Further, the depth information calculation expression of the labeled blood vessel is as follows:
wherein χ represents a deformation coefficient, I 1 Representing a first infrared waveLong strength, Δi 1 Representing the fat absorption intensity at the first infrared wavelength, i.e. the difference between the emitted light intensity and the reflected light intensity at the first infrared wavelength, I 2 Represents the intensity of the second infrared wavelength, ΔI 2 Representing the fat absorption intensity at the second infrared wavelength, namely the difference between the emitted light intensity and the reflected light intensity at the second infrared wavelength, eta is the light intensity absorption correction coefficient, mu 1 Sum mu 2 Representation I 1 And I 2 Absorption coefficient of fat for light at both light intensities.
Further, mu 1 Sum mu 2 By a general mu a The representation is:
where k is the imaginary part of the refractive index of water,lambda is the wavelength of incident light, which is the volume fraction of water in the tissue.
Further, the preprocessing comprises gray processing of the image, binarization processing of the image through a maximum inter-class variance method, and enhancement of the characteristics of venous blood vessels in the image through a multiscale filtering method based on a Hessian matrix.
Further, the first infrared wavelength is 850nm and the second infrared wavelength is 940nm.
Further, in the step S1, the infrared imaging camera respectively shoots a plurality of infrared imaging pictures at the first infrared wavelength and the second infrared wavelength, and the sharpest image is selected as the first infrared imaging picture and the second infrared imaging picture.
8. A vein blood vessel identification positioning system based on near infrared imaging, which is characterized by comprising a processor and a memory, wherein the processor calls a program in the memory to execute the following steps:
s1, respectively shooting infrared imaging pictures at a first infrared wavelength and a second infrared wavelength through a near infrared imaging camera, wherein the infrared imaging pictures are respectively a first infrared imaging picture and a second infrared imaging picture;
s2, preprocessing a first infrared imaging picture, and enhancing the characteristics of vein blood vessels to obtain an enhanced image;
s3, inputting the enhanced image into a first vein blood vessel segmentation network model, and carrying out blood vessel distribution recognition to obtain a segmentation image;
s4, inputting the segmented image into a second vein blood vessel segmentation network model, and labeling blood vessels suitable for puncture to obtain labeled blood vessels; meanwhile, extracting the positions of the communicating domains from the marked blood vessel to obtain the central points of the communicating domains in the marked blood vessel, rotating a plurality of straight lines passing through the communicating domains at different angles by taking each central point as the center, selecting the straight line which is most intersected with the communicating domains from the straight lines, taking the central point corresponding to the selected straight line as a target puncture point, and taking the slope of the straight line as the optimal puncture angle;
and S5, calculating the depth of the body surface fat layer of the marked blood vessel according to the light intensity change of the marked blood vessel in the first infrared imaging picture and the second infrared imaging picture, namely the depth information of the marked blood vessel.
9. The near infrared imaging-based venous blood vessel identification positioning system according to claim 8, wherein the depth information calculation expression of the labeled blood vessel is:
wherein χ represents a deformation coefficient, I 1 Represents the intensity of the first infrared wavelength, ΔI 1 Representing the fat absorption intensity at the first infrared wavelength, i.e. the difference between the reflected intensity and the emitted intensity at the first infrared wavelength, I 2 Represents the intensity of the second infrared wavelength, ΔI 2 Representing the fat absorption intensity at the second infrared wavelength, namely the difference between the reflected light intensity and the emitted light intensity at the second infrared wavelength, eta is the light intensity absorption correction coefficient, mu 1 Sum mu 2 Representation ofI 1 And I 2 Absorption coefficient of fat for light at both light intensities.
10. The near infrared imaging-based venous blood vessel identification positioning system as claimed in claim 9, wherein μ 1 Sum mu 2 By a general mu a The representation is:
where k is the imaginary part of the refractive index of water,lambda is the wavelength of incident light, which is the volume fraction of water in the tissue.
Compared with the prior art, the invention has the following beneficial effects:
1. the recognition of the blood vessel position and the depth can be realized only by adopting near infrared data, additional ultrasonic equipment is not needed, and meanwhile, the shooting data of the same camera and the same position are adopted, so that the high unification of the depth information and the blood vessel position can be ensured, and the recognition precision and the robustness are obviously improved.
