Endoscopic OCT elastography device and method

Abstract

The application discloses an endoscopic OCT elastography device, and relates to the technical field of functional imaging and image processing, and comprises a frequency domain OCT module, a proximal rotation scanning module, a pressure loading module and a calculation processing module. The application also discloses an endoscopic OCT elastography method, which comprises the following steps: S100, training a NURD correction model; S200, deploying the NURD correction model; S300, selecting a reference frame; S400, selecting a loading frame; S500, calculating the correlation degree of the reference frame and the loading frame; and S600, calculating a two-dimensional elastic distribution. The application improves the stability of OCT elastography, realizes more comprehensive evaluation of a target cavity, is beneficial to disease detection and grading, and is applicable to most commercial endoscopic OCT systems without increasing cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of functional imaging and image processing, and in particular to an endoscopic OCT elastography device and method. BACKGROUND

[0002] Elastography is a functional extension of medical imaging modalities used to assess the mechanical properties of tissues, the principle of which is that diseased tissues often exhibit different elastic properties compared to healthy tissues. By measuring the displacement or strain induced by an applied force, elastography can generate images representing the stiffness or elasticity of the examined tissue. The success of elastography lies in its ability to detect early-stage diseases, which often manifest as changes in tissue elasticity before structural changes become visible. Ultrasound and magnetic resonance-based elastography has shown high sensitivity and specificity in detecting and staging liver fibrosis, which is crucial for the treatment of chronic liver diseases. Ultrasound-based elastography has been shown to improve the differentiation of malignant and benign breast lesions.

[0003] However, ultrasound and magnetic resonance-based elastography have spatial resolutions ranging from hundreds of micrometers to millimeters, which face limitations in characterizing the mechanical properties of small lesions, such as atherosclerotic plaques that can form on the inner surface of blood vessels, thus monitoring is crucial as they have the potential to cause cardiovascular events. Limited resolution can hinder detailed assessment of plaque vulnerability, which is a key factor in predicting plaque rupture and subsequent complications such as heart attack and stroke.

[0004] On the other hand, optical coherence elastography (OCE), an optical coherence tomography (OCT)-based optical coherence elastography technique, inherits the resolution advantage of OCT and has been successfully applied in various medical environments. In the field of oncology, OCE has been shown to be able to distinguish between healthy and cancerous breast tissue by revealing heterogeneous mechanical contrast that cannot be observed with OCT alone. This is of great significance in improving the accuracy of tumor identification during surgery and may reduce the risk of residual tumors after breast-conserving surgery. In addition, OCE also shows promise in the field of ophthalmology, it has been used to measure the corneal elastic modulus of patients with keratoconus, a disease characterized by thinning and bulging of the cornea. By providing detailed tissue elasticity maps, OCE can help monitor the effectiveness of treatments such as collagen cross-linking, which aims to harden the cornea and stop the progression of the disease.

[0005] Endoscopic versions of OCE extend its applications beyond extracorporeal examination and ophthalmology, providing new dimensions for diagnosis and treatment inside the human body. To realize endoscopic OCE, a key technical challenge is to reliably extract the deformation caused by mechanical stimulation from the OCT time series. This is because, unlike free-space OCT optical systems that use a scanning galvanometer or a microelectromechanical system for beam scanning, OCT endoscopes based on fiber transport usually achieve beam scanning through the circumferential rotation and translation of the proximal end of the endoscopic OCT probe. Proximal rotation scanning refers to the placement of a high-speed motor system that rotates the endoscopic OCT probe outside the body, which can simplify the design complexity and reduce the size of the endoscopic OCT probe, and reduce the cost of clinical diagnostic applications. However, this proximal rotation scanning mode has a serious position instability problem (also known as non-uniform rotational disturbance, NURD), due to the following reasons: (1) mechanical friction during catheter bending during proximal rotation scanning, (2) rotation speed difference between the proximal and distal ends of the probe, and (3) synchronization error between image acquisition and scanning.

[0006] To solve the NURD problem of proximal rotation scanning to realize endoscopic OCE, various technical attempts have been made. M-mode imaging uses a static fiber probe to collect a time A-line sequence, which has been used to measure tissue deformation at only a single angle or lateral position [Sci. Reports, 2017.7, pp:4731]. Adjacent A-lines (non-adjacent B-frames) have been used to calculate tissue deformation [Phys. Med. & Biol, 2007.52, pp:2445]. This method requires the A-line rate of the OCT to be synchronized with the period of the applied mechanical stimulation. Another method to characterize the tissue stiffness of the lumen structure is to track the change in lumen shape, which minimizes the influence of NURD, but cannot obtain a two-dimensional strain distribution [Opt. Express, 2019.27, pp:16751-16766]. In addition, Wang et al. [Biomed. Opt. Express, 2022.13, pp:5418-5433] did not use proximal rotation scanning, but instead used distal rotation scanning of a double-layer micro motor to minimize the instability of beam scanning, thereby realizing intravascular OCE based on phase. However, such distal scanning imaging probes are still difficult to popularize in clinical applications due to their high cost.

[0007] Therefore, those skilled in the art are committed to developing an endoscopic OCT elastography device and method. SUMMARY

[0008] In view of the above defects of the prior art, the technical problem to be solved by the present application is to alleviate the non-uniform rotational disturbance caused by the proximal rotation scanning endoscopic OCT probe, and reliably extract the elastic deformation of the target lumen caused by pressure stimulation.

[0009] The inventor analyzes that, in order to realize endoscopic OCE, the key to realize in-vivo OCT imaging is to combine an imaging probe which is easy to bend with a catheter, an endoscope and a probe, miniaturize the imaging device and form an endoscopic OCT probe which can enter the human body. Compared with a micro motor arranged at the end of the probe to realize remote driving scanning, a near-end rotary scanning mode in which a driving source is placed outside the body to drive the whole probe to rotate has the advantages of low cost, miniaturization and safety, and in the clinic, the near-end rotary scanning mode still occupies a dominant position.

