TcPO2 curve trend analysis method based on k-cosine corner detection

Abstract

The invention is suitable for the field of biomedical signal processing, and provides a TcPO2 curve trend analysis method based on k-cosine corner detection, and the method comprises the following steps: collecting and recording TcPO2 data in real time, constructing a chain curve structure, and preprocessing the obtained data; based on a k-cosine method, all angular points in the curve are screened out; the difference slope and neighborhood characteristics of the curve are analyzed, and the curve is divided into a plurality of straight line sections and curve sections based on neighborhood corner point intensity saliency and a corner point intensity threshold value; identifying the trend type of each curve section based on a trend type identification algorithm of a curve section and a straight line section; according to the method, a more visual and reliable signal analysis tool can be provided for medical staff, and the accuracy of clinical diagnosis and intervention is improved.

Description

Technical Field

[0001] The invention belongs to the field of biomedical signal processing, and in particular relates to a TcPO2 curve trend analysis method based on k-cosine corner point detection. Background Art

[0002] Transcutaneous oxygen partial pressure (TcPO2) monitoring technology is an important non-invasive monitoring method used to measure the oxygen partial pressure of tissues beneath the skin surface. This technology uses a small sensor attached to the skin to measure the oxygen partial pressure of tissues beneath the skin through heating and electrochemical reactions, thereby assessing the oxygenation status of tissues.

[0003] TcPO2 monitoring is widely used in clinical practice to assess the patient's oxygenation status, especially in intensive care, surgery and chronic disease management. It is of great significance for judging the patient's oxygenation level and tissue oxygen supply status.

[0004] Although TcPO2 technology provides important physiological data, trend analysis and feature extraction of its data still face challenges, especially in dynamic monitoring and real-time analysis.

[0005] In the application field of interactive curve modeling, the problem of curve segmentation based on trend features has always been a hot topic of research. Corner points on digital plane curves are robust features that provide important clues for shape representation and trend description. If the corner points contained in a curve can be correctly identified, the curve can be represented in an efficient and compact way with sufficient accuracy.

[0006] There are many methods for detecting corners and inflection points of curves in two-dimensional space. For example: Patent CN202010619850.5 discloses a contour-based corner detection algorithm. This method is applied to the field of image processing, and detects corners by eight-neighborhood contour tracking; Patent CN202410272676.X is an image corner detection method, which first extracts the edge map through the Canny detector, and then uses the point curvature on the edge to extract the corners in the image; Patent CN202010933216.9 is a method for determining the inflection point of the press-fitting curve, which belongs to the field of press-fitting detection, and the inflection point is judged by setting the number of continuous points Num and the slope threshold k judged by the slope threshold; Patent CN201010191596.X also belongs to the image feature extraction algorithm, which detects the inflection point by encoding the curve with Freeman chain code and calculating the difference cumulative value point by point.

[0007] However, if the above methods are applied to extract feature points of TcPO2 curve, they have their own shortcomings:

[0008] 1. The contour-based corner detection algorithm of CN202010933216.9 is designed for spatial contours in two-dimensional images, while the TcPO2 curve is a time series curve. Directly using the eight-neighborhood method will cause a mismatch between space and data types, which may lead to a decrease in corner detection accuracy.

[0009] 2. The detection method of CN202410272676.X has a strong dependence on curvature calculation. Therefore, if the edge detection effect is not good, it will directly affect the detection and positioning accuracy of subsequent corner points.

[0010] 3. The method of CN202010933216.9 mainly identifies the inflection point of the curve based on the obvious degree of slope change. If the slope threshold k and the number of consecutive points Num are set improperly, it may lead to missed detection or over-detection. Therefore, the parameters need to be adjusted carefully, and it may still not be ideal.

[0011] 4. Although the method of CN201010191596.X has certain advantages in capturing local changes in curves, Freeman chain code is usually used for contours or lines in two-dimensional images, which may increase the computational complexity for single-variable time series such as TcPO2.

[0012] In addition, none of the above methods take into account the problem of the concavity of the curve, while TcPO2 must analyze its concavity to help distinguish the nature of the upward or downward trend of oxygen partial pressure.

[0013] On the other hand, the identification of stable, rising and falling sections in the TcPO2 curve helps to assess the patient's oxygenation status, monitor treatment effects and predict potential health risks. This link occupies the same important position as corner point detection in the trend analysis of the TcPO2 curve.

[0014] Currently, in the medical field, especially for the trend analysis of TcPO2 curves, there is no systematic analysis method that combines several geometric features such as corners, concavity, slope, stable trend, rising trend and falling trend. Existing research mainly focuses on the analysis of a single or a few geometric features, and is mostly used in the field of computer vision or image processing, while the multi-feature comprehensive analysis method for physiological signals has not been widely used, which leads to the lack of accurate and comprehensive trend interpretation tools in clinical applications. Summary of the invention

[0015] The purpose of the present invention is to provide a TcPO2 curve trend analysis method based on k-cosine corner point detection, combining the geometric features of the corner points, the concavity and convexity of the corner point neighborhood, the slope, the stable section, the rising section and the falling section of the curve, so as to help medical workers and scientific researchers quickly and accurately segment the complex TcPO2 curve, understand the behavioral characteristics of the curve in different time periods, reveal the changes in physiological state, and ultimately provide more powerful support for clinical decision-making and patient monitoring.

[0016] The present invention is implemented as follows: a TcPO2 curve trend analysis method based on k-cosine corner point detection, the method comprising:

[0017] Collect and record TcPO2 data in real time, build a chain curve structure, and pre-process the acquired data;

[0018] The corner point strength (or average corner point strength) and the corner point strength index (or average corner point strength index) of each point on the curve are calculated based on the k-cosine method, and all corner points in the curve are identified based on the corner point strength (or average corner point strength), the corner point strength index (or average corner point strength index) and the corner point strength radius defined thereby;

[0019] Analyze the neighborhood features of each point on the TcPO2 curve with respect to corner points, and segment the TcPO2 curve into several straight line segments and curve segments based on the neighborhood corner point intensity significance and corner point intensity threshold;

[0020] The trend type discrimination algorithm based on curve segments and straight line segments identifies the trend type of each segment of the TcPO2 curve;

[0021] The accuracy of this method in TcPO2 curve trend analysis was evaluated through quantitative analysis experiments.