2. The two neural network models are adopted for twice recognition, the first time is used for vascular distribution recognition, the second time is used for vascular selection, compared with the traditional neural network recognition, the network detection precision of double-in and double-out is far better than the precision of single-in and single-out from the result of an actual experiment, and the network detection precision is applied to medical recognition and has higher reliability and accuracy.
3. The venous vessel segmentation network model adopts a TransUNet neural network model, and the model combines CNN and a Transformer as encoders, wherein the CNN and the Transformer pay attention to global information, and the TransUNet neural network model encodes local details. In addition, skip-connection in the Unet network is reserved, the characteristic diagram generated by coding and the characteristic diagram restored by the corresponding decoding part are spliced together through long connection, and characteristic information loss caused by deepening the network layer number in the downsampling construction is avoided.
Drawings
Fig. 1 is a general flow chart of a near infrared imaging-based blood vessel image recognition method of the present invention.
Fig. 2 is a training flow chart of the vessel segmentation network of the present invention.
Fig. 3 is a view showing the effect of preprocessing a blood vessel image according to the present invention.
Fig. 4 is a view showing the effects of the vessel segmentation network according to the present invention.
Detailed Description
The invention will now be described in detail with reference to the drawings and specific examples. The present embodiment is implemented on the premise of the technical scheme of the present invention, and a detailed implementation manner and a specific operation process are given, but the protection scope of the present invention is not limited to the following examples.
As shown in fig. 1, the present embodiment provides a vein blood vessel identification positioning method based on near infrared imaging, which includes the following steps:
s1, respectively shooting infrared imaging pictures at a first infrared wavelength and a second infrared wavelength through a near infrared imaging camera, wherein the infrared imaging pictures are a first infrared imaging picture and a second infrared imaging picture;
s2, preprocessing a first infrared imaging picture, and enhancing the characteristics of vein blood vessels to obtain an enhanced image;
s3, inputting the enhanced image into a first vein blood vessel segmentation network model, and carrying out blood vessel distribution recognition to obtain a segmentation image;
s4, inputting the segmented image into a second vein blood vessel segmentation network model, and labeling blood vessels suitable for puncture to obtain labeled blood vessels;
and S5, calculating the depth of the body surface fat layer of the marked blood vessel according to the light intensity change of the marked blood vessel in the first infrared imaging picture and the second infrared imaging picture, namely the depth information of the marked blood vessel.
As shown in fig. 4, the collected infrared images can be specifically trained and processed through the above five steps, and finally suitable puncture points, puncture angles and puncture positions are output.
The following are specific developments:
1. step S1
When the near infrared imaging camera shoots pictures, respectively shooting a plurality of infrared imaging pictures at a first infrared wavelength and a second infrared wavelength, and selecting an image with highest definition as the first infrared imaging picture and the second infrared imaging picture. In this embodiment, the first infrared wavelength is 850nm, and the second infrared wavelength is 940nm; the peak value of absorbing infrared light is near 940nm wavelength, and the effect of absorbing infrared light is smaller at 850nm wavelength; these two wavelengths are most useful for detecting fat depth.
2. Step S2
The preprocessing of the first infrared imaging picture is as follows
S21, cutting off a black area on the first infrared imaging picture and machine information;
s22, carrying out gray processing on the picture, and if the input image is a color image, processing the image into a black-and-white image, wherein the gray value range is 0-255, and the brightness is from deep to light, and the color is from black to white. Normalizing the picture pixels;
the specific formula of normalization is:
wherein g represents the pixel value of each pixel point in the picture;
s23, performing binarization processing on the gray level map obtained in the S22 through an OTSU algorithm (maximum inter-class variance method);
s24, performing blood vessel enhancement processing on the blood vessel image by using a multiscale filtering method based on a Hessian matrix.
Through the binarization processing and filtering of the first infrared imaging picture, clearer sample input is provided for subsequent vein network segmentation, and a better vein network segmentation effect is achieved.
3. Step S3 and step S4
In the two steps, a TransUNet neural network model is adopted for the first vein blood vessel segmentation network model and the second vein blood vessel segmentation network model.
As shown in fig. 2, the first venous vessel segmentation network model aims at segmenting the contours of the blood vessels. The method comprises data processing, model building, model training, result output and the like.
