Method for evaluating benign and malignant tumor incisal margins during operation

Abstract

The invention relates to the technical field of medical treatment, in particular to an intraoperative tumor incision margin benign and malignant evaluation method, which comprises the following steps: acquiring optical coherence tomography data of historical incision margin tissues; extracting an A-line reflectivity curve based on the optical coherence tomography data of the historical cutting edge tissue; calculating and extracting target features based on an A-line reflectivity curve, wherein the target features comprise a depth feature, a slope feature, a local maximum distance feature, an intensity rapid fluctuation feature, a frequency feature and an optical attenuation coefficient feature; based on the target features and benign and malignant labels of the A-line reflectivity curve corresponding to the target features, training a preset classifier to obtain a trained classifier; and inputting target features extracted based on the optical coherence tomography data of the to-be-detected cutting edge tissue into the trained classifier to obtain an evaluation result. Therefore, the accuracy and the real-time performance of benign and malignant tumor incision edge evaluation can be greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a method for evaluating the benign or malignant nature of a tumor resection margin during surgery. Background Art

[0002] With the continuous expansion of the cancer patient population, surgical treatment has become the main treatment method. In this process, the accuracy of intraoperative margin assessment is crucial to prevent patients from undergoing secondary surgery, especially the key link to ensure the success of the operation and patient recovery. Achieving negative intraoperative margins can not only ensure that the tumor is completely removed, but also effectively reduce damage to healthy tissues, significantly improving the patient's quality of life and recovery speed after surgery. Therefore, strengthening the accuracy of intraoperative margin assessment is of great significance for the treatment of cancer patients.

[0003] Optical Coherence Tomography (OCT) is a cutting-edge non-invasive imaging technology that uses subtle changes in the light scattering properties of different tissues to reveal the three-dimensional microstructural details of tissues in a label-free, real-time manner. OCT has a unique penetrating ability, capable of penetrating about 2mm below the tissue surface while maintaining a spatial resolution of up to 10-20 microns, thereby generating images of similar resolution scale to that of histopathology (histopathology resolution can reach 0.25 microns (40X), 0.5 microns (20X), OCT resolution is 10-20 microns, and the current image quality based on OCT technology (fullfield OCT resolution is 1 micron). Given that a 2mm margin is usually used as an important basis for determining a negative tumor margin in clinical practice, OCT has become the best choice for intraoperative evaluation of a negative tumor margin with its advantages of high resolution and real-time imaging.

[0004] OCT-based tumor assessment technology has been widely studied and has been shown to have a high correlation with histopathological examination. Currently, this technology is mainly used to train different specialists to identify suspicious and non-suspicious areas at the edge of tumor specimens during surgery. However, due to the huge amount of OCT scan data collected during surgery and the need to make real-time judgments on suspicious areas, relying solely on the naked eye evaluation of clinicians is time-consuming and laborious. Therefore, developing an algorithm that can automatically identify and mark these suspicious areas can not only effectively improve the evaluation efficiency, but also ensure the accuracy of the evaluation. The application of this algorithm will play a key role in real-time guidance of tumor surgery. Summary of the invention

[0005] In view of this, an object of the present invention is to provide a method for intraoperative assessment of the benign and malignant nature of tumor resection margins, so as to overcome the current problems of poor accuracy and real-time performance in intraoperative assessment of the benign and malignant nature of tumor resection margins.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions: The present application provides a method for evaluating the benign or malignant nature of a tumor resection margin during surgery, comprising: Obtain optical coherence tomography data of historical resection margin tissue; Extracting an A-line reflectivity curve based on the optical coherence tomography data of the historical resection edge tissue, wherein the A-line reflectivity curve is a curve showing changes in image intensity with the depth of the A-line; Calculating and extracting target features based on the A-line reflectivity curve, the target features include depth features, slope features, distance features between local maxima, rapid intensity fluctuation features, frequency features, and optical attenuation coefficient features; Based on the target feature and the benign and malignant labels of the A-line reflectivity curve corresponding to the target feature, a preset classifier is trained to obtain a trained classifier; The target features extracted based on the optical coherence tomography data of the incision margin tissue to be measured are input into the trained classifier to obtain an evaluation result.

[0007] Further, in some embodiments of the present application, the depth feature includes a penetration depth of a valid signal interval, and the process of extracting the penetration depth of the valid signal interval includes: Smoothing the current A-line reflectivity curve to obtain a current A-line reflectivity smoothing curve; Determine the effective signal interval of the current A-line reflectivity smoothing curve; The difference between the depths of the starting point and the end point of the effective signal interval is used as the penetration depth of the effective signal interval of the current A-line reflectivity curve.

[0008] Further, in some embodiments of the present application, determining the valid signal interval of the current A-line reflectivity smooth curve includes: The depth position corresponding to the first time the image intensity reaches the target value in the current A-line reflectivity smoothing curve is taken as the starting point of the effective signal interval; The part after the starting point of the effective signal interval of the current A-line reflectivity smooth curve is selected by preset sliding window. When the linear fitting slope of the signal in the window tends to 0 or the signal in the window is completely 0, the starting depth position in the current window is taken as the end point of the effective signal interval.

