A GPU-accelerated phase correction method for swept-source OCT imaging system

Abstract

This invention discloses a phase correction method for a GPU-accelerated swept-frequency OCT imaging system, belonging to the field of signal processing technology. The method constructs an SS-OCT system and acquires MZI reference signals and OCT interferometric signals; it locally upsamples the MZI signals, calculates the sub-pixel shift of each column of signals through threshold detection and reduction operations; it uses frequency-domain phase compensation to correct the nonlinear error of the OCT signal; and it achieves wavenumber domain resampling based on the Hilbert transform of the reference columns. This invention employs CUDA parallel computing to optimize the upsampling, frequency-domain shift, and phase unwrapping processes, and combines local computation strategies to reduce redundant operations, significantly reducing the processing time per frame and increasing the scanning rate to 2.48 times that of traditional methods. This improvement is significant and can meet the data processing requirements of higher-speed SS-OCT.

Description

Technical Field

[0001] This invention relates to a phase correction method for a GPU-accelerated swept-frequency OCT imaging system, belonging to the field of signal processing technology. Background Technology

[0002] Optical coherence tomography (OCT) is a high-resolution, non-contact optical imaging technique that can acquire the three-dimensional structure of objects. It is widely used in biomedical imaging fields such as ophthalmology (retina, cornea), cardiovascular medicine, and dermatology. Swept-source OCT (SS-OCT) uses a tunable laser as its light source, offering greater imaging depth, faster scanning speed, and a higher signal-to-noise ratio compared to previous generations of technology.

[0003] SS-OCT signals need to undergo a series of signal processing steps to be converted into diagnostic images. The most important step is wavenumber correction. However, compared to previous generations of technology, SS-OCT systems introduce greater phase noise, which affects the denoising results of SS-OCT imaging and subsequent applications that utilize phase information, such as blood flow imaging based on SS-OCT systems. Therefore, this is a major challenge for SS-OCT systems to perform real-time imaging.

[0004] Currently, various software and hardware methods have been developed to address this issue. Choi WJ et al. used a method of adding a fiber Bragg grating in front of a balanced detector to detect specific wavelengths. This method increases the complexity of the hardware, and because it uses hardware wavenumber sampling, it can only correct offsets that are multiples of 2π, which is not conducive to extending the imaging depth (Choi WJ, Potsaid B, Jayaraman V, et al. Phase-sensitive swept source OCT imaging of the human retina with a VCSEL light source[J]. Optics Letters,2013,38(3).DOI:10.1364 / ol.38.000338.). Braaf B et al. constructed a Mach-Zehnder interferometer (MZI) using another optical path and analyzed the phase space of the MZI signal using Hilbert transform. However, this method truncates the phase interval and performs a one-to-many Hilbert transform on the signal, which greatly increases the computational load (Braaf B et al. Phase-stabilized optical frequency domain imaging at 1-m for the measurement of blood flow in the human choroid[J]. Optics Express,2011,19(21):20886-20903.DOI:10.1364 / OE.19.020886.). Shangguan Ziwei et al. proposed using cross-correlation in the Hilbert transform domain to improve the rate, but this method did not solve the computational burden of performing fast Fourier transform (FFT) and phase unwrapping for each MZI signal. In addition, this method cannot correct sub-pixel level phase shifts (Shangguan Ziwei, Shen Yi, Li Peng, et al. Wavenumber correction and phase measurement of swept-frequency optical coherence tomography system [J]. Acta Physica Sinica, 2016(3):6.DOI:10.7498 / aps.65.034201.).Fan Jinyu et al. improved the cross-correlation method in the previous method and proposed to use time-domain cross-correlation instead of cross-correlation after Hilbert transform of the Mzi signal. However, in order to correct the sub-pixel level phase shift, the method uses frequency-domain upsampling, which also has a considerable amount of computation, which is not conducive to the real-time signal processing of higher-speed SS-OCT (Fan Jinyu, Gao Feng, Kong Wen, et al. Full-spectrum resampling method for multi-faceted rotating mirror laser sweeping optical coherence tomography system [J]. Acta Physica Sinica, 2017, 66(11):10. DOI:10.7498 / aps.66.114204.).