[0010] The inventor analyzes that, in order to realize endoscopic OCE, the key technical challenge is to reliably extract the deformation caused by mechanical stimulation from the OCT time sequence. However, the near-end rotary scanning mode has a position instability problem (also known as non-uniform rotary disturbance, abbreviated as NURD) due to factors such as friction and motor speed stability, which causes the A-line (single-beam scanning data in the depth direction) in the scanning image to be distorted and offset, making it difficult to obtain accurate elastic distribution information in the two-dimensional direction (radial and circumferential), and greatly affecting the elastic calculation imaging result.

[0011] The inventor extracts the real distortion vector based on the A-line corresponding to the near-end rotary scanning of the endoscopic OCT probe, and then constructs a reference frame-distortion frame pairing data with the known real distortion vector, which is used to train a data-driven NURD correction model. The NURD correction model trained based on the real distortion vector can align the A-line information corresponding to the same position imaging between different frames, improve the imaging stability, realize the accurate alignment and correction of the same position between the frames of the elastic deformation process imaging sequence, and perform elastic calculation on the NURD corrected sequence through the elastic calculation model, so as to obtain accurate elastic imaging results.

[0012] In one embodiment of the present application, an endoscopic OCT elastic imaging device is provided, comprising:

[0013] The frequency domain OCT module emits coherent light to the target cavity and receives a return light signal scattered by the target cavity, and converts the return light signal into an OCT signal;

[0014] The near-end rotary scanning module performs coherent light rotary scanning on the target cavity before and after the target cavity is subjected to pressure;

[0015] The pressure loading module applies pressure to the target cavity to cause elastic deformation of the target cavity;

[0016] The computing and processing module processes the OCT signal to obtain an OCT image sequence, performs real-time NURD correction on the OCT image sequence, performs correlation calculation and elastic calculation on the OCT images of the same imaging position before and after the target cavity is subjected to pressure, and obtains the time and spatial elastic distribution of the target cavity;

[0017] The frequency domain OCT module is coupled and physically connected with the proximal rotating scanning module, the pressure loading module is coupled and physically connected with the proximal rotating scanning module, and the frequency domain OCT module, the proximal rotating scanning module and the pressure loading module are respectively in communication connection with the computing processing module.

[0018] The frequency domain OCT module and the proximal rotating scanning module are started, the target cavity is subjected to proximal rotating scanning, the frequency domain OCT module receives a return light signal and converts it into an OCT signal, and sends the OCT signal to the computing processing module for processing, so as to obtain an OCT image sequence before pressure is applied and to perform real-time NURD correction; the pressure loading module is started, pressure is applied to the target cavity, the proximal rotating scanning module performs proximal rotating scanning on the target cavity, the frequency domain OCT module receives a return light signal and converts it into an OCT signal, and sends the OCT signal to the computing processing module for processing, so as to obtain an OCT image sequence after pressure is applied and to perform real-time NURD correction, the computing processing module performs correlation calculation and elasticity calculation on OCT image pairs at the same imaging position before and after pressure is applied to the target cavity, and obtains time and spatial elasticity distribution of the target cavity.

[0019] Optionally, in the endoscopic OCT elastography device in the above embodiment, the frequency domain OCT module comprises a coherent light source, a coupler and a light receiver.

[0020] Optionally, in the endoscopic OCT elastography device in any of the above embodiments, the proximal rotating scanning module comprises a fiber rotating connector, an endoscopic OCT probe and a protective catheter, the fiber rotating connector drives the endoscopic OCT probe to transmit in the protective catheter, and the target cavity is subjected to coherent light rotating scanning.

[0021] Optionally, in the endoscopic OCT elastography device in any of the above embodiments, the pressure loading module comprises a pressure controller, a pressure loader and a pressure transmission component, the pressure controller controls pressure output by the pressure loader, and the pressure transmission component is used to apply pressure to the target cavity.

[0022] Further, in the endoscopic OCT elastography device in the above embodiment, the pressure loading module is coupled and physically connected with the proximal rotating scanning module through an internal passage of the endoscopic OCT probe inserted into the pressure transmission component.

[0023] Further, in the endoscopic OCT elastography device in the above embodiment, the pressure loader uses a water pump or an air pump.

[0024] Further, in the endoscopic OCT elastography device in the above embodiment, the pressure transmission component uses a balloon or a catheter.

[0025] Optionally, in the endoscopic OCT elastography device in any of the above embodiments, the target cavity comprises a human body cavity.

[0026] Further, in the endoscopic OCT elastography device in the above embodiment, the body lumen includes a blood vessel, a trachea, and an esophagus.

[0027] Optionally, in the endoscopic OCT elastography device in any of the above embodiments, the pressure loading module can design different pressure loading schemes according to different target lumens.

[0028] Further, in the endoscopic OCT elastography device in the above embodiment, when the target lumen is a blood vessel, a gas pump or a water pump is used to inflate or fill the balloon with water to expand the balloon to load pressure on the blood vessel.

[0029] Further, in the endoscopic OCT elastography device in the above embodiment, when the target lumen is a trachea, a gas pump is used to inflate the trachea through a catheter to load pressure.

[0030] Further, in the endoscopic OCT elastography device in the above embodiment, when the target lumen is an esophagus, a water pump is used to fill the esophagus with a safe liquid through a catheter to load pressure.

[0031] Optionally, in the endoscopic OCT elastography device in any of the above embodiments, the computing processing module includes:

[0032] a signal processing unit, which processes the OCT signal, acquires an OCT image sequence before pressure loading and an OCT image sequence after pressure loading of the target lumen;

[0033] a NURD correction unit, which performs NURD correction on the OCT image sequence before pressure loading and the OCT image sequence after pressure loading;

[0034] a correlation calculation unit, which performs correlation calculation on a reference frame and a loaded frame image pair, judges whether a correlation threshold (greater than the correlation threshold) is met, the reference frame is an image after the OCT image before pressure loading is corrected by a NURD correction model, and the loaded frame is an image after the OCT image after pressure loading is corrected by the NURD correction model;

[0035] a preprocessing unit, which pre-processes the reference frame and the loaded frame image pair that meet the correlation threshold requirement;

[0036] an elasticity calculation unit, which performs elasticity calculation on the pre-processed reference frame and loaded frame image pair;

[0037] a post-processing unit, which post-processes the reference frame and loaded frame image pair after elasticity calculation to obtain the time and spatial elasticity distribution of the target lumen;

[0038] The signal processing unit, the NURD correction unit, the correlation calculation unit, the preprocessing unit, the elasticity calculation unit and the post-processing unit are sequentially and communicatively connected.