[0009] Further, in some embodiments of the present application, the slope feature includes: an average slope of a valid signal interval, an overall slope of a valid signal interval, and an overall intercept of a valid signal interval, and the process of extracting the slope feature includes: The effective signal interval of the current A-line reflectivity smooth curve is selected by sliding a preset sliding window to obtain multiple data subsets; Calculate the linear fitting slope corresponding to each data subset respectively, and take the average of the linear fitting slopes of all data subsets as the average slope of the effective signal interval of the current A-line reflectivity curve; Calculate the linear fitting slope and intercept of the effective signal interval of the current A-line reflectivity smooth curve, use the slope as the overall slope of the effective signal interval of the current A-line reflectivity curve, and use the intercept as the overall intercept of the effective signal interval of the current A-line reflectivity curve.

[0010] Further, in some embodiments of the present application, the distance feature between local maxima includes: a distance mean and a distance standard deviation, and the process of extracting the distance feature between local maxima includes: Determine the local maximum value of the image intensity in the valid signal interval of the current A-line reflectivity curve; Calculating the difference in depths corresponding to adjacent local maxima; The mean of all the differences is taken as the distance mean of the current A-line reflectivity curve, and the standard deviation of all the differences is taken as the distance standard deviation of the current A-line reflectivity curve.

[0011] Further, in some embodiments of the present application, the rapid intensity fluctuation feature includes: a rapid intensity fluctuation standard deviation and a rapid intensity fluctuation peak value, and the process of extracting the rapid intensity fluctuation feature includes: Smoothing the current A-line reflectivity curve to obtain a current A-line reflectivity smoothing curve; Determining a difference in rapid intensity fluctuations between the current A-line reflectivity curve and the current A-line reflectivity smooth curve; Calculating the standard deviation of the difference as the standard deviation of the intensity rapid fluctuation of the current A-line reflectivity curve; A difference curve is generated based on the difference, and a local maximum value with the largest absolute value in the difference curve is used as the intensity rapid fluctuation peak value of the current A-line reflectivity curve.

[0012] Further, in some embodiments of the present application, the frequency-related features include: a weighted average of a window frequency spectrum, an area of ​​a main peak of a window frequency spectrum, the number of main peaks of a window frequency spectrum, and a standard deviation of a window frequency spectrum. The process of extracting the frequency-related features includes: The current A-line reflectivity curve is selected by sliding a preset sliding window to obtain multiple groups of A-line reflectivity curve information; For each group of A-line reflectivity curve information, it is converted from the time domain to the frequency domain by fast Fourier transform to determine the frequency spectrum of each frequency point in the group, and the weighted average of the frequency spectrum of the current group of A-line reflectivity curve information is calculated by the following formula:

[0013] Where k is the number of each frequency point; N is the total number of frequency components; f k is the frequency at the k frequency point; F (f k ) is f k The frequency spectrum corresponding to the frequency; The average of the weighted average frequency spectra of all groups of A-line reflectivity curve information is used as the weighted average of the window frequency spectrum of the current A-line reflectivity curve; For each group of A-line reflectivity curve information, determine the target frequency point in the group whose frequency spectrum is greater than the first preset threshold, and accumulate the frequency spectra corresponding to all the target frequency points in the group to obtain the main peak area of ​​the frequency spectrum of the current group of A-line reflectivity curve information; take the average of the main peak areas of the frequency spectrum of all groups of A-line reflectivity curve information as the main peak area of ​​the window frequency spectrum of the current A-line reflectivity curve; For each group of A-line reflectivity curve information, determine the number of frequency points in the group whose frequency spectrum is greater than the second preset threshold, and obtain the number of main peaks in the frequency spectrum of the current group of A-line reflectivity curve information; take the average of the number of main peaks in the frequency spectrum of all groups of A-line reflectivity curve information as the number of main peaks in the window frequency spectrum of the current A-line reflectivity curve; For each group of A-line reflectivity curve information, the standard deviation of the frequency spectrum in the group is calculated to obtain the frequency spectrum standard deviation of the current group of A-line reflectivity curve information; the average of the frequency spectrum standard deviations of all groups of A-line reflectivity curve information is used as the window frequency spectrum standard deviation of the current A-line reflectivity curve.

[0014] Furthermore, in some embodiments of the present application, the frequency-related features also include a weighted average of the overall frequency spectrum, an area of ​​a main peak of the overall frequency spectrum, the number of main peaks of the overall frequency spectrum, and a standard deviation of the overall frequency spectrum. The process of extracting the frequency-related features includes: All information of the current A-line reflectivity curve is converted from the time domain to the frequency domain through fast Fourier transform to determine the frequency spectrum of each frequency point, and the overall frequency spectrum weighted average of the current A-line reflectivity curve is calculated by the following formula:

[0015] Where k is the number of each frequency point; N is the total number of frequency components; fk is the frequency at the k frequency point; F (f k ) is f k The frequency spectrum corresponding to the frequency; Determine a target frequency point whose frequency spectrum is greater than a first preset threshold value in all information of the current A-line reflectivity curve, and accumulate the frequency spectra corresponding to the target frequency point to obtain a main peak area of ​​the overall frequency spectrum of the current A-line reflectivity curve; Determine the number of frequency points whose frequency spectrum is greater than a second preset threshold value in all information of the current A-line reflectivity curve, and obtain the number of main peaks of the overall frequency spectrum of the current A-line reflectivity curve; The standard deviation of the frequency spectrum in all information of the current A-line reflectivity curve is calculated to obtain the overall frequency spectrum standard deviation of the current A-line reflectivity curve.