[0022] As a further solution of the present invention, the real-time collection and recording of TcPO2 data, the construction of a chain curve structure, and the pre-processing of the acquired data specifically include:

[0023] A plane rectangular coordinate system is established with the acquisition time of TcPO2 value as the horizontal axis and the TcPO2 value as the vertical axis;

[0024] Record the TcPO2 value and the corresponding UNIX timestamp at uniform time intervals, and record the original UNIX timestamp as In seconds, the corresponding original TcPO2 value is recorded as The unit is mmHg, and the sample size of each data collection is recorded as n (n≥30);

[0025] Discrete digital point array Perform time-series connection to form a chain curve structure in the plane rectangular coordinate system

[0026] Using the min-max normalization method Scale to the range of 0 to 1;

[0027] The structure of the standardized TcPO2 chain curve is: i (x i ,y i )|i=1,…,n;x i ,y i ∈[0,1]}, which can be simply written as {p i (x i ,y i )} or {p i};

[0028] The Savitzky-Golay filter is used to smooth the normalized chain digital curve point sequence.

[0029] As a further solution of the present invention, the corner point strength and corner point strength index of each point on the curve are calculated based on the k-cosine method, and all corner points in the curve are screened out based on the corner point strength, corner point strength index and the corner point strength radius defined thereby, specifically including:

[0030] Use k-cosine corner detection algorithm to convert {p i (x i ,x i )} for every point p i Several predecessor points or successor points of p i The vector formed is defined as p i arm, predecessor or successor point to p i The number of intervals is defined as p i arm length;

[0031] Get with p i The pair of points whose arm lengths are all k <p i-k ,p i+k >, defined as p i The two arms of Defined as p i Forearm, vector Defined as p i The rear arm;

[0032] Use your forearms and rear arm Define p i The k-vector:

[0033]

[0034] The corresponding k-cosine is defined as:

[0035]

[0036] Where -1≤c ik ≤1;

[0037] The longest arm length to be considered is defined as p i The k-radius is denoted as m (m>0);

[0038] Calculate c based on m i1 ,c i2 ,…,c i,m , then define c i0 = -1, c i0 ,c i1 ,c i2 ,…,c i,m It is denoted as k-cosine sequence with radius m;

[0039] According to the corner point strength determination theorem, the curve {p i}P i Corner strength and the corner point strength index value j;

[0040] According to p i The j value is used to calculate the corner strength radius h i , δ(p i ,h i ) is denoted as p′ k (|ki|≤h i ), calculate δ(p i ,h i ) for every point p′ k of

[0041] According to the corner point determination theorem, p i Is it a corner point? If so, set p i Add corner point array {q l (x l ,y l )|l=1,…,n′}(n′ is the number of corner points detected), the corner point sequence can also be abbreviated as {q l (x l ,y l )} or {q l};

[0042] Repeat the above three steps until {p i All the points in} have been determined.

[0043] As a further solution of the present invention, in actual work, since the sampling points of the TcPO2 curve are too dense or too sparse, the detection results of the above method will be incorrect. Therefore, the improved average k-cosine method is used to calculate the average corner point intensity and average corner point intensity index of each point on the curve, and all corner points in the curve are screened out based on the average corner point intensity, the average corner point intensity index and the corner point intensity radius defined thereby, specifically including:

[0044] Define point p i The neighborhood radius r i as follows:

[0045]

[0046] p i The average k-cosine It is calculated in the following way:

[0047]

[0048] Then p i of Treated as c ik , under the condition that the k-radius is m, we can calculate According to the corner point strength determination theorem, p i The average corner strength and the average corner intensity index j;

[0049] use replace Similarly, the corner point determination theorem is used to determine the curve {p i} all corner points.

[0050] As a further solution of the present invention, the corner point strength determination theorem includes:

[0051] In k-p with radius m i The k-cosine sequence c i0 ,c i1 ,c i2 ,…,c im , from the rightmost c im The last one counting from the left is larger than the c of its neighbor on the right ij Determined as p i The corner strength of m is denoted as

[0052] The corner point determination theorem includes:

[0053] Suppose the curve is about p i (x i ,yi ) has a neighborhood of δ(p i ,h i ), h i For p i The corner strength radius of (|ki|≤h i )Hengyou but is a local maximum, at which point p i is a corner point of the curve.

[0054] As a further solution of the present invention, the curve segment recognition algorithm is used to analyze the neighborhood characteristics of each point on the TcPO2 curve with respect to the corner point, and the curve {p i (x i ,y i )} is divided into a number of straight line segments and curve segments, and the specific steps include:

[0055] The curve {p i (x i ,y i )} is recorded as a union form The union of all curve segments is denoted by C p , the union of all straight line segments is denoted as L p ,but

[0056] For the above corner point sequence {q l (x l ,y l )}, define the corner point intensity threshold T c , the corner point q l The strength of nearby corner points is greater than or equal to T c The neighborhood of is defined as the curve segment far away from the corner point and where the strength of the corner point is less than T c The area of ​​is defined as a straight line segment;

[0057] Set the corner point q l For {p i (x i ,y i )} i ,q l About r neighborhoodδ(q l ,r) is defined as:

[0058]

[0059] For {q l Each point q in l , calculate q l About δ(q l,1),δ(q l ,2), the mean value of the corner intensity of…

[0060] Through the infinite sequence of the above r neighborhood-corner point intensity mean pairs Get The corresponding δ(q l ,r l ), and δ(q l ,r l ) is set as the corner point q l The corresponding curve segment;

[0061] Repeat the above two steps until {q l}, and the corresponding curve segment set {δ(q l ,r l )|l=1,…,n′} is recorded as a union:

[0062] Find the set of all straight line segments (the complement of the curve segment set) and record them in the form of union as follows:

[0063]

[0064] As a further solution of the present invention, a trend type discrimination algorithm based on a curve segment identifies a curve {p i (x i ,y i )}, the specific steps include:

[0065] Define a point p on the curve i The slope of the s-order difference is (y i+s -y i ) / (x i+s -x i ), where s is the difference step size, and s must be greater than 1;

[0066] The present invention sets the corner point strength radius h i As the difference step size s, that is, p i The differential slope formula is:

[0067]

[0068] The curve segments can be divided into two types according to the trend type: upward concave segments (CTT_Convexity) and downward concave segments (CTT_Concavity);

[0069] For the curve segment set C p Each segment δ(q l ,r l), respectively calculate and obtain the difference slope mean sequence on the left: The difference slope mean series on the right:

[0070] Calculate δ(q l ,r l )The mean of the difference slopes between the left and right parts and δ(q l ,r l ) is written as the curve point sequence The corresponding difference slope is recorded in the form of a sequence: Similarly, δ(q l ,r l ) The difference slope sequence of the right half is recorded as: but:

[0071]

[0072] Set the judgment conditions for the upward concave section and the downward concave section:

[0073]

[0074] According to the above judgment conditions, the segment δ(q l ,r l )’s trend type;

[0075] Repeat the above process to derive C p Trend type for all curve segments above.