The construction process of the first vein blood vessel segmentation network model comprises the following steps:
and (3) taking a series of pictures by a near infrared imaging camera and preprocessing in the step S2 to obtain a training data set. Then, the method comprises the following steps of: 3: the 3 scale divides the training data set into a target training set, a target validation set and a target test set. In order to increase the number of the training sisters in the network, the situation that the network is over-fitted is avoided, and operations such as rotation, translation, random scaling, shielding and the like are performed on the target training set. The division ratio is only an example in the present embodiment, and is not limited thereto; the dataset is in this embodiment submitted to labeling of the venous vessel edges by a specialist.
Building a structure of a TransUNet neural network model;
inputting the target training set into a TransUNet network, and training the target training set;
and inputting the target verification set and the target test set into a TransUNet network, and training the TransUNet network, so as to obtain the target segmentation effect of the network. Alternatively, in this embodiment, the effect of vein segmentation is evaluated using a dice value and a Hausdorff distance. Further modifications and improvements are made according to the implementation effect of the object.
In this embodiment, the TransUNet neural network model is a U-shaped network with five layers of encoding and decoding portions.
Specifically, the improved encoder in the U-shaped network of the model is as follows:
the encoder of the present network combines CNN, which focuses on local information, and a transducer, which encodes global information;
since the input of the transducer needs to be a sequence, the input image information needs to be subjected to serialization processing; assume that a near infrared ray P is obtained by shooting and is passed through a chartOperation such as preprocessing graying and the like to become single-channel gray scale map P W×H . Then the picture is sliced into pictures with resolution A×A, and the original picture can be decomposed intoLength picture sequence->
After the serialization of the pictures is completed, the near infrared picture slice sequence is compressed to the M dimension (becomes picture T using a fully connected layer W×H×M )。
At the same time, postion embedding in VIT is referenced to encode the absolute position of the picture to form one-dimensional picture position information code to form T pos 。
So that for a specific i-th picture patch, its coding function is:
at the moment of obtaining the output vector z 0 And then input to the final encoding function. The system specifically comprises an L-layer interactive multi-head attention Module (MA) and a multi-head perceptron Module (MLP), and the specific functions are as follows:
y' j ＝MSA(LN(y j-1 ))+y j-1
y j ＝MLP(LN(y' j ))+y' j
wherein y is j An output vector representing the encoded j-th layer, LN representing layer normalization, for reducing the effects of high correlation due to sample variation;
and meanwhile, coding sequences obtained by CNN corresponding to transformers of different layers are adopted to generate characteristic images of corresponding dimensions, and the characteristic images are cascaded to form a CNN-transformer coder.
In the present network, the upsampling encoding is performed five times for the near infrared picture sequence. Meanwhile, the network reserves a skip-connection structure of the Unet structure, and correspondingly, performs five downsampling and decoding, namely, gradually obtains a final decoded image by adopting five layers of decoders;
specifically, the U-shaped network decoder of the present model is as follows:
in the present network, a concatenated decoder of CUP is used, which conceals features by upsampling multiple times and concatenating multiple decodersBy means of the cascade of up-sampling blocks CUP, achieve +.>Is a full resolution of (a).
For each specific upsampled block, comprising two upsampling operators, a 3 x 3 convolutional layer, the loss function uses the ReLU function:
and after the decoded sequence is connected with the feature map acquired by the CNN network in a long distance, so that the feature map is aggregated under the aggregation resolution, and the low level features are kept, thereby realizing a better blood vessel contour recognition effect.
As in fig. 4, the first venous vascular segmentation network model may be trained from the labeled vascular profile dataset; and inputting the near infrared picture obtained after data preprocessing into a trained model, and segmenting out the vein blood vessel outline.
As shown in fig. 3, the objective of the second venous blood vessel segmentation network model is to obtain a blood vessel region suitable for puncture, and calculate and select an optimal puncture point and a corresponding puncture angle by using the connected domain.
The construction process of the second vein blood vessel segmentation network model comprises the following steps:
after the contour recognition of the vessel edge is completed, the vessel segments with too high or too thin edge curvature need to be removed. Specifically, the positions of vein bifurcation, large curvature vein segment, vein segment near the imaging edge of the near infrared camera, short vein segment and the like which are not suitable for needle insertion are manually erased, and the accurate positions suitable for needle insertion are obtained by manual marking and used as training data sets. Then, the method comprises the following steps of: 3: the 3 scale divides the training data set into a target training set, a target validation set and a target test set.