[0016] Furthermore, in some embodiments of the present application, the optical attenuation coefficient-like features include: an attenuation coefficient and an attenuation coefficient amplitude, and the process of extracting the optical attenuation coefficient-like features includes: The axial confocal diffusion function and additive background noise of the OCT system are introduced into the light attenuation model based on Beer's law, and the tissue optical attenuation model based on the OCT system is obtained. Performing nonlinear fitting on the optical coherence tomography data of the historical resection edge tissue by the least square method to determine the target parameters in the tissue optical attenuation model based on the OCT system; Wherein, the target parameters include the attenuation coefficient and the attenuation coefficient amplitude.

[0017] Further, in some embodiments of the present application, the tissue optical attenuation model based on the OCT system is:

[0018] Where I(z) is the light intensity at depth z; cf z is the depth of the focal plane from the tissue boundary; R is the Rayleigh length; A1 is the OCT signal amplitude; µ t is the attenuation coefficient; A2 is the additive background noise.

[0019] The present application relates to the field of medical technology, and specifically to a method for intraoperative assessment of the benign and malignant nature of tumor margins, the method comprising: obtaining optical coherence tomography data of historical margin tissues; extracting an A-line reflectivity curve based on the optical coherence tomography data of historical margin tissues, the A-line reflectivity curve being a curve showing changes in image intensity with the depth of the A-line; extracting target features based on the A-line reflectivity curve, the target features including depth features, slope features, distance features between local maxima, rapid intensity fluctuation features, frequency features, and optical attenuation coefficient features; training a preset classifier based on the target features and the benign and malignant labels of the A-line reflectivity curve corresponding to the target features to obtain a trained classifier; inputting the target features extracted based on the optical coherence tomography data of the margin tissue to be tested into the trained classifier to obtain an assessment result. In this way, the accuracy and real-time performance of the assessment of the benign and malignant nature of tumor margins can be greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is a flow chart of a method for evaluating the benign and malignant nature of a tumor resection margin during surgery provided by an embodiment of the present invention; Figure 2 It is an OCT image in the method for evaluating benign and malignant tumor resection margins during surgery provided by an embodiment of the present invention; Figure 3 Schematic diagram of the principle of the method for evaluating the benign and malignant nature of tumor resection margins during surgery provided by an embodiment of the present invention; Figure 4 is a schematic diagram of extracting depth features in a method for evaluating benign and malignant tumor margins during surgery provided by an embodiment of the present invention; Figure 5 It is a schematic diagram of extracting slope features in the method for evaluating benign and malignant tumor resection margins during surgery provided by an embodiment of the present invention; Figure 6 is a schematic diagram of extracting the distance feature between local maxima in the method for evaluating the benign and malignant nature of tumor resection margins during surgery provided by an embodiment of the present invention; Figure 7 is a schematic diagram of extracting rapid intensity fluctuation features in the method for evaluating benign and malignant tumor margins during surgery provided by an embodiment of the present invention; Figure 8 is a schematic diagram of extracting frequency-based features in the method for evaluating benign and malignant tumor margins during surgery provided by an embodiment of the present invention; Fig. 9 It is a typical representative image among the six categories of features of two types of samples, benign and malignant, in the method for evaluating the benign and malignant nature of intraoperative tumor resection margins provided by an embodiment of the present invention; Fig.10 The diagram is a verification effect diagram for evaluating a classifier trained using the intraoperative tumor resection margin benign and malignant assessment method provided in the present application. DETAILED DESCRIPTION

[0022] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0023] Figure 1 This is a flow chart of the method for evaluating the benign and malignant nature of tumor resection margins during surgery provided by an embodiment of the present invention. Figure 1 , this embodiment may include the following steps: S101. Obtain optical coherence tomography data of historical resection edge tissue.

[0024] S102. Extracting an A-line reflectivity curve based on the optical coherence tomography data of the historical resection edge tissue.

[0025] It should be noted that the A-line reflectivity curve mentioned in the present application is consistent with the A-line reflectivity curve in the prior art, which is a curve showing how the image intensity changes with the depth of the A-line. That is, in the OCT imaging modality, generally multiple A-line reflectivity curves constitute a reflectivity profile of a B-scan, and the numerous B-scans within the scanning range constitute a three-dimensional reflectivity body with an en face cross-section.

[0026] S103, extracting target features based on the A-line reflectivity curve calculation.

[0027] Specifically, in the present application, the target features calculated and extracted based on the A-line reflectivity curve include depth features, slope features, distance features between local maxima, rapid intensity fluctuation features, frequency features, and optical attenuation coefficient features.

[0028] S104 , based on the target features and the benign and malignant labels of the A-line reflectivity curve corresponding to the target features, a preset classifier is trained to obtain a trained classifier.

[0029] Specifically, in the present application, the historical resection margin tissue or A-line reflectivity curve can be used to mark the benign or malignant nature of the tumor based on the actual existing information data. After completing the above-mentioned marking, the A-line reflectivity curve extracted based on the optical coherence tomography data of the historical resection margin tissue, and the target features extracted based on the A-line reflectivity curve will have corresponding benign or malignant labels, thereby training the preset classifier by calculating the determined target features and the corresponding labels to obtain the trained classifier.

[0030] S105, inputting the target features extracted based on the optical coherence tomography data of the measured cutting edge tissue into the trained classifier to obtain an evaluation result.

[0031] Using the same target feature extraction principle, the features extracted from the optical coherence tomography data of the resection margin tissue to be tested are input into the trained classifier to obtain the evaluation results, thereby realizing the intraoperative evaluation of the benign and malignant nature of the tumor resection margin.