[0076] As a further solution of the present invention, a trend discrimination algorithm of a straight line segment is used to determine the difference slope of each point of the TcPO2 curve and the difference slope threshold T st Identify the curve {p i (x i ,y i )}The trend type of each straight line segment, the specific steps include:

[0077] The straight line segments can be divided into three types according to the trend type: stable segment (CTT_Stable), rising segment (CTT_Upward) and falling segment (CTT_Downward);

[0078] Set the judgment conditions for the stable section, rising section, and falling section:

[0079]

[0080] For the straight line segment L p Every point p in i , first calculate p iκ(h i ), and then derive p according to the judgment conditions of the stable section, rising section, and falling section i Trend type;

[0081] Connect all adjacent points that belong to the stable segment, and get a total of n1″ stable segments, denoted as

[0082] Connect all adjacent points that belong to the ascending segment, and get a total of n′2′ ascending segments, denoted as

[0083] Connect all adjacent points that belong to the descending segment, and get a total of n3″ ascending segments, denoted as

[0084] The k-cosine corner point detection algorithm or the improved k-cosine corner point detection algorithm, the curve segment recognition algorithm, the curve segment trend type discrimination algorithm, and the straight line segment trend type discrimination algorithm are encapsulated into filters and connected in the form of an algorithm pipeline to obtain an algorithm for comprehensive trend analysis of the TcPO2 curve, which is called the k-cosine and differential slope collaborative discrimination algorithm.

[0085] The beneficial effects of the present invention are:

[0086] (1) The beneficial effects of the k-cosine corner detection algorithm include: being able to accurately identify corner points in the TcPO2 curve, reflecting the key change locations of the signal, helping to extract important features of the physiological signal, and contributing to a deeper understanding of the physiological state.

[0087] (2) The beneficial effects of the improved k-cosine corner detection algorithm include: enhanced robustness to noise and abnormal data, improved accuracy and reliability of corner detection, and more practicality in clinical applications.

[0088] (3) The beneficial effects of the curve segment recognition algorithm include: being able to effectively segment the TcPO2 curve into curve segments and straight line segments, laying the foundation for further trend analysis and feature extraction.

[0089] (4) The beneficial effects of the trend type discrimination algorithm of the curve segment include: being able to reveal the changing pattern of the data, helping doctors identify different physiological states, and providing support for clinical decision-making.

[0090] (5) The trend type discrimination algorithm of the straight line segment focuses on the analysis of the straight line segment. Its beneficial effects include: being able to identify the stable period or consistent change of the signal, which helps to judge the normal range of the physiological signal and its deviation.

[0091] (6) The k-cosine and differential slope collaborative discrimination algorithm encapsulates the k-cosine corner detection algorithm with the curve segment recognition and trend type discrimination algorithm into a filter and connects them in the form of an algorithm pipeline to achieve a comprehensive analysis of the TcPO2 curve.

[0092] Its main beneficial effects include:

[0093] 1) Enhanced trend identification capability: The algorithm can more accurately identify key change points and trend types in the curve, and through the synergy of different algorithms, it improves the ability to capture complex data patterns.

[0094] 2) Improve analysis efficiency: By integrating multiple algorithms into one, the data processing process is simplified, making trend analysis more efficient and able to quickly respond to changes in physiological signals.

[0095] 3) Enhanced robustness: The collaborative discrimination algorithm shows stronger robustness in dealing with noise and abnormal data, reducing the risk of misjudgment and missed judgment, and ensuring the reliability of the analysis results.

[0096] 4) Clinical application value: It provides medical staff with a more intuitive and reliable signal analysis tool to help them more effectively monitor the physiological status of patients, thereby improving the accuracy of clinical diagnosis and intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 For the upward trend point column {p i (1)} curve graph;

[0098] Figure 2 For the downward trend point column {p i (2)} curve graph;

[0099] Figure 3 The trend point sequence is first rising and then falling. i (3)} curve graph;

[0100] Figure 4 It is a curve diagram of the TcPO2 chain curve point series (the position and strength of all corner points and inflection points on the curve are marked in the figure);

[0101] Figure 5 It is the morphological characteristic diagram of 5 types of curve sections such as concave upward;

[0102] Figure 6 It is a flow chart of the TcPO2 curve trend analysis method based on k-cosine corner point detection;

[0103] Figure 7This is the flow chart of the k-cosine corner detection algorithm;

[0104] Figure 8 Flowchart of the improved k-cosine corner detection algorithm;

[0105] Fig. 9 is a flow chart of the curve segment recognition algorithm;

[0106] Fig.10 It is a flow chart of the algorithm for distinguishing the trend type of the curve segment;

[0107] Fig.11 It is a flow chart of the algorithm for distinguishing the trend type of straight line segments;

[0108] Fig.12 Flowchart of the k-cosine and differential slope co-discrimination algorithm. DETAILED DESCRIPTION

[0109] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0110] like Figure 6 As shown, a TcPO2 curve trend analysis method based on k-cosine corner point detection, the method comprising:

[0111] S1. Collect and record TcPO2 data in real time, construct a chain curve structure, and pre-process the acquired data.

[0112] One of the most common methods for detecting corner points of a TcPO2 curve is to use the second derivative (curvature) of the curve. Curvature is defined as the rate at which the slope changes with the arc length. For a curve y=f(x), it can be expressed as:

[0113]

[0114] Although the concept of curvature of a continuous curve is well defined in differential calculus and can be expressed by an exact analytical formula, its digital representation (such as a chain curve) lacks a generally accepted definition. Therefore, it is not immediately clear how to define the discrete analog of curvature.

[0115] In order to circumvent the problems in digital curvature estimation, the present invention adopts a method that is different from the analytical method but conceptually related. A prerequisite for the effective use of this method is to establish a chain structure that can store digital information and to reasonably pre-process the acquired data.

[0116] S2. Based on the k-cosine method, the corner point strength (or average corner point strength) and corner point strength index (or average corner point strength index) of each point on the curve are calculated, and all corner points in the curve are identified based on the corner point strength (or average corner point strength), corner point strength index (or average corner point strength index) and the corner point strength radius defined thereby.