[0005] In summary, the various SS-OCT signal resampling and phase correction schemes proposed so far suffer from high system costs and large computational loads, which are not conducive to the construction of SS-OCT systems with faster scanning rates, deeper imaging depths, and higher signal-to-noise ratios. Summary of the Invention

[0006] To reduce system cost and computational load, increase scanning speed, and construct an SS-OCT system with deeper imaging depth and higher signal-to-noise ratio, this invention provides a phase correction method for a GPU-accelerated swept-frequency OCT imaging system. The technical solution is as follows:

[0007] Step 1: Build the SS-OCT hardware system, including a beam splitting module, MZI optical path, sample optical path and reference optical path. The swept laser source is split by the beam splitting module and then input into the MZI optical path, sample optical path and reference optical path respectively. The MZI optical path signal and the sample / reference optical path signal are received by the balanced detector and then transmitted to the host computer through the data acquisition card.

[0008] Step 2: For each frame of OCT image, select the first N points of the MZI reference signal in each column and perform linear upsampling interpolation.

[0009] Step 3: Set the MZI reference signal amplitude threshold Th, and obtain the coordinate index of each column of signal reaching the threshold;

[0010] Step 4: Calculate the sub-pixel level translation of each column of signal relative to the MZI reference signal:

[0011]

[0012] Where, m k m represents the coordinate index of the k-th column that reaches the threshold. reference This represents the coordinate index where the MZI reference signal reaches the threshold, L represents the upsampling factor, and Δt k This refers to the calculated subpixel-level translation amount;

[0013] Step 5: Perform translation compensation on each column of the original OCT signal in the frequency domain:

[0014]

[0015] in, Indicates the inverse Fourier transform, y k [n] represents the original OCT signal in the k-th column, y′ k [n] represents the original OCT signal in the k-th column after translation, where n is the signal position index, and N... ori It is the length of each column of the original OCT signal;

[0016] Step 6: Perform Hilbert transform and phase unwrapping on the MZI reference signal to obtain the calibration phase signal;

[0017] Step 7: Perform wavenumber domain resampling based on the inverse mapping relationship between the calibrated phase signal and the signal index n to obtain the corrected signal;

[0018] Step 8: Perform Fourier transform and post-processing on the corrected signal to generate an OCT image.

[0019] Optionally, step 2 employs a single linear upsampling interpolation:

[0020]

[0021] Where x[n] is the value of the original MZI reference signal, L is the upsampling factor, x[n′] is the upsampled MZI reference signal, and n′, n are the signal position indices.

[0022] Optionally, the expression for obtaining the phase signal in step 6 is:

[0023]

[0024] in This represents the Hilbert transform, where i represents the imaginary number. This represents the MZI signal after Hilbert transform, where Im represents taking the real part, Re represents taking the imaginary part, unwrap represents unwinding, and Δφ n This represents the final Hilbert phase.

[0025] Optionally, step 7 utilizes the Hilbert phase Δφ n Perform wavenumber domain calibration based on n and Δφ n The relationship between them leads to the conclusion that Δφ is equal to Δφ. n The index that needs to be transformed at intervals The relationship, here Δφ n Divide into equally spaced k1,…k N , obtain a new index Calculated The relationship between the two signals is used to interpolate the original signal:

[0026]

[0027] f -1 Represents n and Δφ n Inverse function between Indicates the interpolation method, y final [n] is the signal used for final processing.

[0028] A second objective of this invention is to provide a phase correction system for a swept-frequency OCT imaging system, the system comprising:

[0029] Spectrometer module, MZI optical path, sample optical path, and reference optical path;

[0030] Two balanced detectors are connected to the MZI optical path and the sample / reference optical path, respectively.

