Disclosure of Invention
The invention aims to provide a device for evaluating the state of a vein, and aims to solve the technical problems that the evaluation of the state of the vein in the prior art is basically based on the judgment of doctor experience, and the evaluation accuracy of the state of the vein is influenced due to the fact that a hemogram signal is weak and sensitive, muscle shakes and tenses, and the signal distortion of the hemogram is caused.
To achieve the above object, the present invention provides a device for evaluating the status of a vein, comprising:
acquiring venous blood flow graph signals S _ f1 and S _ f2 at f1 Hz and f2 Hz;
storing the obtained venous blood flow map signal S _ f1 after digital-to-analog conversion to obtain a venous blood flow map S1_ f 1;
performing Lagrange interpolation on the venous blood flow map S1_ f1 to obtain a venous blood flow map S2_ f1 with equal sampling intervals;
decomposing the venous blood flow map S2_ f1 to obtain a smooth venous blood flow map signal S3_ f1, a blood flow map signal S4_ f1 and a capillary impedance change rate b _ f 1;
after data processing is carried out on a blood flow graph signal S4_ f1, heart rate variability HRV _ f1 is obtained;
carrying out Pearson similarity calculation on the venous blood flow map signal S3_ f1 and the f1 Hertz effective blood flow map set to obtain a similarity value rho _ f 1;
when the heart rate variability HRV _ f1 is higher than the threshold value d1 and the similarity value rho _ f1 is higher than the threshold value d2, judging that the blood flow graph signal S3_ f1 is effective, otherwise, judging that the blood flow graph signal S3_ f1 is ineffective;
performing GAF transformation on the effective venous blood flow map signal S3_ f1 to obtain GASF _ f 1;
repeating the operation on the S _ f2 to obtain a smooth venous blood flow graph signal S3_ f2, a capillary impedance change rate b _ f2 and a GAF graph GASF _ f 2;
constructing a CNN neural network N1 model, and calculating probability P1 by adopting effective venous blood flow graph signals S3_ f1 and S3_ f2 for combined input;
constructing a CNN neural network N2 model with an attention mechanism, and calculating probability P2 by adopting combined input of GASF _ f1 and GASF _ f 2;
and (3) combining the parameters b _ f1, b _ f2, P1 and P2, constructing a logistic regression model, and obtaining the venous vascular state score.
Wherein, in the step of acquiring venous blood flow graph signals S _ f1 and S _ f2 at f1 Hz and f2 Hz, the device further comprises,
the f1 and the f2 are two frequencies with the magnitude difference level in a beta frequency dispersion section, the range of the beta frequency dispersion section is dozens of kilohertz to dozens of megahertz, and the study on the electrical characteristics of biological tissues in the frequency section can reflect the characteristics of extracellular fluid and intracellular fluid and show a plurality of physiological and pathological changes; the penetration degree of the excitation signals with different frequencies to the cell membrane is different, so that the obtained physiological information is also different; the starting point of the venous blood flow graph signal is the starting point of vein occlusion, and the ending point is the 10 th second after the cuff is deflated.
Wherein in the step of storing the obtained venous blood flow map signal S _ f1 after D/A conversion to obtain a venous blood flow map S1_ f1, the apparatus further comprises,
and sequentially carrying out digital-to-analog conversion, 50Hz trap and Kalman filtering on the venous blood flow graph signal S _ f1, and then transmitting and storing the venous blood flow graph signal into a database through an acquisition card.
Wherein in decomposing the venous blood flow map S2_ f1 to obtain a smoothed venous blood flow map signal S3_ f1, a blood flow map signal S4_ f1, a capillary impedance rate of change b _ f1, the apparatus further comprises,
low-pass and high-pass filtering are respectively carried out on the venous blood flow diagram S2 to obtain a venous blood flow diagram signal S3 'and a blood flow diagram signal S4, and then the venous blood flow diagram signal S3' is decomposed to obtain a smooth venous blood flow diagram signal S3 and a capillary impedance change rate b.
Wherein in the "re-resolving venous blood flow map signal S3', the apparatus further comprises,
and extracting a starting point of ascending branch and an inflection point of ending descending branch from the venous blood flow diagram signal S3 ', correcting a capillary liquid balance curve of the venous blood flow diagram signal S3' through the two points to obtain a smooth venous blood flow diagram signal S3, and obtaining a capillary impedance change rate b by utilizing coordinates of the two points.
Wherein in the capillary liquid balance curve correction of venous flowsheet signal S3' through the two points, the apparatus further comprises,
and establishing a linear equation by using two-point coordinates, constructing a curve signal by using the linear equation, carrying out difference on the venous blood flow diagram signal S3 ', and eliminating a capillary liquid self-balancing signal to obtain a smooth venous blood flow diagram signal S3'.