[0117] The k-cosine corner detection algorithm is an algorithm that uses cosine similarity as a smoothness metric to detect corners in a curve. The core idea is to determine the location of the corner by calculating the local gradient vector of each point on the curve and using the cosine of the vector angle to evaluate the degree and direction of the gradient change.

[0118] S3. Analyze the neighborhood features of each point on the TcPO2 curve with respect to corner points, and segment the TcPO2 curve into several straight line segments and curve segments based on the neighborhood corner point intensity significance and corner point intensity threshold.

[0119] The k-cosine technology can be used to detect the corner points (peaks and troughs) of the TcPO2 curve. These feature points can be used to segment the curve into different "support areas" that depend on the local curve characteristics. These support areas represent the change trend and shape characteristics of the TcPO2 curve in different time intervals, such as: a raised rising segment, a sunken falling segment, and a gentle stable segment. By identifying these areas, clear boundaries can be provided for subsequent feature extraction, pattern recognition, and trend analysis. This segmentation method has important application value in physiological signal analysis, and helps to deeply understand the changing characteristics of the signal from two dimensions: overall trend and local characteristics.

[0120] S4. The trend type discrimination algorithm based on the curve segments and the straight line segments identifies the trend type of each segment of the TcPO2 curve.

[0121] Based on the k-cosine method, the differential slope of each point and the mean of the differential slope of the left and right parts are combined. and Differential slope threshold T for stable region determination st Parameters such as PO2 and TP2 can be used to express each section of the TcPO2 curve in a more refined way.

[0122] S5. The accuracy of this method in TcPO2 curve trend analysis was evaluated through quantitative analysis experiments.

[0123] Experimental design was used to evaluate the accuracy of the k-cosine and differential slope co-discrimination algorithm in TcPO2 curve trend analysis, aiming to verify the effectiveness of the algorithm in curve feature point detection and trend segmentation.

[0124] In S1:

[0125] S1.1. The present invention establishes a plane rectangular coordinate system with the collection time of the TcPO2 value (oxygen partial pressure value) as the horizontal axis and the TcPO2 value as the vertical axis to draw the TcPO2 curve.

[0126] After establishing the plane rectangular coordinate system, it is necessary to collect TcPO2 data. The principle of data collection is to ensure that the TcPO2 value and the corresponding time value are recorded at uniform time intervals under specific experimental conditions. These data will provide the necessary input for subsequent analysis.

[0127] Assume that the sample size of each data collection is n (n ≥ 30), and the collection time is recorded in UNIX timestamp (in seconds), then the original UNIX timestamp can be recorded as The corresponding original TcPO2 value (in mmHg) is recorded as

[0128] The above discrete digital point Connect in time sequence to form a chain curve structure The chain curve can not only effectively store and represent the complex discrete characteristics of the TcPO2 curve, such as sharp ups and downs, tiny fluctuation details, etc., but also accurately capture and reproduce these key information through its unique piecewise linear approximation characteristics.

[0129] S1.2. The present invention uses the minimum-maximum normalization method to Scaled to the range of 0 to 1. The minimum and maximum values ​​of the UNIX timestamp and the original TcPO2 are calculated as follows:

[0130]

[0131] Then use the following formula to convert each point on the original curve Normalize to the range [0,1]:

[0132]

[0133] Finally, the standardized TcPO2 chain curve structure is: i (x i ,y i )|i=1,…,n;x i ,y i ∈[0,1]}.

[0134] S1.3. In order to reduce the impact of noise on the analysis results, the standardized chain digital curve point sequence can be smoothed and filtered.

[0135] The present invention uses a Savitzky-Golay filter, which is essentially equivalent to a low-pass filter and can retain high-frequency details, thereby reducing the interference of high-frequency noise on the signal.

[0136] Example 1:

[0137] The following is an example of the application of Savitzky-Golay filter. Assume that a certain TcPO2 curve {p i (x i ,y i )|i=1,…,n} i The sequence of values ​​is:

[0138] [0.12,0.15,0.13,0.19,0.21,0.18,0.23,0.25,0.24,0.30,0.28,0.26,0.31,0.34,0.33]

[0139] The steps of filtering it using a second-order Savitzky-Golay filter are as follows:

[0140] Step 1: Given a window size W = 5 and a polynomial order d = 2, the coefficient vector of the Savitzky-Golay filter is obtained by the least squares method as follows:

[0141]

[0142] Step 2: Smoothing filtering is performed using a sliding window. For each data point y i , use the left and right points for smoothing, let V sg The component indices of are {-2,-1,0,1,2}, then the smoothing value The calculation formula is as follows:

[0143]

[0144] Applying this formula to the sliding window of each data point yields the smoothed TcPO2 curve.

[0145] Practice has shown that when the data acquisition of the TcPO2 curve is relatively stable, adding a smoothing filter does not have a significant effect on improving the accuracy of the algorithm. Therefore, this step is not a necessary preprocessing step and can sometimes be omitted in specific work.

[0146] In S2:

[0147] S2.1. Definition of k-cosine:

[0148] The k-cosine corner detection algorithm converts {pi (x i ,y i )|i=1,…,n} for each point p i Several predecessors and successors of p i The vector formed is regarded as p i The "arm", the predecessor or successor point to p i The number of intervals is called p i The arm length will be equal to p i The pair of points whose arm lengths are all k <p i-k ,p i+k > as p i The two "arms" (k>0), vector called p i The "forearm", vector called p i The "hind arm".

[0149] Use your forearms and rear arm We can define p i The k-vector is as follows:

[0150]

[0151] The corresponding k-cosine is defined as:

[0152]

[0153] Where -1≤c ik ≤1. If a ik and b ik The angle is close to 0°, then c ik Close to 1; if a ik and b ik The angle is close to 180°, then c ik is close to -1. In other words, when the curve turns quickly, c ik Larger, when the curve is relatively straight, c ik Smaller.

[0154] The k-cosine corner detection algorithm considers every point on the curve as a potential corner point, and needs to examine the k-cosine values ​​of several arm lengths to determine the possibility of the point being a corner point (referred to as the corner point strength in this article). The maximum arm length that needs to be examined is called p i The k-radius is denoted as m (m>0). For example, when m=3, then p i It is necessary to calculate 1-cosine, 2-cosine, and 3-cosine to form a k-cosine sequence, and obtain p through this sequence i Further knowledge of angularity.

[0155] S2.2. Corner point strength determination theorem:

[0156] Based on a priori knowledge of the known TcPO2 curve, the k-radius can usually be set between 10 and 15.