[0039] Further, in the endoscopic OCT elastography device in the above embodiment, the NURD correction unit is configured with a NURD correction model to perform real-time rotation distortion correction on the OCT image sequence obtained by the proximal rotation scanning, to restore the image distortion, distortion and distortion caused by the NURD phenomenon, and to improve the stability of the scanning results.

[0040] Further, in the endoscopic OCT elastography device in the above embodiment, the preprocessing operation includes:

[0041] Cavity foreground extraction, suppress background noise interference;

[0042] The image header and tail are respectively supplemented with a plurality of A lines at the tail and the head, and the image continuity during A line deformation elasticity calculation of the head and the tail is maintained.

[0043] Further, in the endoscopic OCT elastography device in the above embodiment, the elasticity calculation unit performs dense displacement and strain calculation along the radial and circumferential directions of the target cavity on the reference frame and the loaded frame image pair after NURD correction, to obtain the elasticity result image, i.e. the displacement and strain elasticity parameter distribution along the radial and circumferential directions of the cavity with time.

[0044] Further, in the endoscopic OCT elastography device in the above embodiment, the post-processing includes:

[0045] Crop, crop the elasticity result image corresponding to the plurality of A lines supplemented at the tail and the head in the preprocessing, to restore the original size;

[0046] Polar coordinate transformation, the result of the B scan mode is transformed into polar coordinates to obtain the elasticity result of the rotation scanning mode. The B scan mode refers to one-dimensional A line data obtained by sequentially arranging the two-dimensional image data of the cavity rotation scanning in one column.

[0047] In another embodiment of the present application, an endoscopic OCT elastography method is provided, comprising:

[0048] S100, training a NURD correction model, extracting a real distortion vector of a proximal rotation scanning module, applying to an OCT image, constructing a reference frame-distorted frame image with known real distortion vector, training a data-driven NURD correction model until the NURD correction model converges;

[0049] S200, deploying the NURD correction model, deploying the NURD correction model in the endoscopic OCT elastography device;

[0050] S300, selecting a reference frame, real-time NURD correction of the OCT sequence before pressure loading and selection of a reference frame, the endoscopic OCT elastography device acquires an OCT image sequence before pressure loading of the target lumen, real-time inference is performed through the NURD correction model to obtain a NURD corrected OCT image sequence before pressure loading, and a NURD corrected OCT image before pressure loading is selected as a reference frame;

[0051] S400, selecting a loading frame, real-time NURD correction of the OCT sequence after pressure loading while pressure is applied to the target lumen, and selecting a loading frame, the endoscopic OCT elastography device acquires an OCT image sequence after pressure loading of the target lumen, real-time inference is performed through the NURD correction model to obtain a NURD corrected OCT image sequence after pressure loading, and the NURD corrected OCT image after pressure loading is sequentially selected as a loading frame;

[0052] S500, reference frame-loading frame correlation calculation, different loading frames and the same reference frame are sequentially formed into a reference frame-loading frame image pair, correlation calculation is performed, and it is determined whether the correlation threshold is met;

[0053] S600, two-dimensional elastic distribution calculation, the reference frame and the loading frame image pair that meet the correlation threshold are sequentially subjected to elastic calculation according to the loading frame acquisition time sequence, and the displacement and strain elastic parameter distribution along the radial and circumferential directions of the lumen are obtained as the pressure loading changes.

[0054] Optionally, in the endoscopic OCT elastography method in the above embodiment, the real distortion vector of the endoscopic OCT probe is based on the offset of the A-line.

[0055] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, step S100 comprises:

[0056] S110, selecting an imaging phantom, using a phantom with circumferential angle characteristics and axial invariant properties as a target lumen for extracting a real distortion vector caused by NURD;

[0057] S120, phantom imaging and adjacent frame preprocessing, using the phantom to simulate target lumen path transmission, collecting B-scan mode images and preprocessing adjacent frames;

[0058] S130, trajectory feature segmentation, pixel-level labeling of part of the OCT image, training a binary classification image segmentation network, the purpose is to obtain the circumferential distribution trajectory feature of the phantom OCT image, and to distinguish the inner and outer regions of the trajectory, and to apply the binary classification image segmentation network after training and convergence to segment the adjacent frames of the OCT image, and to obtain a binary classification segmentation mask based on the inner and outer trajectories;

[0059] S140, mask sliding window and coincidence degree calculation, the binary classification segmentation mask corresponding to the adjacent frames of the OCT image is calculated based on the sliding window coincidence degree, and the potential distortion vector based on the A line between the adjacent frames of the OCT image is obtained;

[0060] S150, potential distortion vector filtering, the potential distortion vector is subjected to Gaussian filtering smoothing operation, abnormal value interference is reduced, and the real distortion vector is obtained;

[0061] S160, reconstructing the reference frame-distortion frame pairing data with the known real distortion vector, applying the real distortion vector to the arbitrary endoscopic OCT image, taking the OCT image before the pressure as the reference frame, rearranging the A line data according to the offset value of the same position of the real distortion vector corresponding to each A line position, and taking the rearranged A line data as the new endoscopic OCT image as the distortion frame;

[0062] S170, repeating S110-S160, generating the reference frame-distortion frame pairing data based on the real distortion vector, the real distortion vector as a known label, for training the data-driven NURD correction model until the NURD correction model converges.

[0063] Further, in the endoscopic OCT elastography method in the above embodiment, the phantom includes a square quartz tube.

[0064] Further, in the endoscopic OCT elastography method in the above embodiment, the preprocessing in S120 includes radial cutting of the phantom trajectory region, adjusting the image to a uniform size and contrast enhancement.