[0031] Data acquisition card, used to synchronously acquire signals output by the balance detector;

[0032] The host computer, configured with a CUDA-enabled GPU, is used to execute a phase correction method, the method comprising:

[0033] Step 1: Build the SS-OCT hardware system, including a beam splitting module, MZI optical path, sample optical path and reference optical path. The swept laser source is split by the beam splitting module and then input into the MZI optical path, sample optical path and reference optical path respectively. The MZI optical path signal and the sample / reference optical path signal are received by the balanced detector and then transmitted to the host computer through the data acquisition card.

[0034] Step 2: For each frame of OCT image, select the first N points of the MZI reference signal in each column and perform linear upsampling interpolation.

[0035] Step 3: Set the MZI reference signal amplitude threshold Th, and obtain the coordinate index of each column of signal reaching the threshold;

[0036] Step 4: Calculate the sub-pixel level translation of each column of signal relative to the MZI reference signal:

[0037]

[0038] Where, m k m represents the coordinate index of the k-th column that reaches the threshold. reference This represents the coordinate index where the MZI reference signal reaches the threshold, L represents the upsampling factor, and Δt k This refers to the calculated subpixel-level translation amount;

[0039] Step 5: Perform translation compensation on each column of OCT signals in the frequency domain:

[0040]

[0041] in, Indicates the inverse Fourier transform, y k [n] represents the original OCT signal in the k-th column, y′ k [n] represents the original OCT signal in the k-th column after translation, where n is the signal position index, and N... ori It is the length of each column of the original OCT signal;

[0042] Step 6: Perform Hilbert transform and phase unwrapping on the MZI reference signal to obtain the calibration phase signal;

[0043] Step 7: Perform wavenumber domain resampling based on the inverse mapping relationship between the calibrated phase signal and the signal index n to obtain the corrected signal;

[0044] Step 8: Perform Fourier transform and post-processing on the corrected signal to generate an OCT image.

[0045] Optionally, step 2 employs a single linear upsampling interpolation:

[0046]

[0047] Where x[n] is the value of the original MZI reference signal, L is the upsampling factor, x[n′] is the upsampled MZI reference signal, and n′, n are the signal position indices.

[0048] Optionally, the expression for obtaining the phase signal in step 6 is:

[0049]

[0050] in This represents the Hilbert transform, where i represents the imaginary number. This represents the MZI signal after Hilbert transform, where Im represents taking the real part, Re represents taking the imaginary part, unwrap represents unwinding, and Δφ n This represents the final Hilbert phase.

[0051] Optionally, step 7 utilizes the Hilbert phase Δφ n Perform wavenumber domain calibration based on n and Δφ n The relationship between them leads to the conclusion that Δφ is equal to Δφ. n The index that needs to be transformed at intervals The relationship, here Δφ n Divide into equally spaced k1,…k N, obtain a new index Calculated The relationship between the two signals is used to interpolate the original signal:

[0052]

[0053] f -1 Represents n and Δφ n Inverse function between Indicates the interpolation method, y final [n] is the signal used for final processing.

[0054] A third object of the present invention is to provide an electronic device including a memory and a GPU; the memory being used to store a computer program; and the GPU being used to implement the method as described in any of the preceding claims when the computer program is executed.

[0055] A fourth objective of the present invention is to provide a computer-readable storage medium storing a computer program that, when executed by a GPU, implements the method described in any of the preceding claims.

[0056] The beneficial effects of this invention are:

[0057] This invention employs GPU parallel acceleration, with key steps optimized using CUDA. Compared to traditional CPU serial computation, the computation time is significantly reduced. Furthermore, it only upsamples the first N points of the MZI reference signal (rather than the entire column of data), further reducing redundant computation. In addition, it quickly calculates the threshold coordinates through reduction operations, reducing the complexity of comparison computations. Experiments have shown that this invention has a faster processing speed, with the number of frames processed per second reaching more than 2.48 times that of existing methods, demonstrating a significant improvement and meeting the data processing requirements of higher-speed SS-OCT. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a structural block diagram of the sweep frequency OCT imaging phase correction system of the present invention.

[0060] Figure 2 This is a flowchart of the sweep frequency OCT imaging phase correction method of the present invention.