In order to increase the number of the training sisters in the network, the situation that the network is over-fitted is avoided, and operations such as rotation, translation, random scaling, shielding and the like are performed on the target training set. The division ratio is only an example in the present embodiment, and is not limited thereto; the dataset is in this embodiment submitted to labeling of the venous vessel edges by a specialist.
Building a structure of a TransUNet neural network model;
inputting the target training set into a TransUNet network, and training the target training set;
and inputting the target verification set and the target test set into a TransUNet network, and training the TransUNet network, so as to obtain the target segmentation effect of the network. Alternatively, in this embodiment, the effect of vein segmentation is evaluated using a dice value and a Hausdorff distance. Further modifications and improvements are made according to the implementation effect of the object.
As shown in fig. 4, the second venous vascular segmentation network model may be trained with labeled blood vessel data sets suitable for puncture; the output result of the first vein blood vessel segmentation network model can be used as the input of the second vein blood vessel segmentation network model, the blood vessel region suitable for puncture is identified, and the optimal puncture point and puncture angle are further calculated through a connected domain algorithm.
Inputting the result of the first vein blood vessel segmentation network model into a second vein blood vessel segmentation network model to obtain a blood vessel region suitable for puncture, and setting the output result picture as Q. And extracting the position of the connected domain of Q, wherein the method comprises the following steps of:
cv2.findcontours (grayimg. Cv2.Retr_tree, cv2. Chan_approx_none) function.
After each connected domain is obtained, it is assumed that the connected domain set is s= { S 1 ,s 2 ,……,s n And (n is the number of communication domains).
For each connected domain, calculating the corresponding center point of the connected domain, and calculating the center point of each connected domain by using a picture center moment calculation function in opencv:
cv2.moments(contours[j],binaryImage＝True)
assume that the set of center points corresponding to each connected domain is p= { P 1 ,p 2 ,……,p n And (n is the number of communication fields), these center points being the candidate set of the best puncture points.
For each center point p i A plurality of straight lines passing through the communicating domain at different angles rotate around the connecting domain as a center; selecting and connecting the domains s in the straight lines i Straight line l with most intersection i And take l i Slope theta of (2) i As the optimum puncture angle for the communicating region.
Through the steps, the l corresponding to each communication domain can be obtained i And (5) a straight line. Selecting l 0 ＝max{l i ,i∈[1,n]A target puncture point in the original image is a connected domain center point P corresponding to the straight line o And the angle corresponding to the straight line is taken as the puncture angle theta 0 。
4. Step S5
The depth information of the identified blood vessel is acquired by using the first infrared imaging picture and the second infrared imaging picture, and the principle is as follows:
since the depth of the vein of the human body is generally determined by the thickness of the epidermis, the thickness of the dermis, and the thickness of the fat layer overlying the vein. The skin epidermis layer thickness of a human arm is typically around 0.15 mm, the total skin thickness is typically around 2 mm, and the fat layer thickness varies from individual to individual. The arm vein depth is determined primarily by the fat layer thickness.
The skin epidermis layer of the human body has good transmission characteristics for infrared light. Near infrared rays irradiated to the skin of a human body can pass through the epidermis to enter subcutaneous adipose tissues mostly. Also, the backscattered light of fat is mostly able to pass through the skin. There are also two roles of scattering and absorption of light in fat. But the fat absorption coefficient is small and stable and differs at two near infrared wavelengths. There is a peak of infrared absorption around 940nm wavelength, whereas at 850nm wavelength, the effect of infrared absorption is smaller.
Since the output brightness of the infrared image is nonlinear with the input signal amplitude, gamma correction is used to calculate the brightness of the image. Assuming that a pixel point of a picture has pixels in three dimensions of R, G and B, the brightness of the pixel point can be calculated correspondingly as follows:
since the infrared light is radiated uniformly outward in a wave mode, it is assumed here that the distance the light passes when radiated to the blood vessel is d. Since it is uniformly divergent, the brightness should exhibit a linear relationship with the intensity of light reflected by the picture. Assuming that the infrared light intensity reflected by the picture is I', the following formula can be derived:
I’＝γL
the coefficient γ here is a correction coefficient of a linear relationship, and can be obtained by linear regression. The above linear regression may be performed experimentally in advance.