[0065] Further, in the endoscopic OCT elastography method in the above embodiment, step S140 includes:

[0066] S141, setting a reference window, the binary classification segmentation mask contains W A lines, each A line contains H pixels, and an Hxw size reference window is taken with the i-th A line as the center position in the n-1 frame (w

[0067] S142, calculating a mask sliding window, the same center position i and window size Hxw are taken in the n-th frame, and the sliding window operation from i-d / 2 to i+d / 2 is performed, wherein d represents the window sliding range, and the value of d is determined according to the maximum offset of the same position A line of the adjacent frames of the endoscopic OCT probe, and d / 2>the maximum offset;

[0068] S143, calculate the degree of coincidence, calculate the degree of coincidence D between the reference window and each mask sliding window, find the window with the highest degree of coincidence, the offset between the center position and i is the offset of the i-th A-line pair caused by NURD, the expression of the mask window coincidence calculation is as follows:

[0069]

[0070] Wherein, p i is the reference window in the n-1 frame with the i-th A-line as the center position, g j is the sliding window in the n frame with the j-th (j∈[i-d / 2, i+d / 2]) A-line as the center position; D(i,j) is the degree of coincidence of the reference window and the sliding window, which ranges from 0 (completely not coincident) to 1 (completely coincident);

[0071] S144, calculate the real distortion vector based on A-line, repeat S141-S143 to get the offset of each A-line as the potential distortion vector based on A-line between adjacent frames of OCT image.

[0072] Further, in the endoscopic OCT elastography method in the above embodiment, the parameters W=1024, H=512, w=32, and d=50 are set.

[0073] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, the pressure applied to the target cavity adopts an endoscopic OCT probe combined with a flexible surgical instrument, and one or more of a plurality of forms of pressure is directly or indirectly applied to the target cavity, so that the target cavity is elastically deformed.

[0074] Further, in the endoscopic OCT elastography method in the above embodiment, the flexible surgical instrument includes a catheter, a guide wire, a balloon, and a stent.

[0075] Further, in the endoscopic OCT elastography method in the above embodiment, the plurality of forms of pressure includes water pressure, air pressure, sound pressure, and light pressure.

[0076] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, the correlation degree calculation formula of the reference frame and the loaded frame in step S500 is as follows:

[0077]

[0078] Wherein, the pixel value of the reference frame is x i , the pixel value of the loaded frame is y i , wherein i=1, 2, …, m, and m is the number of image pixels; is the average value of the reference frame, is the average value of the loading frame; r is the correlation between the reference frame and the loading frame, and ranges from -1 to 1, r = 1 represents complete correlation, r = 0 represents no correlation, and r = -1 represents complete negative correlation.

[0079] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, the reference frame and the loading frame that meet the threshold requirement are preprocessed.

[0080] Further, in the endoscopic OCT elastography method in the above embodiments, the preprocessing includes:

[0081] Cavity foreground extraction suppresses background noise interference;

[0082] The image head and tail are respectively supplemented with a plurality of A lines at the tail and the head, so as to maintain the image continuity during the A line deformation elasticity calculation of the head and the tail.

[0083] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, the correlation threshold is set to 0.6, and when the correlation between the reference frame and the loading frame is greater than 0.6, it is judged that the correlation threshold is met, which indicates that the correlation is good, and the elasticity calculation result has reference significance.

[0084] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, the step S600 further includes post-processing the elasticity calculation result.

[0085] Further, in the endoscopic OCT elastography method in the above embodiments, the post-processing includes:

[0086] The elasticity result image is cropped to restore the original size corresponding to the A line position supplemented by the head and tail in the preprocessing operation;

[0087] The elasticity result of the B scan mode is subjected to polar coordinate transformation to obtain the elasticity result of the rotating scan mode, and the B scan mode refers to one-dimensional A line data obtained by rotating the cavity around the scan one time and arranged in columns to form two-dimensional image data.

[0088] Optionally, in the endoscopic OCT elastography method in any of the above embodiments, the elasticity calculation assumes that the image only has elastic deformation in a two-dimensional plane (i.e., circumferential and radial), and takes the speckle intensity and distribution as the calculation basis.

[0089] The present application extracts the real distortion vector of the proximal rotation scanning endoscopic OCT elastography device based on A-line, constructs the reference frame-distortion frame pairing data with known real distortion vector, trains the data-driven NURD correction model, more accurately aligns the A-line information at the same position between different frames, and improves the stability of the OCT elastography; the present application extracts the radial and circumferential displacement and strain elastic distribution of the target lumen, realizes more comprehensive evaluation of the target lumen, and is beneficial to disease detection and grading. The present application is suitable for most commercial endoscopic OCT systems without increasing the cost, assists doctors in making correct diagnosis to reduce the misdiagnosis rate, and reduces the medical expenditure of patients and the pressure of national medical insurance.

[0090] The concept, specific structure and technical effects of the present application will be further described below in combination with the drawings, so as to fully understand the purpose, features and effects of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0091] Figure 1 is a structural diagram of an endoscopic OCT elastography device according to an exemplary embodiment;

[0092] Figure 2 is a flowchart of an endoscopic OCT elastography method according to an exemplary embodiment;

[0093] Figure 3 is a quantitative result comparison diagram of a NURD correction network model trained based on real and virtual distortion vectors according to an exemplary embodiment;

[0094] Figure 4 is a qualitative comparison diagram of NURD correction by a model trained based on real and virtual distortion vectors according to an exemplary embodiment;

[0095] Figure 5 is an effect evaluation diagram of the influence of NURD correction on elastic displacement measurement according to an exemplary embodiment;

[0096] Figure 6 is an elastic distribution visualization diagram of the dynamic response of a blood vessel phantom during pressure application according to an exemplary embodiment; DETAILED DESCRIPTION

[0097] The present application can be embodied in many different forms, and the scope of protection of the present application is not limited to the embodiments described herein.

[0098] In the drawings, components of the same structure are denoted by the same reference numerals, and components similar in structure or function are denoted by similar reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the size and thickness of each component are not limited in the present application. In order to make the drawing clearer, the thickness of the components is appropriately exaggerated in some places in the drawing.

[0099] The inventor designs an endoscopic OCT elastography device, as shown in the accompanying drawings, comprising: Figure 1

[0100] A frequency domain OCT module emits coherent light to the target cavity and receives a return light signal scattered by the target cavity, and converts the return light signal into an OCT signal. The frequency domain OCT module comprises a coherent light source, a coupler, a light receiver, and a reference arm.