[0032] The method for evaluating the benign and malignant nature of tumor resection margins provided by the present application is to obtain the optical coherence tomography data of historical resection margin tissues; extract the A-line reflectivity curve based on the optical coherence tomography data of historical resection margin tissues, and the A-line reflectivity curve is a curve showing the change of image intensity with the depth of the A-line; calculate and extract target features based on the A-line reflectivity curve, and the target features include depth features, slope features, distance features between local maxima, rapid intensity fluctuation features, frequency features, and optical attenuation coefficient features; train a preset classifier based on the target features and the benign and malignant labels of the A-line reflectivity curve corresponding to the target features to obtain a trained classifier; input the target features extracted based on the optical coherence tomography data of the resection margin tissue to be tested into the trained classifier to obtain an evaluation result. In this way, the accuracy and real-time performance of the evaluation of the benign and malignant nature of tumor resection margins can be greatly improved.

[0033] It should be noted that the number of A-line reflectivity curves extracted based on the optical coherence tomography data of historical resection edge tissue is generally multiple, so in this application, a set consisting of a large number of A-line reflectivity curves can be referred to as an A-line reflectivity curve set.

[0034] In some embodiments of the present application, for a specific optical coherence tomography data based on historical cutting edge tissue, the A-line reflectivity curve is extracted by removing DC, performing window function, dispersion correction and Fourier transform on the original signal collected by the swept-frequency OCT, thereby obtaining the OCT A-line reflectivity curve.

[0035] In some embodiments of the present application, the process of extracting and labeling the A-line reflectivity curve may include: first extracting the A-line reflectivity curve sets of normal tissue and cancer tissue OCT respectively. The process of determining the benign and malignant labels of the A-line curve may be: first, the OCT specialist combines the OCT image (such as Figure 2 The overall contour, gross structural area and microscopic area features are matched one by one with the pathological H&E image; then the pathologist labels the pathological H&E image by category; then the OCT specialist labels the OCT image by category based on the category labeling of the pathological H&E image; finally, the OCT specialist and the pathologist finally confirm the benign and malignant label of the A-line curve based on the labeling results of both parties (as shown in the figure). Figure 3 as shown).

[0036] On this basis, the process of extracting each target feature in this application is introduced in detail below.

[0037] First, the depth feature may specifically include the penetration depth of the effective signal interval. The process of extracting the penetration depth of the effective signal interval includes: smoothing the current A-line reflectivity curve to obtain the current A-line reflectivity smooth curve; determining the effective signal interval of the current A-line reflectivity smooth curve; using the difference in depth between the starting point and the end point of the effective signal interval as the penetration depth of the effective signal interval of the current A-line reflectivity curve (e.g. Figure 4 as shown).

[0038] And determining the effective signal interval of the current A-line reflectivity smooth curve, which may specifically include: taking the depth position corresponding to the first time the image intensity reaches the target value in the current A-line reflectivity smooth curve as the starting point of the effective signal interval; selecting the part after the starting point of the effective signal interval of the current A-line reflectivity smooth curve through a preset sliding window, and when the linear fitting slope of the signal in the window tends to 0 or the signal in the window is completely 0, taking the starting depth position of the current window as the end point of the effective signal interval.

[0039] In practical applications, the following operations are performed for each A-line reflectivity curve: Firstly, the A-line reflectivity curve is smoothed by using a Savitzky-Golay filter through polynomial fitting to deal with the noise while retaining the peak and valley characteristics of the data, and the corresponding A-line reflectivity smooth curve is obtained.

[0040] Then, the starting position of the effective signal, i.e., the starting point of the effective signal interval, is determined on the A-line reflectivity smooth curve, including: reaching the target value for the first time, such as the maximum image intensity value of all signals in the current A-line reflectivity smooth curve The depth of 10% is taken as the starting point of the effective signal interval. . And further determine the end position of the effective signal, that is, the end point of the effective signal interval. Specifically, a sliding, fixed-size window can be used to determine the end position of the effective signal. In practical applications, the fixed length L of the sliding window can be set to 20% of the length of the entire A-line reflectivity curve, and the two coverage areas before and after the window have a 50% overlap. When the slope k of the linear fit of the signal in the window tends to zero, or the signal on the entire window is completely zero, the starting position of the current window is determined to be the end point of the effective signal interval. By determining the starting point and the end point, the penetration depth of the effective signal interval is obtained. Among them, formula 1 describes the specific calculation method of the penetration depth of the effective signal interval: (1) In the formula, D represents the penetration depth of the effective signal interval, d end Indicates the depth of the end point of the valid signal interval, d start Indicates the depth of the start point of the valid signal interval.

[0041] Furthermore, the slope feature includes an average slope of the effective signal interval, an overall slope of the effective signal interval, and an overall intercept of the effective signal interval. For each A-line reflectivity smooth curve, the process of extracting the slope feature includes: sliding and selecting the effective signal interval of the current A-line reflectivity smooth curve through a preset sliding window to obtain multiple data subsets; calculating the linear fitting slope corresponding to each data subset respectively, and taking the average value of the linear fitting slopes of all data subsets as the average slope of the effective signal interval of the current A-line reflectivity curve; calculating the linear fitting slope and intercept of the effective signal interval of the current A-line reflectivity smooth curve, taking the slope as the overall slope of the effective signal interval of the current A-line reflectivity curve, and taking the intercept as the overall intercept of the effective signal interval of the current A-line reflectivity curve (such as Figure 5 as shown).