[0061] Figure 3This is a schematic diagram of the sampling offset of the MZI reference signal provided in Embodiment 1 of the present invention.

[0062] Figure 4 This is a schematic diagram of the corrected signal provided in Embodiment 1 of the present invention.

[0063] Figure 5 This is the corrected phase difference diagram provided in Embodiment 1 of the present invention.

[0064] Figure 6 This is a diagram showing the signal processing results before and after processing, as provided in Embodiment 1 of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0066] Example 1:

[0067] This embodiment provides a phase correction method for a GPU-accelerated swept-frequency OCT imaging system, including the following steps:

[0068] Step 1: First, set up as follows Figure 1 The SS-OCT hardware system shown in this embodiment adopts a classic SS-OCT system scheme, including: a beam splitting module, an MZI optical path, a sample optical path, and a reference optical path. The swept-frequency laser source is split into an MZI optical path, a sample optical path, and a reference optical path by the beam splitting module. The interference signal of the MZI optical path is received by a balanced detector and then input to a data acquisition card, and then transmitted to a host computer. The sample optical path and the reference optical path are received by another balanced detector, input to a data acquisition card, and then transmitted to a host computer.

[0069] The OCT reference signal (MZI reference signal) acquired in the MZI optical path must not exceed 1 / 10 of the sampling bandwidth of the acquisition card to reduce the influence of noise. The acquired MZI reference signal is as follows: Figure 3 As shown.

[0070] Step 2: Sample the MZI reference signal.

[0071] For each frame of an OCT image, assuming the number of columns is M, the MZI reference signal corresponding to each column can be obtained. The first N points in each column of the MZI reference signal are selected for upsampling (to save efficiency, it is not necessary to upsample the entire column of reference signal). In order to optimize the rate as much as possible, this embodiment uses a single linear upsampling interpolation:

[0072]

[0073] Where x[n] is the value of the original MZI reference signal, L is the upsampling factor, x[n′] is the upsampled signal, and n′, n are the indices of the corresponding positions. Upsampling can be optimized and accelerated by using texture memory in CUDA-enabled graphics cards; other upsampling methods, including but not limited to cubic spline interpolation, higher-order polynomial interpolation, and frequency-domain Fourier interpolation, can achieve higher accuracy, but will correspondingly sacrifice time efficiency.

[0074] Step 3: Set the threshold Th for the MZI reference signal amplitude.

[0075] The wavelength distribution of a swept-frequency laser is relatively stable and will not cause distortion of the interference signal. Taking advantage of this, the amplitude threshold Th is manually set to find the coordinate index of each individual MZI signal reaching the threshold:

[0076] m = {k|x[k] = Th}

[0077] To simplify the calculation, the Th value is set to the maximum value of the current signal. A CUDA-enabled graphics card is used, and the reduction operation method can significantly reduce the time required for this step.

[0078] Step 4: Use the obtained threshold coordinate index corresponding to the value {m k The subtraction of the signal with respect to the MZI reference signal is obtained by subtracting the signal from the MZI reference signal, where k = 1, 2, ..., M.

[0079]

[0080] Where Δt k It calculates the shift relative to the MZI reference signal, m. reference This is the coordinate index where the MZI reference signal reaches the threshold. CUDA can be used to maximize parallel computation to obtain the desired value.

[0081] Step 5: Use the translation amount Δt obtained in the previous step k The OCT signal in each column is shifted. In particular, for accurate signal shifting, this embodiment uses a frequency domain approach:

[0082]

[0083] in Indicates Fourier transform, Indicates the inverse Fourier transform, * indicates the dot product of discrete sequences, y k [n] represents the original discrete OCT signal in the k-th column, y′ k [n] represents the original OCT signal in the k-th column after translation, where n is the index of the signal in each column, and N... oriThis represents the original length of each signal column. Using a CUDA-enabled graphics card and writing code using the Cufft library can significantly reduce the required computation time.