[0101] A proximal end rotation scanning module performs coherent light rotation scanning on the target cavity before and after applying pressure to the target cavity. The proximal end rotation scanning module comprises a fiber rotation connector, an endoscopic OCT probe, and a protective catheter. The fiber rotation connector rotates to drive the endoscopic OCT probe to move in the protective catheter, and performs coherent light rotation scanning on the target cavity.

[0102] A pressure loading module applies pressure to the target cavity to cause elastic deformation of the target cavity. The target cavity includes a human body cavity, such as a blood vessel, a trachea, and an esophagus. The pressure loading module comprises a pressure controller, a pressure loader, and a pressure transmission component. The pressure controller controls the pressure output by the pressure loader, and the pressure is applied to the target cavity through the pressure transmission component. The pressure loader uses a water pump or an air pump, and the pressure transmission component uses a balloon or a catheter. The pressure loading module is coupled and physically connected with the proximal end rotation scanning module through an endoscopic OCT probe inserted into an internal passage of the pressure transmission component. The pressure loading module can design different pressure loading schemes according to different target cavities. Specifically, when the target cavity is a blood vessel, an air pump or a water pump is used to inflate or fill water in the balloon to expand the balloon and apply pressure to the blood vessel; when the target cavity is a trachea, an air pump is used to inflate the trachea through a catheter to achieve pressure loading; and when the target cavity is an esophagus, a water pump is used to fill the esophagus with a safe liquid through a catheter to achieve pressure loading.

[0103] A computing and processing module processes the OCT signal to obtain an OCT image sequence, performs real-time NURD correction on the OCT image sequence, and performs correlation calculation and elasticity calculation on OCT images of the same imaging position before and after applying pressure to the target cavity to obtain the time and spatial elasticity distribution of the target cavity. The computing and processing module comprises:

[0104] A signal processing unit processes the OCT signal to obtain an OCT image sequence before pressure loading and an OCT image sequence after pressure loading. ​

[0105] NURD correction unit, NURD correction is performed on the OCT image sequence before loading and the OCT image sequence after pressure loading, and a NURD correction model is deployed in the NURD correction unit to perform real-time rotational distortion correction on the OCT image sequence obtained by proximal rotation scanning, to restore the image distortion, distortion and distortion caused by NURD phenomenon, and to improve the stability of the scanning result;

[0106] Correlation calculation unit, the correlation of the reference frame and the loading frame image pair is calculated, and it is judged whether the correlation threshold (greater than the correlation threshold) is satisfied, the reference frame is the image after the OCT image before pressure loading is corrected by the NURD correction model, and the loading frame is the image after the OCT image after pressure loading is corrected by the NURD correction model;

[0107] Preprocessing unit, the reference frame and the loading frame image pair meeting the correlation threshold requirement are preprocessed; the preprocessing operation includes:

[0108] Cavity foreground extraction, suppress background noise interference;

[0109] The image head and tail are respectively supplemented with a plurality of A lines at the tail and the head, and the image continuity during A line deformation elasticity calculation of the head and the tail is maintained.

[0110] Elasticity calculation unit, the reference frame and the loading frame image pair after preprocessing are subjected to elasticity calculation; the elasticity calculation unit is subjected to dense displacement and strain calculation along the radial and circumferential directions of the target cavity after the NURD correction of the reference frame and the loading frame image pair, to obtain the elasticity result image, i.e. the displacement and strain elasticity parameter distribution along the radial and circumferential directions of the cavity with time;

[0111] Post-processing unit, the reference frame and the loading frame image pair after elasticity calculation are subjected to post-processing to obtain the time and space elasticity distribution of the target cavity; the post-processing includes:

[0112] Crop, crop the elasticity result image corresponding to the plurality of A lines supplemented at the tail and the head in the preprocessing, to restore the original size;

[0113] Polar coordinate transformation, the result of B scan mode is subjected to polar coordinate transformation to obtain the elasticity result of rotation scanning mode, and the B scan mode refers to one-dimensional A line data obtained by rotating scanning of the cavity for one revolution arranged in columns to form two-dimensional image data.

[0114] The signal processing unit, the NURD correction unit, the correlation calculation unit, the preprocessing unit, the elasticity calculation unit and the post-processing unit are sequentially connected in communication.

[0115] The frequency domain OCT module is coupled and physically connected with the proximal rotating scanning module, the pressure loading module is coupled and physically connected with the proximal rotating scanning module, and the frequency domain OCT module, the proximal rotating scanning module and the pressure loading module are respectively in communication connection with the computing processing module;

[0116] The frequency domain OCT module and the proximal rotating scanning module are started, the target cavity is subjected to proximal rotating scanning, the frequency domain OCT module receives a return light signal and converts it into an OCT signal, and the OCT signal is sent to the computing processing module for processing, so that an OCT image sequence before pressure application is obtained and real-time NURD correction is performed; then the pressure loading module is started, pressure is applied to the target cavity, the proximal rotating scanning module performs proximal rotating scanning on the target cavity, the frequency domain OCT module receives a return light signal and converts it into an OCT signal, and the OCT signal is sent to the computing processing module for processing, so that an OCT image sequence after pressure application is obtained and real-time NURD correction is performed, and the computing processing module performs correlation calculation and elasticity calculation on OCT image pairs at the same imaging position before and after pressure application to the target cavity, so that the time and spatial elasticity distribution of the target cavity is obtained.