Thereby, the reflected light intensity I of the infrared picture can be obtained 1 The relationship between' and the brightness of the picture, and then the difference of the infrared energy absorbed at two wavelengths is used to calculate the fat thickness. Since the adipocyte size is much larger than the wavelength of near infrared light, there is strong back scattering in fat. The thickness of fat can be estimated using the back-scattered light difference. First, two kinds of light are at the fat layer on the surface of the human body, and the absorption coefficient of the light is:
where k is the imaginary part of the refractive index of water,for the volume fraction of water in the tissue (for a fixed constant of 0.81) Lambda is the wavelength of the incident light.
The depth of the body surface fat layer of the human body can be calculated through infrared light with two wavelengths, and the method specifically comprises the following steps:
wherein I is 1 Represents the initial light intensity, ΔI, of an infrared light wave having a wavelength of 850mm 1 Representing the intensity of fat absorption (difference between emitted and reflected light), I 2 Represents the initial light intensity, ΔI, of an infrared light wave having a wavelength of 940mm 2 Representing the intensity of fat absorption (difference between emitted and reflected light), I 1 ' representing the reflected light intensity of the first infrared imaging picture, I 2 ' represents the reflected light intensity of the second infrared imaging picture, eta is the light intensity absorption correction coefficient, mu 1 ，μ 2 The absorption coefficient of fat for light at both light intensities is shown. In the formula, specific correction coefficients can be determined by regression.
The foregoing describes in detail preferred embodiments of the present invention. It should be understood that numerous modifications and variations can be made in accordance with the concepts of the invention by one of ordinary skill in the art without undue burden. Therefore, all technical solutions which can be obtained by logic analysis, reasoning or limited experiments based on the prior art by the person skilled in the art according to the inventive concept shall be within the scope of protection defined by the claims.
Claims (9)
1. A vein blood vessel identification positioning method based on near infrared imaging, which is characterized by comprising the following steps:
s1, respectively shooting infrared imaging pictures at a first infrared wavelength and a second infrared wavelength through a near infrared imaging camera, wherein the infrared imaging pictures are respectively a first infrared imaging picture and a second infrared imaging picture;
s2, preprocessing a first infrared imaging picture, and enhancing the characteristics of vein blood vessels to obtain an enhanced image;
s3, inputting the enhanced image into a first vein blood vessel segmentation network model, and carrying out blood vessel distribution recognition to obtain a segmentation image;
s4, inputting the segmented image into a second vein blood vessel segmentation network model, and labeling blood vessels suitable for puncture to obtain labeled blood vessels; meanwhile, extracting the positions of the communicating domains from the marked blood vessel to obtain the central points of the communicating domains in the marked blood vessel, rotating a plurality of straight lines passing through the communicating domains at different angles by taking each central point as the center, selecting the straight line which is most intersected with the communicating domains from the straight lines, taking the central point corresponding to the selected straight line as a target puncture point, and taking the slope of the straight line as the optimal puncture angle;
s5, calculating the depth of the body surface fat layer of the marked blood vessel according to the light intensity change of the marked blood vessel in the first infrared imaging picture and the second infrared imaging picture, namely the depth information of the marked blood vessel;
the first vein blood vessel segmentation network model and the second vein blood vessel segmentation network model both adopt a TransUNet neural network model.
2. The vein blood vessel identification positioning method based on near infrared imaging according to claim 1, wherein the calculation expression of the depth information of the labeled blood vessel is:
wherein χ represents a deformation coefficient, I 1 Represents the intensity of the first infrared wavelength, ΔI 1 Representing the fat absorption intensity at the first infrared wavelength, i.e. the difference between the emitted light intensity and the reflected light intensity at the first infrared wavelength, I 2 Represents the intensity of the second infrared wavelength, ΔI 2 Representing the fat absorption intensity at the second infrared wavelength, namely the difference between the emitted light intensity and the reflected light intensity at the second infrared wavelength, eta is the light intensity absorption correction coefficient, mu 1 Sum mu 2 Representation I 1 And I 2 Absorption coefficient of fat for light at both light intensities.
3. The near infrared imaging-based vein blood vessel identification positioning method as claimed in claim 2, wherein μ is 1 Sum mu 2 By a general mu a The representation is:
where k is the imaginary part of the refractive index of water,lambda is the wavelength of incident light, which is the volume fraction of water in the tissue.
4. The vein blood vessel identification positioning method based on near infrared imaging according to claim 1, wherein the preprocessing comprises gray processing of an image, binarization processing is carried out through a maximum inter-class variance method, and finally a multiscale filtering method based on a Hessian matrix is adopted to enhance the characteristics of vein blood vessels in the image.