[0157] For example, if m = 10, then find p i The steps of the k-cosine sequence are as follows:

[0158] Step 1: Use formula (6) to calculate c i1 ,…,c i,10 ;

[0159] Step 2: Define c i0 = -1;

[0160] Step 3: Combine the above 11 k-cosines into a sequence c i1 ,…,c i,10 .

[0161] That is, k-p with radius m i The k-cosine sequence is:

[0162] c i0 ,c i1 ,c i2 ,…,c im (7)

[0163] Theorem 1. Corner point strength determination theorem: In the case of k-radius m, i The k-cosine sequence c i1 ,…,c im From the rightmost c im The last one counting from the left is larger than the c of its neighbor on the right ij It was judged to be p i The corner strength of m is denoted as Right now:

[0164]

[0165] Here c i0 = -1 ensures that no matter what value m takes, must exist, that is, index j must exist and 1≤j≤m. Here, index j is called the corner strength index.

[0166] Theorem 1 can be used to determine the corner strength of each point on the TcPO2 curve with respect to m. The following example illustrates this:

[0167] Example 2:

[0168] Assume m = 10, and construct the curve point series of the rising trend, the falling trend, and the rising trend followed by the falling trend respectively. As shown in the following table:

[0169] Table 1 Curve points of rising trend, falling trend, and rising then falling trend

[0170]

[0171] From the above table, we can see that x in i is the standardized time value, which is equally spaced from 0 to 1. i It is the standardized TcPO2 value, and its value range is between [0,1].

[0172] Figure 1 A curve that reflects an upward trend The corresponding curve graph shows The y in i Follow x i increases, gradually changing from 0.1 to 1.0. And y i The growth rate of x i The uniform increase gradually increases.

[0173] Figure 2 It is a descending trend curve point series An example diagram of y i Follow x i It increases and decreases, the value gradually decreases from 1.0 to 0.1, and the decreasing rate gradually increases.

[0174] Figure 3 Shows a trend point sequence that first rises and then falls At the beginning, y i =0.0, with x i Increases and gradually increases, when x i = 0.5, reaching a peak value of 1.0; when x i >5, y i Follow x i increases and gradually decreases. When x i =1.0 i =0.0.

[0175] Consider the three curve points listed above at x 10 =0.50 corresponding to the 1-cosine, 2-cosine, ..., 10-cosine values. According to formula (6), the following table can be obtained:

[0176] Table 2 In x 10= 0.50 corresponding to the 1-cosine, 2-cosine, ..., 10-cosine values

[0177]

[0178] As can be seen from the table above, in the rising and falling trend curves, k-cosine is mostly negative values ​​close to -1, indicating that the curve is not very curved and is relatively straight; while in the curve that rises first and then falls, k-cosine gradually changes from negative to positive, indicating that the curve is 10 =A trend change occurred around 0.50.

[0179] According to Theorem 1, we can find the three curve points listed in x 10 = 0.50 corner strength as follows:

[0180]

[0181] S2.3. Corner determination theorem and k-cosine corner detection algorithm:

[0182] Due to the corner strength The curve is at p i The curvature substitute value at is a good measure of curvature, so it can be used To identify the corner points and inflection points of the curve (in fact, the present invention only uses the corner point detection part, but in order to maintain the integrity of the theorem, the inflection point detection will still be mentioned in the text).

[0183] At a known point p i Under the premise that the corner point strength index is j, define p i Corner point strength radius h i as follows:

[0184]

[0185] For example, in Example 2 the curve In the figure, we examine k-cosine from right to left. Since 10-cosine is greater than 9-cosine, 10-cosine is the corner strength, so j = 10,

[0186] According to the above definition, the determination theorem of corner points and inflection points is given as follows:

[0187] Theorem 2. Corner point determination theorem: Let the curve be about p i (x i ,y i ) has a neighborhood of δ(p i ,h i ), hi For p i The corner strength radius of (|ki|≤h i )Hengyou but is a local maximum, at which point p i is a corner point of the curve.

[0188] Theorem 3. Inflection point determination theorem: Let δ(p i ,h i ) As before, if for Hengyou but is a local minimum, at which point p i is an inflection point of the curve.

[0189] like Figure 7 As shown, based on Theorem 2, the corner points on the curve can be identified, and the algorithm is as follows:

[0190]

[0191]

[0192] S2.4. Improved k-cosine corner detection algorithm:

[0193] When the sampling points of the TcPO2 curve are too dense or too sparse, the algorithm for obtaining the corner point strength in the previous section may lead to incorrect detection results. Therefore, in actual work, an improved method is used to perform data enhancement to improve the robustness of the algorithm.

[0194] Since this improved method requires the calculation of p i The average k-cosine method is the mean of the k-cosine in a neighborhood, so it is called the average k-cosine method. The neighborhood radius r i It is usually set to 1 / 2 of k:

[0195]

[0196] And p i The average k-cosine It is calculated in the following way:

[0197]

[0198] For example, when k = 3, p i Need to calculate c i1 ,c i2 ,c i3 , and from (10) we can know that r i =1, from (11) we get

[0199] Then p i of Treated as c ik , calculate p in turn under k-radius m i of And use Theorem 1 to determine that p i The average corner strength and corner strength index j.

[0200] like Figure 8 As shown, use replace Theorem 2 can also be used to identify the curve {p i}, based on The k-cosine corner detection algorithm is as follows:

[0201]

[0202] The following example illustrates how to use the average corner point intensity to identify the corner points of a curve.

[0203] Example 3:

[0204] A standardized TcPO2 chain curve point sequence with a sample size of n = 101 was constructed, and m = 10 was set. The average corner point intensity of each point was calculated. as follows:

[0205] Table 3 List of average corner point strengths of TcPO2 chain curve points (n=101)

[0206]

[0207]

[0208]

[0209] It is known that the corner point strength radius at i=26,51,76 in the above table is 5. It can be seen from the column that when i=26 and 76, under the condition of |ik|≤5|, According to the corner point determination theorem, i = 26 and 76 are corner points; similarly, when i = 51, under the condition of |ik|≤5, there is always According to the inflection point determination theorem, i=51 is the inflection point.

[0210] Figure 4 The curve diagram of the normalized TcPO2 point sequence with n=101 shows that two corner points are detected on the curve, which are the peak and the trough of the curve; and one inflection point is detected, which is the junction of the peak and the trough of the curve.