[0084] Step 6: After shifting the signal, the standard OCT signal resampling and correction process can be performed, interpolating the time-domain sampled signal back to the wavenumber-domain sampled OCT signal. First, perform a Hilbert transform on the MZI reference signal and unwrap the phase to obtain the phase signal used for calibration.

[0085]

[0086]

[0087] in This represents the Hilbert transform, where i represents the imaginary number. This represents the MZI signal after Hilbert transform, where Im represents taking the real part, Re represents taking the imaginary part, unwrap represents unwinding to remove phase stacking caused by the 2π period, and Δφ. n This represents the final Hilbert phase.

[0088] Step 7: Use the obtained Hilbert phase signal to perform wavenumber domain calibration, based on n and Δφ. n The relationship between them leads to the conclusion that Δφ is equal to Δφ. n The index that needs to be transformed at intervals The relationship, here Δφ n Divide into equally spaced k1,…k N , obtain a new index Calculated The relationship between the two signals is used to interpolate the original signal:

[0089]

[0090] f -1 Represents n and Δφ n Inverse function between The interpolation method can be represented by polynomial fitting or cubic spline fitting to obtain the corresponding relationship, y final [n] is the signal obtained for final processing.

[0091] Step 8: Finally, calculate the obtained y final [n] Perform a Fourier transform, take the logarithmic transform domain, and normalize to obtain the components of each column in the OCT image used for display. The entire process can be run on a GPU using a CUDA-enabled graphics card.

[0092] To further illustrate the effects of the present invention, a series of related experiments and comparative analyses were conducted.

[0093] The experimental conditions are as follows:

[0094] An SS-OCT system with a sweep rate of 100kHz was built. A 1GHz sampling frequency acquisition card was used in the experiment, and the acquired OCT image size was 1024*8192. Based on the sweep rate, the acquisition time for each frame was approximately 10ms. The computing platform consisted of a 64-bit Windows operating system, an Intel i9 14900kf CPU, and an Nvidia GeForce RTX 4070 Super GPU with 12GB of video memory.

[0095] The collected signals are as follows Figure 3 As shown, there is a time delay between different signal sequences due to non-integer sampling intervals, which is the sub-pixel level sampling error described in this invention. The signal corrected using the phase correction method of this invention is shown below. Figure 4 As shown, alignment is basically achieved in the time domain, and analysis in the phase domain is as follows: Figure 5 As shown, the phase error is significantly reduced, with the standard deviation decreasing from 0.199 rad to 0.02 rad, which is sufficient for typical OCT signal processing.

[0096] Finally, the theoretical computational complexity and actual running time of the algorithm mentioned in this invention and the paper "Fan Jinyu, Gao Feng, Kong Wen, et al. Full-spectrum resampling method for multi-faceted rotating mirror laser sweeping optical coherence tomography system [J]. Acta Physica Sinica, 2017, 66(11):10. DOI:10.7498 / aps.66.114204." were analyzed. Assuming the upsampling factor is L, the size of an image is M*N, and the length of the truncated signal used for upsampling is K, if only the number of time-consuming multiplication operations is considered, as shown in Table 1:

[0097] Table 1 Comparative Analysis of the Two Algorithms

[0098]

[0099]

[0100] As can be seen from Table 1, the present invention requires less computation in theory and has a faster processing speed in practice, with the number of frames processed per second reaching more than 2.48 times, showing a significant improvement. This method can meet the data processing requirements of higher-speed SS-OCT.

[0101] Example 2:

[0102] This embodiment provides a phase correction system for a swept-frequency OCT imaging system, including:

[0103] Spectrometer module, MZI optical path, sample optical path, and reference optical path;

[0104] Two balanced detectors are connected to the MZI optical path and the sample / reference optical path, respectively.

[0105] Data acquisition card, used to synchronously acquire signals output by the balance detector;

[0106] The host computer, configured with a CUDA-enabled GPU, is used to execute a phase correction method, the method comprising:

[0107] Step 1: Build the SS-OCT hardware system, including a beam splitting module, MZI optical path, sample optical path and reference optical path. The swept laser source is split by the beam splitting module and then input into the MZI optical path, sample optical path and reference optical path respectively. The MZI optical path signal and the sample / reference optical path signal are received by the balanced detector and then transmitted to the host computer through the data acquisition card.