5. The method for identifying and locating a vein based on near infrared imaging according to claim 1, wherein said first infrared wavelength is 850nm and said second infrared wavelength is 940nm.
6. The vein blood vessel identification positioning method based on near infrared imaging according to claim 1, wherein in the step S1, the infrared imaging camera respectively shoots a plurality of infrared imaging pictures at a first infrared wavelength and a second infrared wavelength, and the sharpest image is selected as the first infrared imaging picture and the second infrared imaging picture.
7. A vein blood vessel identification positioning system based on near infrared imaging, comprising a processor and a memory, wherein the processor calls a program in the memory to execute the following steps:
s1, respectively shooting infrared imaging pictures at a first infrared wavelength and a second infrared wavelength through a near infrared imaging camera, wherein the infrared imaging pictures are respectively a first infrared imaging picture and a second infrared imaging picture;
s2, preprocessing a first infrared imaging picture, and enhancing the characteristics of vein blood vessels to obtain an enhanced image;
s3, inputting the enhanced image into a first vein blood vessel segmentation network model, and carrying out blood vessel distribution recognition to obtain a segmentation image;
s4, inputting the segmented image into a second vein blood vessel segmentation network model, and labeling blood vessels suitable for puncture to obtain labeled blood vessels; meanwhile, extracting the positions of the communicating domains from the marked blood vessel to obtain the central points of the communicating domains in the marked blood vessel, rotating a plurality of straight lines passing through the communicating domains at different angles by taking each central point as the center, selecting the straight line which is most intersected with the communicating domains from the straight lines, taking the central point corresponding to the selected straight line as a target puncture point, and taking the slope of the straight line as the optimal puncture angle;
s5, calculating the depth of the body surface fat layer of the marked blood vessel according to the light intensity change of the marked blood vessel in the first infrared imaging picture and the second infrared imaging picture, namely the depth information of the marked blood vessel;
the first vein blood vessel segmentation network model and the second vein blood vessel segmentation network model both adopt a TransUNet neural network model.
8. The near infrared imaging-based venous blood vessel identification positioning system according to claim 7, wherein the depth information calculation expression of the labeled blood vessel is:
wherein χ represents a deformation coefficient, I 1 Represents the intensity of the first infrared wavelength, ΔI 1 Representing the firstFat absorption intensity at infrared wavelength, i.e. the difference between the reflected and emitted light intensity at the first infrared wavelength, I 2 Represents the intensity of the second infrared wavelength, ΔI 2 Representing the fat absorption intensity at the second infrared wavelength, namely the difference between the reflected light intensity and the emitted light intensity at the second infrared wavelength, eta is the light intensity absorption correction coefficient, mu 1 Sum mu 2 Representation I 1 And I 2 Absorption coefficient of fat for light at both light intensities.
9. The near infrared imaging-based venous blood vessel identification positioning system as claimed in claim 8, wherein μ 1 Sum mu 2 By a general mu a The representation is:
where k is the imaginary part of the refractive index of water,lambda is the wavelength of incident light, which is the volume fraction of water in the tissue.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111353446.9A CN114176514B (en) | 2021-11-16 | 2021-11-16 | Vein blood vessel identification positioning method and system based on near infrared imaging |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111353446.9A CN114176514B (en) | 2021-11-16 | 2021-11-16 | Vein blood vessel identification positioning method and system based on near infrared imaging |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114176514A CN114176514A (en) | 2022-03-15 |
| CN114176514B true CN114176514B (en) | 2023-08-29 |
Family
ID=80540225
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111353446.9A Active CN114176514B (en) | 2021-11-16 | 2021-11-16 | Vein blood vessel identification positioning method and system based on near infrared imaging |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114176514B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116746926B (en) * | 2023-08-16 | 2023-11-10 | 深圳市益心达医学新技术有限公司 | Automatic blood sampling method, device, equipment and storage medium based on image recognition |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100920251B1 (en) * | 2008-12-30 | 2009-10-05 | 동국대학교 산학협력단 | A method for restoring infrared vein image blurred by skin scattering |