[0211] In S3:

[0212] S3.1. Differential slope of the curve:

[0213] It is known that the original TcPO2 curve will become a coordinate point sequence p1(x1,y1),…,p1 in [0,1]×[0,1] after preprocessing. n (x n ,y n ), let Δx i =|x i+1 -x i |,Δy i =|y i+1 -y i |, according to the characteristics of time series, we know that 0<Δx i =Δx i+1 <1,0≤Δy i ≤1.

[0214] Now if we define p by simply replacing the derivative in formula (1) with the first-order difference i The slope of the slope is calculated by the method of first-order difference, which may be affected by local extreme values ​​and change points, resulting in inaccurate slope calculation.

[0215] This computational difficulty can be reduced by using the slope measurement after quadratic smoothing. For example: define p i The slope is (y i+s -y i ) / (x i+s -x i ), the slope is called the s-order differential slope, where s is called the differential step size, and s must be greater than 1. Using the s-order differential slope instead of the first-order differential slope can effectively improve the accuracy of the slope calculation.

[0216] Obviously, if this method is adopted, how to choose the difference step length s is a key issue. i Corner point strength radius h i , the corresponding differential slope formula is as follows:

[0217]

[0218] S3.2.TcPO2 curve characterization tool:

[0219] The TcPO2 curve characterization tool (referred to as TcPO2_RepTool) is a container-type data structure with data processing functions, used to store:

[0220] 1. The segment information of the curve, including:

[0221] (1) Number of segments

[0222] (2) Data point columns for each segment

[0223] (3) Trend type of each segment (concave upward, concave downward, stable, rising, falling, etc.)

[0224] (4) Segment location information (starting point index, segment length, etc.)

[0225] 2. Characteristic information of the curve, including:

[0226] (1) The s-order differential slope κ(h i )

[0227] (2) Corner point strength of each point on the curve

[0228] 3. A collection of algorithms for curve feature extraction and segmentation, including:

[0229] (1) k-cosine corner detection algorithm

[0230] (2) Improved k-cosine corner detection algorithm

[0231] (3) Curve segment recognition algorithm

[0232] (4) Trend type identification algorithm of curve segments

[0233] (5) Trend type identification algorithm for straight line segments

[0234] (6) k-cosine and differential slope collaborative discrimination algorithm

[0235] In the algorithm for trend type identification, the curve trend type and type identifier (enumerated type represented by CTT) defined in TcPO2_RepTool are as follows, where the upward concave and downward concave trends are given shape representations:

[0236] 1. Stable (CTT_Stable)

[0237] 2. CTT_Upward

[0238] 3. Downward (CTT_Downward)

[0239] 4. Concave upward (CTT_Convexity, shape: )

[0240] 5. Concave downward (CTT_Concavity, shape: )

[0241] The corresponding segmented curves are called the steady segment, the rising segment, the falling segment, the upward concave segment, and the downward concave segment respectively.

[0242] TcPO2_RepTool is the data container of the TcPO2 curve trend analysis program. When the TcPO2 data is collected, preprocessed and a standardized curve is generated, the corner strength of each point on the curve is calculated one by one. and the s-order difference slope κ(h i ) and stored in TcPO2_RepTool, and use the improved k-cosine corner detection algorithm to identify corners, and then store the corners in TcPO2_RepTool, and finally use the known corner index, κ(h i ) and various trend type identification algorithms partition the curve according to trend characteristics.

[0243] S3.3. Algorithm for distinguishing trend types of curve segments and straight line segments:

[0244] For the curve sequence {p i (x i ,y i )|i=1,…,n} Let the corner point list obtained by Algorithm 2 be {q l (x l ,y l )|l=1,…,n′}(n′ is the number of corner points detected), and the corner point q l The neighborhood with more significant corner point strength is defined as the curve segment, and the area far away from the corner point and with lower corner point strength is defined as the straight line segment. The union of all curve segments is denoted as C p , the union of all straight line segments is denoted as L p , then {p i =C p ∪L p .

[0245] By further subdividing the curve segments and straight line segments, five types of sub-curve segments can be defined:

[0246] The curve segment can be further divided into: upward concave segment (CTT_Convexity), downward concave segment (CTT_Concavity);

[0247] The straight line segment can be further divided into: a stable segment (CTT_Stable), an upward segment (CTT_Upward) and a downward segment (CTT_Downward).

[0248] The morphological characteristics of the above segments are as follows Figure 5 shown.

[0249] From the above segment definition, we can see that {p i (x i ,y i )|i=1,…,n} can be divided into non-overlapping subsets belonging to these five types of curve segments.

[0250] A corner strength threshold T can be defined c To divide the curve segment and the straight line segment, for the TcPO2 curve, T c The empirical threshold of is generally set between -0.40 and -0.55.

[0251] In addition, let the corner point q l For {p i (x i ,y i )|i=1,…,n} i , then q l About r neighborhoodδ(q l ,r) can be defined as:

[0252]

[0253] For example: l The 3-neighborhood δ(q l ,3) is the point set {p i-3 ,p i-2 ,p i-1 ,p i ,p i+1 ,p i+2 ,p i+3}, the corresponding angle intensity set is About δ(q l ,3) is the average angular intensity:

[0254]

[0255] like Fig. 9 As shown, based on T c The curve segment recognition algorithm is as follows:

[0256]

[0257] If the TcPO2 curve is written in the form of a collection Easy-to-understand curve segment union Line segment union

[0258] In S4:

[0259] S4.1. Algorithm for determining trend type of curve segment:

[0260] Further investigation l The curve segment δ(q l ,r l ). Still assume q l =p i , δ(q l ,r l ) is written as a curve point sequence The corresponding difference slope is recorded in the form of a sequence: Similarly, δ(q l ,r l ) The difference slope sequence of the right half is recorded as: The mean of the difference slopes in the left half is recorded as The mean of the difference slopes in the right half is recorded as Right now:

[0261]

[0262] like Fig.10 As shown, according to the properties of the corner points, and Must be of different sign. and When δ(q l ,r l ) is identified as an upward concave area; when and When δ(q l ,r l ) is identified as a concave area, and the trend type discrimination algorithm of the curve segment is as follows:

[0263]

[0264] S4.2. Algorithm for determining the trend type of straight line segments:

[0265] from Deduct The rest of the L consists entirely of straight line segments. p , let the number of straight line segments be n″, and the straight line segments can be divided into stable segments (with CTT_Stable mark), rising segments (with CTT_Upward mark) and falling segments (with CTT_Downward mark), let the stable segments The number is n1″, the rising section The number of descending sections is n′2′. The number of is n3″, so:

[0266]

[0267] You can use L p The κ(h i ) to identify its stability. For example: first define a differential slope threshold T in a stable area st (In the TcPO2 curve, T st The empirical value of is between 0.5 and 0.9); then examine L p Each point p i κ(h i ): When κ(h i )∈[-T st ,T st ], p i belong κ(h i )∈(-∞,-T st ), p i belong κ(h i )∈(T st ,+∞), p i belong Finally, L p Adjacent points with the same upward trend type are connected, and the resulting connected area set is all the straight line segments that have been divided with CTT marks.