[0108] Step 2: For each frame of OCT image, select the first N points of the MZI reference signal in each column and perform linear upsampling interpolation.

[0109] Step 3: Set the MZI reference signal amplitude threshold Th, and obtain the coordinate index of each column of signal reaching the threshold;

[0110] Step 4: Calculate the sub-pixel level translation of each column of signal relative to the MZI reference signal:

[0111]

[0112] Where, m k m represents the coordinate index of the k-th column that reaches the threshold. reference This indicates the coordinate index where the MZI reference signal reaches the threshold, and L represents the upsampling factor;

[0113] Step 5: Perform translation compensation on each column of the original OCT signal in the frequency domain:

[0114]

[0115] in, Indicates the inverse Fourier transform, y k [n] represents the original OCT signal in the k-th column, y′ k [n] represents the original OCT signal in the k-th column after translation, where n is the signal position index, and N... ori It is the length of each column of the original OCT signal;

[0116] Step 6: Perform Hilbert transform and phase unwrapping on the MZI reference signal to obtain the calibration phase signal;

[0117] Step 7: Perform wavenumber domain resampling based on the inverse mapping relationship between the calibrated phase signal and the signal index n to obtain the corrected signal;

[0118] Step 8: Perform Fourier transform and post-processing on the corrected signal to generate an OCT image.

[0119] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A phase correction method for a swept-frequency OCT imaging system, characterized in that, The method accelerates execution on GPUs through CUDA parallel computing, including: Step 1: Build the SS-OCT hardware system, including a beam splitting module, MZI optical path, sample optical path and reference optical path. The swept laser source is split by the beam splitting module and then input into the MZI optical path, sample optical path and reference optical path respectively. The MZI optical path signal and the sample / reference optical path signal are received by the balanced detector and then transmitted to the host computer through the data acquisition card. Step 2: For each frame of OCT image, select the first N points of the MZI reference signal in each column and perform linear upsampling interpolation. Step 3: Set the MZI reference signal amplitude threshold Th, and obtain the coordinate index of each column of signal reaching the threshold; Step 4: Calculate the sub-pixel level translation of each column of signal relative to the MZI reference signal: wherein m k represents the coordinate index of the kth column reaching the threshold, m reference represents the coordinate index of the MZI reference signal reaching the threshold, and L represents the multiple of upsampling; Step 5: Perform translation compensation on each column of the original OCT signal in the frequency domain: in, Indicates the inverse Fourier transform, y k [n] represents the original OCT signal in the k-th column, y′ k [n] represents the original OCT signal in the k-th column after translation, where n is the signal position index, and N... ori It is the length of each column of the original OCT signal; Step 6: Perform Hilbert transform and phase unwrapping on the MZI reference signal to obtain the calibration phase signal; Step 7: Perform wavenumber domain resampling based on the inverse mapping relationship between the calibrated phase signal and the signal index n to obtain the corrected signal; Step 8: Perform Fourier transform and post-processing on the corrected signal to generate an OCT image.

2. The phase correction method for a swept-frequency OCT imaging system according to claim 1, characterized in that, Step 2 employs a single linear upsampling interpolation: Where x[n] is the value of the original MZI reference signal, L is the upsampling factor, x[n′] is the upsampled MZI reference signal, and n′, n are the signal position indices.

3. The phase correction method for a swept-frequency OCT imaging system according to claim 1, characterized in that, The expression for obtaining the calibrated phase signal in step 6 is as follows: in This represents the Hilbert transform, where i represents the imaginary number. This represents the MZI signal after Hilbert transform, where Im represents taking the real part, Re represents taking the imaginary part, unwrap represents unwinding, and Δφ n This represents the final Hilbert phase.