| CN102871645A (en) * | 2011-07-11 | 2013-01-16 | 浙江大学 | Near-infrared imaging ultrasonic vascular therapeutic apparatus |
| CN107748872A (en) * | 2017-10-27 | 2018-03-02 | 孙洪军 | A kind of IMAQ is clear and comprehensive intelligent palm vein identification device |
| CN107812283A (en) * | 2017-10-18 | 2018-03-20 | 北京工商大学 | A kind of method for automatically determining point of puncture position |
| CN109171905A (en) * | 2018-10-11 | 2019-01-11 | 青岛浦利医疗技术有限公司 | Guiding puncture equipment based on infrared imaging |
| CN112022346A (en) * | 2020-08-31 | 2020-12-04 | 同济大学 | Control method of full-automatic venipuncture recognition integrated robot |
| CN113303771A (en) * | 2021-07-30 | 2021-08-27 | 天津慧医谷科技有限公司 | Pulse acquisition point determining method and device and electronic equipment |
-
2021
- 2021-11-16 CN CN202111353446.9A patent/CN114176514B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100920251B1 (en) * | 2008-12-30 | 2009-10-05 | 동국대학교 산학협력단 | A method for restoring infrared vein image blurred by skin scattering |
| CN102871645A (en) * | 2011-07-11 | 2013-01-16 | 浙江大学 | Near-infrared imaging ultrasonic vascular therapeutic apparatus |
| CN107812283A (en) * | 2017-10-18 | 2018-03-20 | 北京工商大学 | A kind of method for automatically determining point of puncture position |
| CN107748872A (en) * | 2017-10-27 | 2018-03-02 | 孙洪军 | A kind of IMAQ is clear and comprehensive intelligent palm vein identification device |
| CN109171905A (en) * | 2018-10-11 | 2019-01-11 | 青岛浦利医疗技术有限公司 | Guiding puncture equipment based on infrared imaging |
| CN112022346A (en) * | 2020-08-31 | 2020-12-04 | 同济大学 | Control method of full-automatic venipuncture recognition integrated robot |
| CN113303771A (en) * | 2021-07-30 | 2021-08-27 | 天津慧医谷科技有限公司 | Pulse acquisition point determining method and device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114176514A (en) | 2022-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Adegun et al. | Deep learning-based system for automatic melanoma detection | |
| US7499576B2 (en) | Method and system for detecting a fiducial in digital projection images | |
| Jiang et al. | Skin lesion segmentation based on multi-scale attention convolutional neural network | |
| CN111243730B (en) | Mammary gland focus intelligent analysis method and system based on mammary gland ultrasonic image | |
| US20220207742A1 (en) | Image segmentation method, device, equipment and storage medium | |
| CN114176514B (en) | Vein blood vessel identification positioning method and system based on near infrared imaging | |
| CN109512464A (en) | A kind of disorder in screening and diagnostic system | |
| CN114494296A (en) | Brain glioma segmentation method and system based on fusion of Unet and Transformer | |
| Rajathi et al. | Varicose ulcer (C6) wound image tissue classification using multidimensional convolutional neural networks | |
| CN114066884B (en) | Retinal blood vessel segmentation method and device, electronic device and storage medium | |
| CN110751636A (en) | Fundus image retinal arteriosclerosis detection method based on improved coding and decoding network | |
| JP2023106142A (en) | Image processing apparatus and image processing method | |
| CN114511581B (en) | Multi-task multi-resolution collaborative esophageal cancer lesion segmentation method and device | |
| CN116503607B (en) | CT image segmentation method and system based on deep learning | |
| CN111681254A (en) | Catheter detection method and system for vascular aneurysm interventional operation navigation system | |
| CN115471470A (en) | Esophageal cancer CT image segmentation method | |
| CN114581474A (en) | Automatic clinical target area delineation method based on cervical cancer CT image | |
| KR20090067141A (en) | Pattern identifying method, registration device, collating device and program | |
| CN117036715A (en) | Deformation region boundary automatic extraction method based on convolutional neural network | |
| Tang et al. | MMMNA-net for overall survival time prediction of brain tumor patients | |
| Kang et al. | 3D active vessel tracking using an elliptical prior | |
| CN110443217A (en) | One kind being based on multispectral fingerprint method for anti-counterfeit and system | |
| CN117455779B (en) | Auxiliary enhancement system of medical ultrasonic imaging instrument | |
| Van Do et al. | Segmentation of hard exudate lesions in color fundus image using two-stage CNN-based methods | |
| CN114155195B (en) | Brain tumor segmentation quality evaluation method, device and medium based on deep learning |
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 |