[0268] like Fig.11 As shown, based on the above description, the trend type discrimination algorithm of the straight line segment is designed as follows:

[0269]

[0270]

[0271] S4.3.k-cosine and differential slope collaborative discrimination algorithm:

[0272] like Fig.12 As shown in the figure, the above k-cosine corner point detection algorithm and the curve segment recognition and trend type discrimination algorithm in this section are encapsulated into a filter and connected in the form of an algorithm pipeline, and the collaborative discrimination algorithm for trend analysis of the TcPO2 curve can be obtained as follows:

[0273]

[0274] In S5:

[0275] 1) Data source: selected from a medical database of a medical institution, including 500 TcPO2 curve data. The number of data points in each data is n, and they are all in the chain digital curve point list {p i (x i ,yi )|i=1,…,n} structure storage, each curve is annotated, including the corner points of the curve and the trend segmentation area (concave downward area, concave upward area, stable area, rising area and falling area). The corner point position is marked with the sequence index i of the curve where the corner point is located, and the trend segmentation area uses the starting sequence index i of the curve segment. s and the end sequence index i e To identify.

[0276] 2) Data distribution: The dataset is divided into three categories according to the subject status:

[0277] Health status (200 copies)

[0278] Mild pathological condition (150 copies)

[0279] Severe pathological conditions (150 copies)

[0280] 3) Experimental steps:

[0281] Step 1: Standardize and preprocess all curve data according to the minimum-maximum standardization method;

[0282] Step 2: Perform region segmentation and trend type identification on the curve according to Algorithm 6;

[0283] Step 3: Record the curve cluster recognition results of each curve, compare them with the manual annotations, and calculate the accuracy of the algorithm recognition results under different conditions.

[0284] Here, a percentage matching threshold Threshold is set to determine whether each curve is successfully recognized.

[0285] Suppose the manually marked interval is The identified interval is The overlapping interval is:

[0286]

[0287] The length of the overlapping interval is recorded as:

[0288] [OverLap] = max(0,OverLap end -OverLap start ) (16)

[0289] Add up the lengths of all overlapping intervals on the curve to get the total overlapping interval [OverLap total If the [OverLap total]>n×Threshold, the curve is successfully identified, otherwise it fails to be identified. The number of curve samples in each experiment is N, and the number of successfully identified curves is N. s ≤N. Then the accuracy of algorithm recognition Acc is defined as:

[0290]

[0291] 4) Experimental results analysis:

[0292] The following table shows the accuracy of algorithm recognition under different pathological conditions of the subjects and different percentage matching thresholds.

[0293] Table 4 Comparison of algorithm recognition accuracy under different pathological conditions and different Thresholds

[0294]

[0295] From the physiological state of the subjects: In the healthy state, the algorithm's best Acc under each Threshold condition reached 97.52%, showing good performance. This shows that under ideal conditions, the algorithm can effectively extract curve geometric features and perform regional classification. For pathological and severe pathological states, the Acc value gradually decreased to 92.34%. This degree of accuracy attenuation should be related to the complexity of the curve shape and the presence of noise in the data.

[0296] From the perspective of percentage matching threshold, the Acc of each state decreases as the Threshold increases. When the Threshold is increased to 0.9, the highest Acc is 90.06% and the lowest is only 83.40%, which is about 10% lower than when Threshold = 0.6.

[0297] This shows that as the Threshold parameter gradually increases, the judgment conditions for whether the curve recognition is successful or not become stricter. For a curve recognition result to be successfully judged, a higher and higher interval recognition matching degree is required, and a higher accuracy requirement is placed on the algorithm.

[0298] 4) Discussion

[0299] (1) Factors affecting performance: Under pathological conditions, the TcPO2 curve may have more noise and fluctuations, resulting in unstable feature extraction. That is, the quality of the data and the complexity of the curve significantly affect the accuracy of the algorithm.

[0300] (2) Further improvement direction: We can consider introducing more feature extraction methods, such as wavelet transform or Fourier transform, to enhance the analysis ability of complex curves.

[0301] (3) Clinical application potential: This algorithm can provide support for real-time analysis of clinical TcPO2 curves, helping to quickly identify patients' physiological status and disease changes.

[0302] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0303] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0304] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A TcPO2 curve trend analysis method based on k-cosine corner point detection, characterized in that: The method comprises: Collect and record TcPO2 data in real time, build a chain curve structure, and pre-process the acquired data; Based on the k-cosine method, all corner points in the curve are screened out; Analyze the slope and neighborhood characteristics of the TcPO2 curve, and segment the TcPO2 curve into several straight line segments and curve segments based on the neighborhood corner intensity significance and corner intensity threshold; The trend type discrimination algorithm based on curve segments and straight line segments identifies the trend type of each segment of the TcPO2 curve.

2. The method according to claim 1, characterized in that The real-time collection and recording of TcPO2 data, construction of a chain curve structure, and preprocessing of the acquired data specifically include: A plane rectangular coordinate system is established with the acquisition time of TcPO2 value as the horizontal axis and the TcPO2 value as the vertical axis; Record the TcPO2 value and the corresponding UNIX timestamp at uniform time intervals, and record the original UNIX timestamp as In seconds, the corresponding original TcPO2 value is recorded as In units of mmHg, the sample size of data collected each time is recorded as n, n ≥ 30; Discrete digital point array Perform time-series connection to form a chain curve structure in the plane rectangular coordinate system Using the min-max normalization method Scale to the range of 0 to 1; The structure of the standardized TcPO2 chain curve is: i (x i ,y i )|i=1,…,n;x i ,y i ∈[0,1]}, which can be simply written as {p i (x i ,y i )} or {p i }; The Savitzky-Golay filter is used to smooth the normalized chain digital curve point sequence.