[0117] Based on the above embodiment, the inventor provides an endoscopic OCT elastography method, as shown in Figure 2 The method comprises the following steps:

[0118] S100, training a NURD correction model, extracting a real distortion vector of the proximal rotating scanning module, applying the real distortion vector to an OCT image, constructing a reference frame-distorted frame image with a known real distortion vector, training a data-driven NURD correction model until the NURD correction model converges; specifically comprising:

[0119] S110, selecting an imaging phantom, using a phantom with circumferential angle characteristics and axial invariance properties as a target cavity for extracting a real distortion vector caused by NURD, and the phantom

[0120] The square quartz tube comprises:

[0121] S120, phantom imaging and adjacent frame preprocessing, using the phantom to simulate target cavity path transmission, collecting images in B-scan mode and preprocessing adjacent frames, and the preprocessing comprises radially cropping a phantom trajectory area, adjusting the image to a uniform size and contrast enhancement;

[0122] S130, trajectory feature segmentation, pixel-level labeling of part of the OCT image, training a binary classification image segmentation network, the purpose being to obtain a circumferential distribution trajectory feature of the phantom OCT image, to distinguish the inner and outer regions of the trajectory, and to apply the trained and converged binary classification image segmentation network to segment adjacent frames of the OCT image, to obtain a binary classification segmentation mask based on the inner and outer regions of the trajectory;

[0123] S140, mask sliding window and coincidence degree calculation, the binary classification segmentation mask corresponding to the adjacent frames of the OCT image is calculated based on the sliding window coincidence degree, and the potential distortion vector based on the A line between the adjacent frames of the OCT image is obtained; specifically including:

[0124] S141, setting a reference window, the binary classification segmentation mask contains W A lines, each A line contains H pixels, and an Hxw size reference window is taken in the nth-1 frame with the ith A line as the center position (w

[0125] S142, calculating the mask sliding window, in the nth frame, the sliding window operation is performed from i-d / 2 to i+d / 2 with the same center position i and window size Hxw, wherein d represents the window sliding range, and the value of d is determined according to the maximum offset of the same position A line of the adjacent frames of the imaging endoscopic OCT probe, d / 2>the maximum offset;

[0126] S143, calculating the coincidence degree, calculating the coincidence degree D between the reference window and each mask sliding window, finding the window with the highest coincidence degree, and the offset Δ between the center position and i is the offset corresponding to the ith A line caused by NURD, and the expression of the mask window coincidence degree calculation is as follows:

[0127]

[0128] wherein, p i is the reference window with the ith A line as the center position in the nth-1 frame, g j is the sliding window with the jth (j∈[i-d / 2, i+d / 2]) A line as the center position in the nth frame; D(i,j) is the coincidence degree of the reference window and the sliding window, and its range is 0 (completely not coincident) to 1 (completely coincident);

[0129] S144, calculating the potential distortion vector based on the A line, repeating S141-S143 to obtain the offset of each A line as the potential distortion vector based on the A line between the adjacent frames of the OCT image.

[0130] S150, potential distortion vector filtering, performing Gaussian filtering smoothing operation on the potential distortion vector to reduce abnormal value interference, and obtaining the real distortion vector;

[0131] S160, reestablishing the reference frame-distortion frame pairing data with the known real distortion vector, applying the real distortion vector to the arbitrary endoscopic OCT image, the OCT image before pressure application as the reference frame, rearranging the A-line data according to the offset value of the same position of the real distortion vector corresponding to each A-line position, the rearranged A-line data forming a new endoscopic OCT image as the distortion frame;

[0132] S170, repeating S110-S160 to generate the reference frame-distortion frame pairing data based on the real distortion vector as the known label for training the data-driven NURD correction model until the NURD correction model converges.

[0133] S200, deploying the NURD correction model, deploying the NURD correction model to the endoscopic OCT elastography device;

[0134] S300, selecting a reference frame, real-time NURD correction of the pre-pressure loading OCT sequence and selecting a reference frame, the endoscopic OCT elastography device acquiring the OCT image sequence before the target lumen pressure loading, obtaining the NURD corrected pre-pressure loading OCT image sequence through real-time inference of the NURD correction model, and selecting a NURD corrected pre-pressure loading OCT image as a reference frame;

[0135] S400, selecting a loading frame, real-time NURD correction of the post-pressure loading OCT sequence while applying pressure to the target lumen and selecting a loading frame, applying pressure to the target lumen to cause elastic deformation, the endoscopic OCT elastography device acquiring the post-pressure loading OCT image sequence, obtaining the NURD corrected post-pressure loading OCT image sequence through real-time inference of the NURD correction model, and the corrected post-pressure loading OCT image as a loading frame in turn; applying pressure to the target lumen using an endoscopic OCT probe combined with a flexible surgical instrument, the flexible surgical instrument including a catheter, a guide wire, a balloon, and a stent, directly or indirectly applying one or more of a variety of pressures to the target lumen, causing the target lumen to elastically deform, the variety of pressures including water pressure, air pressure, sound pressure, and light pressure.

[0136] S500, reference frame-loading frame correlation calculation, sequentially forming reference frame-loading frame image pairs by combining different loading frames with the same reference frame, and performing correlation calculation to determine whether the correlation threshold is met; the correlation calculation formula between the reference frame and the loading frame is as follows:

[0137]

[0138] wherein the pixel value of the reference frame is x i , the pixel value of the loading frame is y i , wherein i=1, 2, …, m, and m is the number of image pixels; is a reference frame average value, is a loading frame average value; r is a correlation degree between the reference frame and the loading frame, and the value range of r is between -1 and 1, r = 1 represents complete correlation, r = 0 represents no correlation, r = -1 represents complete negative correlation, and the correlation degree threshold is set to 0.6. When the correlation degree between the reference frame and the loading frame is greater than 0.6, it is judged that the correlation degree threshold is met, which represents good correlation, and the elastic calculation result has reference significance. The reference frame and the loading frame meeting the threshold requirement are preprocessed, and the preprocessing includes:

[0139] cavity foreground extraction, background noise interference is suppressed;

[0140] The image header and the tail part are respectively supplemented with a plurality of A lines at the tail and the header, and the image continuity in the A line shape change elastic calculation of the header and the tail is maintained.

[0141] S600, two-dimensional elastic distribution calculation, the reference frame and the loading frame image pair meeting the correlation degree threshold are sequentially subjected to elastic calculation according to the loading frame acquisition time sequence. The elastic calculation assumes that the image only has elastic deformation in the two-dimensional plane (i.e. circumferential and radial), and obtains the displacement and strain elastic parameter distribution along the cavity radial and circumferential direction with the change of pressure loading, according to the speckle intensity size and distribution. The elastic calculation result is post-processed, including:

[0142] cropping the elastic result image, corresponding to the A line position supplemented at the header and the tail in the preprocessing operation, the original size is restored;

[0143] The elastic result of the B scan mode is subjected to polar coordinate transformation to obtain the elastic result of the rotary scan mode. The B scan mode refers to one-dimensional A line data obtained by sequentially rotating the cavity for one revolution and arranged in columns to form two-dimensional image data.