Wherein, in the step of obtaining the heart rate variability HRV _ f1 after data processing the blood flow graph signal S4_ f1, the device further comprises,
and (3) extracting minimum values of the blood flow graph signals S4, calculating the time difference between each minimum value point, namely the R-R instantaneous period of the heartbeat, calculating the standard deviation as the HRV (hyper-short-term heart rate variability), and judging whether the testee is relaxed and the emotion is calm.
Wherein, in the process of carrying out Pearson similarity calculation on the venous blood flow diagram signals S3_ f1 and the f1 Hertz effective blood flow diagram set to obtain a similarity value rho _ f1, the device also comprises,
the effective blood flow atlas is judged to be effective by a doctor and comprises several blood flow charts of normal people and patients.
According to the device for evaluating the venous vessel state, the problems of signal mutation, unequal sampling intervals and the like caused by data loss in the transmission process are effectively corrected through the Lagrange interpolation method, and the later filtering effect is further improved; through capillary drift correction, make the blood flow graph wave form more press close to the theory, go up and down a rate of change and judge more accurately. The formation of a large amount of venous shunt may affect the capillary filtration fluid balance characteristic, and the rate of change of the capillary impedance, which characterizes the capillary filtration fluid balance characteristic, may provide a basis for the judgment of patients with venous obstruction accompanied by a large amount of venous shunt or malformed blood vessels. The signal quality of the venous blood flow graph is poor due to the fact that the signal quality of the venous blood flow graph is poor and the characteristics of the venous blood flow graph cannot be effectively judged due to the fact that muscles of a tested person are tensed, the emotion of the tested person is not stable, the waveform similarity is calculated and compared with the similarity between the venous blood flow graph and an effective blood flow graph to judge whether the waveform quality of the blood flow graph is too high, and the accuracy of judging the effectiveness of the blood flow graph can be improved by combining the venous blood flow graph and the effective blood flow graph; the single frequency venous blood flow map signal can only reflect the information of the venous obstruction method under certain degree of penetration of the cell membrane, the venous blood flow map signal is combined with S3_ f1 and S3_ f2, and the spectrogram is combined with S6_ f1 and S6_ f2, and different degrees of penetration of the cell membrane is stimulated by different frequencies to obtain more physiological information; the machine vision CNN network has strong feature extraction capability, and attention mechanism pays attention to important information of the network, so that venous blood flow graph signals can be analyzed from higher dimensionality, and more information can be acquired. Through the logistic regression model, the information contained in the two CNN models and the capillary liquid self-balance is comprehensively fused, the blood vessel occlusion degree is better judged, and the model accuracy is improved.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
In the description of the present invention, it is to be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on the orientations or positional relationships illustrated in the drawings, and are used merely for convenience in describing the present invention and for simplicity in description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed in a particular orientation, and be operated, and thus, are not to be construed as limiting the present invention. In addition, in the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
Referring to fig. 1 to 3, the present invention provides an apparatus for evaluating the status of a vein, comprising:
s101, obtaining venous blood flow graph signals S _ f1 and S _ f2 at f1 Hz and f2 Hz;
the starting point of the venous blood flow graph signal is the starting point of an occlusion method, and the ending point is the 10 th second after release. The f1 and f2 excitation frequency bands need to be in a beta section and have order of magnitude difference, corresponding to different penetration degrees of the excitation on cell membranes, fluctuation rules of corresponding blood flow change in total impedance change are also different, and more physiological information is provided.
S102, storing the obtained venous blood flow diagram signal S _ f1 after digital-to-analog conversion to obtain a venous blood flow diagram S1_ f 1;
specifically, the signals are subjected to digital-to-analog conversion, 50Hz trap and Kalman filtering in sequence, and then are transmitted and stored in a database through an acquisition card. Due to the fact that power frequency signals in the circuit, noise such as external electromagnetic signals can interfere collection of the venous blood flow graph, power frequency interference can be eliminated through 50Hz trapped waves, external noise interference is weakened in a self-adaptive mode through Kalman filtering, and the signal-to-noise ratio of the venous blood flow graph is improved.
S103, carrying out Lagrange interpolation on the venous blood flow graph S1_ f1 to obtain a venous blood flow graph S2_ f1 with equal sampling intervals;
specifically, due to the fact that packet loss occurs in the process of transmitting the venous blood flow graph data, the problems of data loss, signal mutation, unequal sampling intervals and the like are caused, extraction of characteristic points is affected, and meanwhile the later filtering effect is weakened, therefore lagrangian interpolation supplement is conducted on the data, missing points are filled, and the filtering effect is improved. Lagrange interpolation formula, p (x) is the corresponding interpolated value:
s104, decomposing a venous blood flow diagram S2_ f1, and obtaining a smooth venous blood flow diagram signal S3_ f1, a blood flow diagram signal S4_ f1 and a capillary drift impedance change rate b _ f 1;
specifically, two digital filters of low-pass and high-pass are provided to filter the venous blood flow map S2_ f1 to obtain venous blood flow map signals S3' _ f1 and blood flow map signals S4_ f 1. Because the venous blood flow map generally contains pulse wave signals, i.e. blood flow map signals in the general sense, the bandwidth is 0.05-100Hz, and the venous blood flow map signals S3' _ f1 and the blood flow map signals S4_ f1 are decomposed by using the bandwidth. If the venous blood flow map is already filtered of the pulse wave signal during the acquisition process, this step can be omitted and the heart rate variability cannot be obtained.