[0042] Specifically, for each A-line reflectivity smoothing curve, the following operations are performed: First, the A-line reflectivity smooth curve is smoothed and the effective signal interval is determined by the same principle as in the above deep feature extraction content, and the A-line reflectivity smooth curve is selected by sliding the sliding window, where each selection obtains a data subset. On this basis, the signal strength data of the smooth curve in each window is The slope was calculated using the linear fit equation (i.e., Formula 2).

[0043] For a window of length L A linear fit is performed on the data subset within the formula 2, where k1-k n is the linear fitting slope of different window data (that is, different data subsets), c1-c n is the intercept of different windows, d1-d n is the fitted data, where d1=d stat , the specific linear fitting equation is shown in Formula 2: (2) After the window slides through all valid signals, the average value is calculated using Formula 3 to obtain the average slope of the valid signal interval of the A-line reflectivity curve. The specific formula is as follows: (3) in, The total number of window slides.

[0044] Furthermore, for each A-line reflectivity curve, all data of the overall effective signal interval of the A-line reflectivity smoothing curve are directly used to fit using Formula 4 to obtain the overall slope of the effective signal interval of the A-line reflectivity curve and the overall intercept of the effective signal interval.

[0045] (4) Furthermore, the distance feature between local maxima includes a distance mean and a distance standard deviation. The process of extracting the distance feature between local maxima includes: determining the local maximum value of the image intensity in the effective signal interval of the current A-line reflectivity curve; calculating the difference in depth corresponding to each adjacent local maximum value; taking the mean of all differences as the distance mean of the current A-line reflectivity curve, and taking the standard deviation of all differences as the distance standard deviation of the current A-line reflectivity curve (such as Figure 6 as shown).

[0046] Specifically, the following operations can be performed for each A-line reflectivity curve: Calculate the distance between adjacent local maxima of image intensity in the A-line reflectivity curve, and obtain their mean and standard deviation.

[0047] In practical applications, the argrelextrema function in the SciPy library can be used to identify the depth position d1-d of the local maximum of all image intensities in the effective signal interval in the A-line reflectivity curve. m , where m is the number of local maxima. Then the mean distance between the depth positions corresponding to these adjacent image intensities is calculated , expressed by Formula 5, where d iis the depth position of the local maximum value of the i-th image intensity, and the standard deviation of the distance As shown in formula 6: (5) (6) Furthermore, the rapid intensity fluctuation feature includes a rapid intensity fluctuation standard deviation and a rapid intensity fluctuation peak value. The process of extracting the rapid intensity fluctuation feature includes: smoothing the current A-line reflectivity curve to obtain the current A-line reflectivity smooth curve; determining the difference of the rapid intensity fluctuation between the current A-line reflectivity curve and the current A-line reflectivity smooth curve; calculating the standard deviation of the difference as the rapid intensity fluctuation standard deviation of the current A-line reflectivity curve; generating a difference curve based on the difference, and taking the local maximum value with the largest absolute value in the difference curve as the rapid intensity fluctuation peak value of the current A-line reflectivity curve (such as Figure 7 as shown).

[0048] Specifically, the following operations can be performed for each A-line reflectivity curve: First, by calculating the image intensity in the A-line reflectivity curve The image intensity in the corresponding A-line reflectivity smooth curve The difference between the rapid intensity fluctuations between and is used to eliminate the slow intensity changes caused by normal light attenuation, as shown in Formula 7: (7) Wherein, N represents the length of the A-line reflectivity curve or the A-line reflectivity smoothing curve (the two are the same), that is, the number of image depth pixels.

[0049] Then, according to Formula 8, these differences are calculated Standard Deviation : (8) in Is the difference The average value of .

[0050] Use Formula 9 to find the difference curve (based on the difference And the corresponding depth generation, that is Figure 7 The local maximum value with the largest absolute value in the intensity rapid fluctuation curve shown in FIG. 1 is taken as the intensity rapid fluctuation peak value, and formula 9 is as follows: (9) Furthermore, the frequency characteristics include the weighted average of the window frequency spectrum, the main peak area of ​​the window frequency spectrum, the number of main peaks of the window frequency spectrum, and the standard deviation of the window frequency spectrum. The process of extracting the frequency characteristics includes: For the weighted average of the window frequency spectrum: the current A-line reflectivity curve is selected through a preset sliding window to obtain multiple groups of A-line reflectivity curve information; for each group of A-line reflectivity curve information, it is converted from the time domain to the frequency domain through fast Fourier transform to determine the frequency spectrum of each frequency point in the group, and the weighted average of the frequency spectrum of the current group of A-line reflectivity curve information is calculated; the average of the weighted average of the frequency spectra of all groups of A-line reflectivity curve information is used as the weighted average of the window frequency spectrum of the current A-line reflectivity curve.

[0051] For the main peak area of ​​the window frequency spectrum: for each group of A-line reflectivity curve information, determine the target frequency point in the group whose frequency spectrum is greater than the first preset threshold, and accumulate the frequency spectra corresponding to all the target frequency points in the group to obtain the main peak area of ​​the frequency spectrum of the current group of A-line reflectivity curve information; take the average of the main peak areas of the frequency spectrum of all groups of A-line reflectivity curve information as the main peak area of ​​the window frequency spectrum of the current A-line reflectivity curve.