[0144] To verify the effect of the above-mentioned endoscopic OCT elastography device and method, the inventors collected 20,000 real distortion vectors from the experimental endoscopic OCT system, constructed the corresponding 20,000 reference frame-distortion frame pairing data to train the NURD correction network, and conducted experiments on the correction network based on the CNN (Med. Image Anal, 2022, 77, pp: 102355) and Transformer (Biomed. Opt. Express, 2024, 15, pp: 319-335) architectures. The results before NURD correction, the correction results of the NURD correction model trained based on real distortion vectors, and the correction results of the NURD correction model trained based on virtual distortion vectors were compared, wherein the virtual distortion vectors are obtained from the non-training NURD correction algorithm on a large number of OCT sequences, which is a common way to obtain NURD distortion vectors, and the cost of obtaining is to collect a large number of endoscopic OCT image sequences in advance. Since there is a gap between the extracted distortion results and the real values, the virtual distortion vectors are named.

[0145] The NURD correction quantitative results of the two correction networks trained based on real and virtual distortion vectors are shown in FIGS. 8 and 9. Figure 3 As can be seen from the two box plots, the correction results based on real distortion vectors are better than those based on virtual distortion vectors on the correction models of the two popular architectures, and the accuracy of the image distortion correction results can improve the elastic calculation results of the elastic imaging. Figure 4 The qualitative results of different distortion vectors trained based on the Transformer architecture are shown. The OCT imaging target is a square quartz tube, and the endoscopic square tube structure gradually distorts during scanning and acquisition before correction. After NURD correction, the result is closer to the real structure of the imaged square tube compared with before correction. The result based on real distortion vectors is more accurate than that based on virtual distortion vectors.

[0146] To verify the effect of the endoscopic NURD phenomenon on the elastic displacement measurement, a static endoscopic blood vessel phantom image sequence was continuously collected. As shown in FIG. 10, Figure 5 When the static sample collected by the endoscopic OCT probe is calculated by the elastic calculation model, the circumferential direction should be 0 displacement in the ideal case, but there is a large displacement error (see the strip mapping and the digital value representing the average displacement) due to the NURD phenomenon in the actual acquisition process, especially for the circumferential displacement, which brings side effects to the elastic calculation result. After NURD correction, the displacement error is significantly reduced, which is more conducive to reducing the final elastic parameter calculation error.

[0147] The inventors set the target lumen as a blood vessel, adopt the combination of an endoscopic OCT probe and a balloon catheter, use a gas pump to inflate the balloon to make it expand under pressure, i.e., to apply air pressure to give the blood vessel wall radial pressure to make it elastically deform. Figure 6 As shown in FIG. 2, a relatively hard plaque part is simulated in the blood vessel, which is depicted by a white line in the figure. Elastic calculation is performed between the reference frame and the loaded frames in the process of balloon expansion, and the images are collected by the endoscopic OCT system with proximal rotation scanning, with an interval of 0.2s between the loaded frames. As can be seen, with the expansion of the balloon, the radial outward displacement gradually increases except in the plaque simulation area, and the corresponding radial strain indicates that the normal blood vessel part of the blood vessel phantom other than the hard plaque is gradually compressed. As for the circumferential displacement, due to the irregular shape of the blood vessel lumen, the displacement is unevenly distributed, but it can be seen that the displacement also gradually increases with the expansion of the balloon, and the corresponding circumferential strain also shows a similar trend.

[0148] Thus, the mechanical performance of the lumen tissue is obtained by the circumferential and radial displacement and strain elastic distribution obtained by elastic calculation, and the tissue stiffness is distinguished, thereby providing clinical auxiliary diagnosis for the detection of tissue lesions.

[0149] The above describes the preferred embodiments of the present application in detail. It should be understood that those skilled in the art can make many modifications and changes without creative labor according to the concept of the present application. Therefore, any technical solution that can be obtained by logical analysis, reasoning or limited experiment on the basis of the prior art according to the concept of the present application should be within the protection scope defined by the claims.

Claims

1. An endoscopic OCT elastography apparatus, characterized by, The application relates to a time and space elasticity distribution measurement system for a target cavity, which comprises the following modules: a frequency domain OCT module which emits coherent light to a target cavity and receives a return light signal scattered by the target cavity, and converts the return light signal into an OCT signal; a proximal end rotating scanning module which performs coherent light rotating scanning on the target cavity before and after pressure is applied to the target cavity; a pressure loading module which applies pressure to the target cavity to make the target cavity elastically deform; a calculation processing module which processes the OCT signal to obtain an OCT image sequence, performs real-time NURD correction on the OCT image sequence, and performs correlation calculation and elasticity calculation on OCT image pairs at the same imaging position before and after pressure is applied to the target cavity to obtain time and space elasticity distribution of the target cavity; the frequency domain OCT module and the proximal end rotating scanning module are coupled and physically connected through an optical fiber, the pressure loading module and the proximal end rotating scanning module are coupled and physically connected, and the frequency domain OCT module, the proximal end rotating scanning module and the pressure loading module are respectively communicatively connected to the calculation processing module; the calculation processing module comprises: a signal processing unit which processes the OCT signal to obtain an OCT image sequence before pressure is applied to the target cavity and an OCT image sequence after pressure is applied to the target cavity; a NURD correction unit which performs NURD correction on the OCT image sequence before pressure is applied to the target cavity and the OCT image sequence after pressure is applied to the target cavity; a correlation calculation unit which performs correlation calculation on reference frame and loading frame image pairs to determine whether a correlation threshold is met; a preprocessing unit which pre-processes the reference frame and loading frame image pairs meeting the correlation threshold requirement; an elasticity calculation unit which performs elasticity calculation on the pre-processed reference frame and loading frame image pairs; a post-processing unit which post-processes the reference frame and loading frame image pairs after elasticity calculation to obtain time and space elasticity distribution of the target cavity; the signal processing unit, the NURD correction unit, the correlation calculation unit, the preprocessing unit, the elasticity calculation unit and the post-processing unit are sequentially communicatively connected; the NURD correction unit is provided with a NURD correction model which performs real-time rotating distortion correction on the OCT image sequence obtained through proximal end rotating scanning to restore image distortion, distortion and distortion caused by NURD phenomenon; the NURD correction model is trained, a real distortion vector of the proximal end rotating scanning module is extracted and applied to the OCT image to construct a reference frame-distortion frame image with a known real distortion vector, and the NURD correction model based on data driving is trained until the NURD correction model converges.