3. The method according to claim 2, characterized in that The method of filtering out all corner points in the curve based on the k-cosine method is as follows: the corner point strength and corner point strength index of each point on the curve are calculated based on the k-cosine method, and all corner points in the curve are filtered out based on the corner point strength, the corner point strength index and the corner point strength radius defined thereby, including: Use k-cosine corner detection algorithm to convert {p i (x i ,x i )} for every point p i Several predecessor points and p i The vector and p i Several successor points of p i The vectors formed are collectively called p i arm, predecessor or successor point to p i The number of intervals is defined as p i arm length; Get with p i The pair of points whose arm lengths are all k <p i-k ,p i+k >, defined as p i The two arms of Defined as p i Forearm, vector Defined as p i The rear arm; Use your forearms and rear arm Define p i The k-vector: The corresponding k-cosine is defined as: Where -1≤c ik ≤1; The longest arm length to be considered is defined as p i The k-radius is denoted as m, where m>0. Based on the TcPO2 curve prior, the k-radius m is set to an integer between 10 and 15; Calculate c based on m i1 ,c i2 ,…,c i,m , then define c i0 = -1, c i0 ,c i1 ,c i2 ,…,c i,m It is denoted as k-cosine sequence with radius m; According to the corner point strength determination theorem, the curve {p i }P i Corner strength and the corner point strength index value j; According to p i The j value is used to calculate the corner strength radius h i , δ(p i ,h i ) is denoted as p′ k (|ki|≤h i ), calculate δ(p i ,h i ) for every point p′ k of According to the corner point determination theorem, p i Is it a corner point? If so, set p i Add corner point array {q l (x l ,y l )|l=1,…,n′}, denoted as {q l (x l ,y l )}, where n' is the number of corner points detected; Repeat the above three steps until {p i All the points in} have been determined.

4. The method according to claim 2, characterized in that: The method of filtering out all corner points in the curve based on the k-cosine method is specifically as follows: calculating the average corner point strength and the average corner point strength index of each point on the curve based on the k-cosine method, and filtering out all corner points in the curve based on the average corner point strength, the average corner point strength index and the corner point strength radius defined thereby, including: Define point p i The neighborhood radius r i : p i The average k-cosine c ik The calculation steps are: Then p i c ik Treated as c ik , under the condition that the k-radius is m, we can calculate According to the corner point strength determination theorem, p i The average corner strength and the average corner intensity index j; use replace Similarly, the corner point determination theorem is used to determine the curve {p i } all corner points.

5. The method according to claim 3 and claim 4, characterized in that The corner point strength determination theorem includes: In k-p with radius m i The k-cosine sequence c i0 ,c i1 ,c i2 ,…,c im , from the rightmost c im The last one counting from the left is larger than the c of its right neighbor ij Determined as p i The corner strength of m is expressed as The corner point determination theorem includes: Suppose the curve is about p i (x i ,y i ) has a neighborhood of δ(p i ,h i ), h i For p i The corner strength radius of (|ki|≤h i )Hengyou but is a local maximum, at which point p i is a corner point of the curve.

6. The method according to claims 3, 4 and 5, characterized in that The curve segment recognition algorithm is used to analyze the neighborhood features of each point on the TcPO2 curve with respect to corner points, and the TcPO2 curve is segmented into a number of straight line segments and curve segments based on the neighborhood corner point intensity significance and corner point intensity threshold, specifically including: The curve {p i (x i ,y i )} is recorded as a union form The union of all curve segments is denoted by C p , the union of all straight line segments is denoted as L p ,but For the above corner point sequence {q l (x l ,y l )}, define the corner point intensity threshold T c , the corner point q l The strength of nearby corner points is greater than or equal to T c The neighborhood of is defined as the curve segment far away from the corner point and where the strength of the corner point is less than T c The area of ​​is defined as a straight line segment; Set the corner point q l For {p i (x i ,y i )} i ,q l About r neighborhoodδ(q l ,r) is defined as: For {q l Each point q in l , calculate q l About δ(q l ,1),δ(q l ,2), the mean value of the corner intensity of… Through the above r neighborhood δ(q l ,r) is an infinite sequence of pairs of corner point intensity means ,…get inf The corresponding δ(q l ,r l ), and δ(q l ,r l ) is set as the corner point q l The corresponding curve segment; Repeat the above two steps until {q l }, and the corresponding curve segment set {δ(q l ,r l )|l=1,…,n′} is recorded as a union: Find the set of all straight line segments (the complement of the curve segment set) and record them in the form of union as follows:

7. The method according to claim 6, characterized in that The trend type discrimination algorithm based on curve segments and straight line segments identifies the trend type of each segment of the TcPO2 curve, wherein the trend type discrimination algorithm based on curve segments identifies the trend type of each curve segment on the TcPO2 curve, and the specific steps include: Define a point p on the curve i The slope of the s-order difference is (y i+s -y i ) / (x i+s -x i ), where s is the difference step size, and s>1; Set the corner strength radius h i As the difference step size s, then p i The differential slope formula is: The curve segments are divided into upward concave segments CTT_Convexity and downward concave segments CTT_Concavity according to the trend type; For the curve segment set C p Each segment δ(q l ,r l ), respectively calculate and obtain the difference slope mean sequence on the left: And the difference slope mean series on the right: Calculate δ(q l ,r l )The mean of the difference slopes between the left and right parts and Set the judgment conditions for the upward concave section and the downward concave section: According to the judgment conditions, the segment δ(q l ,r l )’s trend type; Repeat the above process to derive C p Trend type for all curve segments above.

8. The method according to claim 6, characterized in that The trend type discrimination algorithm based on curve segments and straight line segments identifies the trend type of each segment of the TcPO2 curve, wherein the trend type discrimination algorithm based on straight line segments identifies the trend type of each straight line segment on the TcPO2 curve, and the specific steps include: Using the differential slope of each point of the TcPO2 curve and the differential slope threshold T st , identify the curve {p i (x i ,y i )}The trend type of each straight line segment includes: The straight line segments are divided into stable segments CTT_Stable, rising segments CTT_Upward and falling segments CTT_Downward according to trend types; Set the judgment conditions for the stable section, rising section, and falling section: For the straight line segment L p Every point p in i , calculate p i κ(h i ), and then derive p according to the judgment conditions of the stable section, rising section, and falling section i Trend type; Connect all adjacent points that belong to the stable segment to obtain n1″ stable segments, denoted as Connect all adjacent points that belong to the ascending segment to obtain n′2′ ascending segments, which are recorded as Connect all adjacent points that belong to the descending segment to obtain n3″ ascending segments, denoted as

Images