[0052] Regarding the number of main peaks in the window frequency spectrum: for each group of A-line reflectivity curve information, determine the number of frequency points in the group whose frequency spectrum is greater than the second preset threshold, and obtain the number of main peaks in the frequency spectrum of the current group of A-line reflectivity curve information; take the average number of main peaks in the frequency spectrum of all groups of A-line reflectivity curve information as the number of main peaks in the window frequency spectrum of the current A-line reflectivity curve.

[0053] For the standard deviation of the window frequency spectrum: for each group of A-line reflectivity curve information, calculate the standard deviation of the frequency spectrum in the group to obtain the standard deviation of the frequency spectrum of the current group of A-line reflectivity curve information; take the average of the standard deviations of the frequency spectrum of all groups of A-line reflectivity curve information as the standard deviation of the window frequency spectrum of the current A-line reflectivity curve (such as Figure 8 as shown).

[0054] Specifically, perform the following operations for each A-line reflectivity curve: First, the signal of each window is converted from the time domain to the frequency domain through fast Fourier transform (that is, the A-line reflectivity curve is selected by sliding the sliding window to obtain information data of multiple windows, where the information data selected by each window is a set of A-line reflectivity curve information, that is, the signal of the window), the amplitude of each frequency point is calculated to obtain the frequency spectrum, and the frequency spectrum is normalized.

[0055] On this basis, the weighted average of the window frequency spectrum is taken: the weighted average of the signal of each window (that is, the information of each group of A-line reflectivity curves) is obtained, the frequency in the normalized result is used as the weight, the weighted average of the frequency spectrum is calculated, and the weighted average of the signal of each window (that is, the information of each group of A-line reflectivity curves) is obtained. The average is taken to obtain the weighted average of the window frequency spectrum corresponding to the A-line reflectivity curve, as shown in Formula 10: (10) Where k is the number of each frequency point; N is the total number of frequency components; f k is the frequency at the k frequency point; F (f k ) is f k The frequency spectrum corresponding to the frequency.

[0056] For the main peak area of ​​the window frequency spectrum: for the signal of each window (that is, each group of A-line reflectivity curve information), by analyzing the frequency spectrum, find the frequency points where the frequency spectrum exceeds the set threshold T, and accumulate the frequency spectra at these frequency points to obtain the total frequency of the peak area of ​​the entire main frequency composition of the signal of each window (that is, each group of A-line reflectivity curve information), that is, to obtain the main peak area Area of ​​the frequency spectrum of the signal of each window (that is, each group of A-line reflectivity curve information). F , as shown in formula 11. Then the average of the main peak areas of the frequency spectrum of the signals of all windows (that is, the information of all groups of A-line reflectivity curves) is taken as the main peak area of ​​the window frequency spectrum of the A-line reflectivity curve.

[0057] (11) For the number of main peaks in the frequency spectrum of the window: for the signal of each window (that is, each group of A-line reflectivity curve information), the number of frequency points in the frequency spectrum density that exceed the set threshold T is calculated as the number of main peaks in the frequency spectrum, as shown in formula 12, and the number of main peaks in the frequency spectrum of the signal of each window (that is, each group of A-line reflectivity curve information) is obtained. F , and the number of main peaks in the window frequency spectrum of the A-line reflectivity curve is obtained by taking the average.

[0058] (12) For the standard deviation of the window frequency spectrum: for the signal of each window (that is, each group of A-line reflectivity curve information), calculate the standard deviation of the frequency spectrum respectively, and obtain the standard deviation characteristics of the frequency spectrum of the signal of each window (that is, each group of A-line reflectivity curve information). , and then take the average to obtain the window frequency spectrum standard deviation of the A-line reflectivity curve, as shown in Formula 13.

[0059] (13) Furthermore, the frequency-based features also include the weighted average of the overall frequency spectrum, the main peak area of ​​the overall frequency spectrum, the number of main peaks of the overall frequency spectrum, and the standard deviation of the overall frequency spectrum. The process of extracting frequency-based features includes: converting all information of the current A-line reflectivity curve from the time domain to the frequency domain by fast Fourier transform to determine the frequency spectrum of each frequency point, and calculating the weighted average of the overall frequency spectrum of the current A-line reflectivity curve; determining the target frequency point whose frequency spectrum is greater than the first preset threshold in all information of the current A-line reflectivity curve, and accumulating the frequency spectrum corresponding to the target frequency point to obtain the main peak area of ​​the overall frequency spectrum of the current A-line reflectivity curve. Determine the number of frequency points whose frequency spectrum is greater than the second preset threshold in all information of the current A-line reflectivity curve, and obtain the number of main peaks of the overall frequency spectrum of the current A-line reflectivity curve.

[0060] The standard deviation of the frequency spectrum in all information of the current A-line reflectivity curve is calculated to obtain the overall frequency spectrum standard deviation of the current A-line reflectivity curve.

[0061] It should be noted that the above-mentioned calculation method of the frequency-type features for the entire effective signal interval is based on the same principle as the calculation method of the frequency-type features for the window, and the only difference is that the operation objects are different (the operation object of the frequency-type features for the entire effective signal interval is the data information of the effective signal interval of the entire A-line reflectivity curve, and the operation object of the frequency-type features for the window is the data of each window of the A-line reflectivity curve and then the average is taken). You can refer to the calculation method of the frequency-type features for the window for understanding, and will not be repeated here.

[0062] Furthermore, the optical attenuation coefficient-like features include attenuation coefficient and attenuation coefficient amplitude. The process of extracting the optical attenuation coefficient-like features includes: introducing the axial confocal diffusion function and additive background noise of the OCT system into the light attenuation model based on Beer's law to obtain a tissue optical attenuation model based on the OCT system; performing nonlinear fitting on the optical coherence tomography data of the historical resection edge tissue by the least squares method to determine the target parameters in the tissue optical attenuation model based on the OCT system; wherein the target parameters include the attenuation coefficient and the attenuation coefficient amplitude.