2. The endoscopic OCT elastography apparatus according to claim 1, wherein the frequency domain OCT module comprises a coherent light source, a coupler and a light receiver, and a reference arm.

3. The endoscopic OCT elastography apparatus according to claim 1, wherein the proximal end rotating scanning module comprises an optical fiber rotating connector, an endoscopic OCT probe and a protective catheter, the optical fiber rotating connector rotates to drive the endoscopic OCT probe to transmit in the protective catheter to perform coherent light rotating scanning on the target cavity.

4. The endoscopic OCT elastography apparatus according to claim 1, wherein The pressure loading module comprises a pressure controller, a pressure loader, and a pressure transmission component, wherein the pressure controller controls the pressure output by the pressure loader and applies pressure to the target cavity through the pressure transmission component.

5. An endoscopic OCT elastography method using the endoscopic OCT elastography apparatus according to any one of claims 1 to 4, characterized by, The method comprises the following steps: S100, training a NURD correction model, extracting a real distortion vector of the proximal rotation scanning module, applying the real distortion vector to an OCT image, constructing a reference frame-distortion frame image with a known real distortion vector, training a data-driven NURD correction model until the NURD correction model converges; S200, deploying the NURD correction model, deploying the NURD correction model in the endoscopic OCT elastography device; S300, selecting a reference frame, real-time NURD correction of an OCT sequence before pressure loading and selecting a reference frame, the endoscopic OCT elastography device acquires an OCT image sequence before pressure loading of the target cavity, obtains a NURD-corrected OCT image sequence before pressure loading through real-time inference of the NURD correction model, and selects a NURD-corrected OCT image before pressure loading as a reference frame; S400, selecting a loading frame, real-time NURD correction of an OCT sequence after pressure loading while applying pressure to the target cavity and selecting a loading frame, the target cavity is elastically deformed after pressure loading, the endoscopic OCT elastography device acquires an OCT image sequence after pressure loading, obtains a NURD-corrected OCT image sequence after pressure loading through real-time inference of the NURD correction model, and the NURD-corrected OCT image after pressure loading is sequentially selected as a loading frame; S500, reference frame-loading frame correlation calculation, sequentially forming a reference frame-loading frame image pair by combining different loading frames with the same reference frame, performing correlation calculation, and determining whether the correlation threshold is met; S600, two-dimensional elastic distribution calculation, sequentially performing elastic calculation on the reference frame and the loading frame image pair that meets the correlation threshold according to the loading frame acquisition time sequence, and obtaining displacement and strain elastic parameter distribution along the radial and circumferential directions of the cavity with pressure loading.

6. The endoscopic OCT elastography method of claim 5, wherein, The step S100 comprises: S110, selecting an imaging phantom, using a phantom with circumferential angle characteristics and axial invariance properties as the target cavity for extracting a real distortion vector caused by NURD; S120, phantom imaging and adjacent frame preprocessing, using the phantom to simulate the target cavity path transmission, collecting images in B-scan mode, and preprocessing adjacent frames; S130, trajectory feature segmentation, pixel-level labeling of part of the OCT image, training a binary classification image segmentation network, obtaining the circumferential distribution trajectory feature of the phantom OCT image, distinguishing the inner and outer regions of the trajectory, and applying the trained and converged binary classification image segmentation network to segment adjacent frames of the OCT image to obtain a binary classification segmentation mask based on the inner and outer trajectories. S140, Mask sliding window and coincidence degree calculation, the binary segmentation mask corresponding to the adjacent frame of the OCT image is calculated based on the sliding window coincidence degree, and the potential distortion vector based on A line between the adjacent frames of the OCT image is obtained; S150, potential distortion vector filtering, the potential distortion vector based on A line is subjected to Gaussian filtering smoothing operation, abnormal value interference is reduced, and a real distortion vector is obtained; S160, reconstructing the reference frame-distortion frame pairing data with known real distortion vector, applying the real distortion vector to any endoscopic OCT image, taking the OCT image before applying pressure as the reference frame, rearranging the A line data according to the offset value size of the same position of the real distortion vector corresponding to each A line position, and the rearranged A line data constitutes a new endoscopic OCT image as the distortion frame; S170, repeating S110-S160, generating reference frame-distortion frame pairing data based on the real distortion vector, the real distortion vector as a known label, training a data-driven NURD correction model until the NURD correction model converges.

7. The endoscopic OCT elastography method of claim 6, wherein, The step S140 comprises: S141, setting a reference window, the binary segmentation mask comprising W A lines, each A line comprising H pixels, in the i-th frame taking the i-th A line as a center position n-1 pixels, in the i-th frame taking the i-th A line as a center position i pixels, in the i-th frame taking the i-th A line as a center position Hxw pixels, in the i-th frame taking the i-th A line as a center position w < W pixels, in the i-th frame taking the i-th A line as a center position S142, calculate a mask sliding window, in the first n frame with the same center position i and window size Hxw perform sliding window operation from i-d / 2 to i+d / 2 ; S143. Calculate the overlap: Calculate the overlap between the reference window and each mask sliding window. D Find the window with the highest overlap, whose center position is... i The offset Δ between them is the first caused by NURD. i The expression for calculating the offset corresponding to each line A and the overlap of the mask window is as follows: ； S144, calculating the real distortion vector based on A line, repeating S141-S143 to obtain the offset of each A line as the potential distortion vector based on A line between the adjacent frames of the OCT image.

8. The endoscopic OCT elastography method of claim 7, wherein, The correlation degree calculation formula of the reference frame and the loaded frame in the step S500 is as follows: 。

Images