Specifically, the specific steps of obtaining the smoothed venous blood flow map signal S3_ f1 and the capillary drift impedance change rate b _ f1 include:
s201, extracting a branch ascending starting point q1 and a branch descending ending point q2 in a venous blood flow graph signal S3' _ f 1;
taking the signal starting point as a ascending branch starting point q1 and the coordinate as (x1, y1), performing inflection point search on the venous blood flow graph signal S3' _ f1, and finding the last inflection point as a descending branch end point q2 and the coordinate as (x2, y 2).
S202 constructs the linear equation f (x) using the q1, q2 coordinates:
obtaining the rate of change of capillary impedance
S203 reconstructs the drift signal using the linear equation f (x), and the coordinate axis x, i.e., the time axis.
S204 obtains a smoothed venous blood flow map signal S3_ f1 by subtracting the venous blood flow map signals S3' _ f1 and canceling the excursion signal.
S105, obtaining heart rate variability HRV _ f1 after data processing on a blood flow graph signal S4_ f 1;
specifically, minimum value extraction is carried out on a blood flow graph signal, time difference between each minimum value point and standard deviation of the time differences are calculated, the standard deviation is taken as ultra-short-time heart rate variability, and whether a tester is relaxed or not and the mood is calm is judged. High heart rate variability, meaning that the more fully prepared, calmer and less stressed the unknown challenge; low heart rate variability means stress.
S106, carrying out Pearson similarity calculation on the venous blood flow map signal S3_ f1 and the f1 Hertz effective blood flow atlas to obtain a similarity value rho _ f 1;
specifically, 1000 points are extracted at equal intervals in S3_ f1 to form a signal X, the signal in the effective blood flow atlas is Yi, each Yi is composed of 1000 points, Pearson correlation coefficient ρ i is calculated for the signals X and Yi, and the minimum value is taken as the similarity ρ _ f 1.
S107, when the heart rate variability HRV _ f1 is higher than the threshold d1 and the similarity is higher than the threshold d2, judging that the blood flow graph signal S3_ f1 is effective, otherwise, judging that the blood flow graph signal S3_ f1 is ineffective;
s108, repeating the operations from S102 to S108 on the S _ f2 to obtain a smooth venous blood flow map signal S3_ f2, a capillary drift impedance change rate b _ f2 and a GAF map GASF _ f 2;
s109, GAF conversion is carried out on the effective venous blood flow map signal S3_ f1 to obtain GASF _ f 1;
specifically, assuming that S3_ f1 is composed of n points, the signal S3_ f1 is scaled to [ -1, 1], and the normalized value is replaced with a cosine function, thereby obtaining the GASF.
S110, constructing a CNN model, and converting an effective venous blood flow graph into probability P1 by using two inputs;
specifically, S3_ f1 and S3_ f2 are input as input data into the CNN model to obtain the probability P1. The CNN model is composed of two channels and a full connection layer with an activation function of Sigmoid, wherein each channel comprises: a feature extraction layer and a connection layer. And the feature extraction layer is formed by connecting two feature extraction submodules in series. The feature extraction submodule is composed of two Conv2D branch activation functions of 'Relu', a merging layer composed of a concatenate layer and Conv2D (the activation function is 'Relu') for merging two branches, a Max Pooling layer, a GlobalAveragePooling2D layer, two fully-connected layer activation functions of 'Relu' and 'sigmoid', a RepeatVector layer, a Reshape layer and a multiplex layer; the connection layer is composed of a Flatten layer, two full connection layers consisting of a Dense activation function of 'Relu' and Dropout, and a full connection layer activation function of 'Sigmoid'.
S111, constructing a CNN model with an attention mechanism, and converting the GASF graph into probability P2;
specifically, GASF _ f1 and GASF _ f2 are input as input data to the CNN to obtain the probability P2. The structure of the CNN model is similar to S110, but a channel attention module is added to each channel to re-assign weights to features and focus on important information.
S112, constructing a logistic regression model by combining parameters b _ f1, b _ f2, P1 and P2, and obtaining a venous vessel state score;
specifically, the b _ f1, b _ f2, P1 and P2 are used for obtaining the vascular state score through a logistic regression model consisting of two connected layers with the activation function of Relu and one connected layer with the activation function of Sigmoid.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.