4. The phase correction method for a swept-frequency OCT imaging system according to claim 1, characterized in that, Step 7 utilizes the Hilbert phase Δφ n Perform wavenumber domain calibration based on n and Δφ n The relationship between them leads to the conclusion that Δφ is equal to Δφ. n The index that needs to be transformed at intervals The relationship, here Δφ n Divide into equally spaced k1,…k N , obtain a new index Calculated The relationship between the two signals is used to interpolate the original signal: f -1 Represents n and Δφ n Inverse function between Indicates the interpolation method, y final [n] is the signal used for final processing.

5. A phase correction system for a swept-frequency OCT imaging system, characterized in that, The system includes: Spectrometer module, MZI optical path, sample optical path, and reference optical path; Two balanced detectors are connected to the MZI optical path and the sample / reference optical path, respectively. Data acquisition card, used to synchronously acquire signals output by the balance detector; The host computer, configured with a CUDA-enabled GPU, is used to execute a phase correction method, the method comprising: Step 1: Build the SS-OCT hardware system, including a beam splitting module, MZI optical path, sample optical path and reference optical path. The swept laser source is split by the beam splitting module and then input into the MZI optical path, sample optical path and reference optical path respectively. The MZI optical path signal and the sample / reference optical path signal are received by the balanced detector and then transmitted to the host computer through the data acquisition card. Step 2: For each frame of OCT image, select the first N points of the MZI reference signal in each column and perform linear upsampling interpolation. Step 3: Set the MZI reference signal amplitude threshold Th, and obtain the coordinate index of each column of signal reaching the threshold; Step 4: Calculate the sub-pixel level translation of each column of signal relative to the MZI reference signal: Where, m k m represents the coordinate index of the k-th column that reaches the threshold. reference This represents the coordinate index of the MZI reference signal reaching the threshold, where L represents the upsampling factor, and Δt k This is the calculated subpixel-level translation amount; Step 5: Perform translation compensation on each column of the original OCT signal in the frequency domain: in, Indicates the inverse Fourier transform, y k [n] represents the original OCT signal in the k-th column, y′ k [n] represents the original OCT signal in the k-th column after translation, where n is the signal position index, and N... ori It is the length of each column of the original OCT signal; Step 6: Perform Hilbert transform and phase unwrapping on the MZI reference signal to obtain the calibration phase signal; Step 7: Perform wavenumber domain resampling based on the inverse mapping relationship between the calibrated phase signal and the signal index n to obtain the corrected signal; Step 8: Perform Fourier transform and post-processing on the corrected signal to generate an OCT image.

6. The phase correction system of the swept-frequency OCT imaging system according to claim 5, characterized in that, Step 2 employs a single linear upsampling interpolation: Where x[n] is the value of the original MZI reference signal, L is the upsampling factor, x[n′] is the upsampled MZI reference signal, and n′, n are the signal position indices.

7. The phase correction system of the swept-frequency OCT imaging system according to claim 5, characterized in that, The expression for obtaining the calibrated phase signal in step 6 is as follows: in This represents the Hilbert transform, where i represents the imaginary number. This represents the MZI signal after Hilbert transform, where Im represents taking the real part, Re represents taking the imaginary part, unwrap represents unwinding, and Δφ n This represents the final Hilbert phase.

8. The phase correction system of the swept-frequency OCT imaging system according to claim 5, characterized in that, Step 7 utilizes the Hilbert phase Δφ n Perform wavenumber domain calibration based on n and Δφ n The relationship between them leads to the conclusion that Δφ is equal to Δφ. n The index that needs to be transformed at intervals The relationship, here Δφ n Divide into equally spaced k1,…k N , obtain a new index Calculated The relationship between the two signals is used to interpolate the original signal: f -1 Represents n and Δφ n Inverse function between Indicates the interpolation method, y final [n] is the signal used for final processing.

9. An electronic device, characterized in that, It includes a memory and a GPU; the memory is used to store a computer program; the GPU is used to implement the method as described in any one of claims 1 to 4 when the computer program is executed.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by the GPU, implements the method as described in any one of claims 1 to 4.

Images