[0063] Specifically, in an embodiment of the present application, the axial confocal spread function (PSF) and additive background noise of the OCT system are introduced into the light attenuation model based on Beer's law to obtain a tissue optical attenuation model based on the OCT system. The distribution of intensity with depth, that is, the model, can be described by Formula 14.

[0064] In practical applications, the optical coherence tomography data of historical resection edge tissue can be nonlinearly fitted by the least squares method to obtain three unknown parameters, namely the target parameters: the OCT signal amplitude A1 at the tissue boundary, the optical attenuation coefficient , and the additive base noise A2.

[0065] In some implementations of the present application, in order to better find the optimal solution for these parameters, the Levenberg-Marquardt numerical optimization iterative algorithm may be used for implementation.

[0066] (14) Where I(z) is the light intensity at depth z, cf is the depth of the focal plane from the tissue boundary, z R is the Rayleigh length, which can be determined from the optical coherence tomography data of the historical resection edge tissue, where the attenuation coefficient amplitude can be obtained based on I(z).

[0067] On the basis of the above embodiment, 18 target features in six categories (i.e., depth features, slope features, distance features between local maxima, rapid intensity fluctuation features, frequency features, and optical attenuation coefficient features) determined based on the optical coherence tomography data of historical resection margin tissue are input together with their corresponding benign and malignant labels into a preset classifier, such as a random forest, support vector machine, naive Bayes classifier, neural network classifier, etc. for training and verification. The trained classifier outputs the probability value of benign and malignant, and has the ability to evaluate the benign and malignant nature of the A-line.

[0068] The trained classifier is used to evaluate the benign and malignant nature of the OCT A-line. During training and verification, the input of the classifier is the 18 target features of the OCT A-line and the corresponding benign and malignant labels (during actual evaluation, the input is the corresponding features extracted from the optical coherence tomography data based on the resection margin tissue to be tested using the same principle as above). In some embodiments of the present application, typical representative images of the six major features of the two types of samples, benign and malignant, are as follows: Fig. 9 shown.

[0069] Fig.10 The validation effect diagram for evaluating the classifier trained by the intraoperative tumor margin benign and malignant assessment method provided in this application (using the scikit-learn library to create an instance of a random forest classifier (RandomForestClassifier), setting the number of decision trees built in the forest n_estimators=50, and using the standard criterion='entropy' when splitting the tree), from Fig.10It can be seen that the classifier trained using the intraoperative tumor margin benign and malignant assessment method provided in the present application has a good assessment effect.

[0070] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0071] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0072] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0073] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0074] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0075] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0076] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0077] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0078] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for evaluating the benign or malignant nature of tumor resection margins during surgery, characterized in that: include: Obtain optical coherence tomography data of historical resection margin tissue; Extracting an A-line reflectivity curve based on the optical coherence tomography data of the historical resection edge tissue, wherein the A-line reflectivity curve is a curve showing changes in image intensity with the depth of the A-line; Calculating and extracting target features based on the A-line reflectivity curve, the target features include depth features, slope features, distance features between local maxima, rapid intensity fluctuation features, frequency features, and optical attenuation coefficient features; Based on the target feature and the benign and malignant labels of the A-line reflectivity curve corresponding to the target feature, a preset classifier is trained to obtain a trained classifier; The target features extracted based on the optical coherence tomography data of the incision margin tissue to be measured are input into the trained classifier to obtain an evaluation result.

2. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The depth feature includes the penetration depth of the effective signal interval, and the process of extracting the penetration depth of the effective signal interval includes: Smoothing the current A-line reflectivity curve to obtain a current A-line reflectivity smoothing curve; Determine the effective signal interval of the current A-line reflectivity smoothing curve; The difference between the depths of the starting point and the end point of the effective signal interval is used as the penetration depth of the effective signal interval of the current A-line reflectivity curve.

3. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 2, characterized in that: The determining of the effective signal interval of the current A-line reflectivity smooth curve includes: The depth position corresponding to the first time the image intensity reaches the target value in the current A-line reflectivity smoothing curve is taken as the starting point of the effective signal interval; The part after the starting point of the effective signal interval of the current A-line reflectivity smooth curve is selected by preset sliding window. When the linear fitting slope of the signal in the window tends to 0 or the signal in the window is completely 0, the starting depth position in the current window is taken as the end point of the effective signal interval.

4. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The slope feature includes: an average slope of a valid signal interval, an overall slope of a valid signal interval, and an overall intercept of a valid signal interval. The process of extracting the slope feature includes: The effective signal interval of the current A-line reflectivity smooth curve is selected by sliding a preset sliding window to obtain multiple data subsets; Calculate the linear fitting slope corresponding to each data subset respectively, and take the average of the linear fitting slopes of all data subsets as the average slope of the effective signal interval of the current A-line reflectivity curve; Calculate the linear fitting slope and intercept of the effective signal interval of the current A-line reflectivity smooth curve, use the slope as the overall slope of the effective signal interval of the current A-line reflectivity curve, and use the intercept as the overall intercept of the effective signal interval of the current A-line reflectivity curve.

5. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The distance feature between local maxima includes: a distance mean and a distance standard deviation. The process of extracting the distance feature between local maxima includes: Determine the local maximum value of the image intensity in the valid signal interval of the current A-line reflectivity curve; Calculating the difference in depths corresponding to adjacent local maxima; The mean of all the differences is taken as the distance mean of the current A-line reflectivity curve, and the standard deviation of all the differences is taken as the distance standard deviation of the current A-line reflectivity curve.

6. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The rapid intensity fluctuation feature includes: a rapid intensity fluctuation standard deviation and a rapid intensity fluctuation peak value. The process of extracting the rapid intensity fluctuation feature includes: Smoothing the current A-line reflectivity curve to obtain a current A-line reflectivity smoothing curve; Determining a difference in rapid intensity fluctuations between the current A-line reflectivity curve and the current A-line reflectivity smooth curve; Calculating the standard deviation of the difference as the standard deviation of the intensity rapid fluctuation of the current A-line reflectivity curve; A difference curve is generated based on the difference, and a local maximum value with the largest absolute value in the difference curve is used as the intensity rapid fluctuation peak value of the current A-line reflectivity curve.

7. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The frequency characteristics include: weighted average of window frequency spectrum, main peak area of ​​window frequency spectrum, number of main peaks of window frequency spectrum and standard deviation of window frequency spectrum. The process of extracting the frequency characteristics includes: The current A-line reflectivity curve is selected by sliding a preset sliding window to obtain multiple groups of A-line reflectivity curve information; For each group of A-line reflectivity curve information, it is converted from the time domain to the frequency domain by fast Fourier transform to determine the frequency spectrum of each frequency point in the group, and the weighted average of the frequency spectrum of the current group of A-line reflectivity curve information is calculated by the following formula: Where k is the number of each frequency point; N is the total number of frequency components; f k is the frequency at the k frequency point; F (f k ) is f k The frequency spectrum corresponding to the frequency; The average of the weighted average frequency spectra of all groups of A-line reflectivity curve information is used as the weighted average of the window frequency spectrum of the current A-line reflectivity curve; For each group of A-line reflectivity curve information, determine the target frequency point in the group whose frequency spectrum is greater than the first preset threshold, and accumulate the frequency spectra corresponding to all the target frequency points in the group to obtain the main peak area of ​​the frequency spectrum of the current group of A-line reflectivity curve information; take the average of the main peak areas of the frequency spectrum of all groups of A-line reflectivity curve information as the main peak area of ​​the window frequency spectrum of the current A-line reflectivity curve; For each group of A-line reflectivity curve information, determine the number of frequency points in the group whose frequency spectrum is greater than the second preset threshold, and obtain the number of main peaks in the frequency spectrum of the current group of A-line reflectivity curve information; take the average of the number of main peaks in the frequency spectrum of all groups of A-line reflectivity curve information as the number of main peaks in the window frequency spectrum of the current A-line reflectivity curve; For each group of A-line reflectivity curve information, the standard deviation of the frequency spectrum in the group is calculated to obtain the frequency spectrum standard deviation of the current group of A-line reflectivity curve information; the average of the frequency spectrum standard deviations of all groups of A-line reflectivity curve information is used as the window frequency spectrum standard deviation of the current A-line reflectivity curve.

8. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The frequency characteristics also include the weighted average of the overall frequency spectrum, the main peak area of ​​the overall frequency spectrum, the number of main peaks of the overall frequency spectrum, and the standard deviation of the overall frequency spectrum. The process of extracting the frequency characteristics includes: All information of the current A-line reflectivity curve is converted from the time domain to the frequency domain through fast Fourier transform to determine the frequency spectrum of each frequency point, and the overall frequency spectrum weighted average of the current A-line reflectivity curve is calculated by the following formula: Where k is the number of each frequency point; N is the total number of frequency components; f k is the frequency at the k frequency point; F (f k ) is f k The frequency spectrum corresponding to the frequency; Determine a target frequency point whose frequency spectrum is greater than a first preset threshold value in all information of the current A-line reflectivity curve, and accumulate the frequency spectra corresponding to the target frequency point to obtain a main peak area of ​​the overall frequency spectrum of the current A-line reflectivity curve; Determine the number of frequency points whose frequency spectrum is greater than a second preset threshold value in all information of the current A-line reflectivity curve, and obtain the number of main peaks of the overall frequency spectrum of the current A-line reflectivity curve; The standard deviation of the frequency spectrum in all information of the current A-line reflectivity curve is calculated to obtain the overall frequency spectrum standard deviation of the current A-line reflectivity curve.

9. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 1, characterized in that: The optical attenuation coefficient-like features include: attenuation coefficient and attenuation coefficient amplitude. The process of extracting the optical attenuation coefficient-like features includes: The axial confocal diffusion function and additive background noise of the OCT system are introduced into the light attenuation model based on Beer's law, and the tissue optical attenuation model based on the OCT system is obtained. Performing nonlinear fitting on the optical coherence tomography data of the historical resection edge tissue by the least square method to determine the target parameters in the tissue optical attenuation model based on the OCT system; Wherein, the target parameters include the attenuation coefficient and the attenuation coefficient amplitude.

10. The method for evaluating the benign and malignant nature of tumor resection margins during surgery according to claim 9, characterized in that: The tissue optical attenuation model based on the OCT system is: Where I(z) is the light intensity at depth z; cf z is the depth of the focal plane from the tissue boundary; R is the Rayleigh length; A1 is the OCT signal amplitude; µ t is the attenuation coefficient; A2 is the additive background noise.

Images