A photoacoustic imaging system based on self-learning

Abstract

The present invention relates to the field of photoacoustic imaging technology, and in particular to a self-learning-based photoacoustic imaging system comprising: an acquisition module, a screening module, an actual generation module, an ideal generation module, an adjustment module, and an imaging module. The present invention achieves dynamic control of photoacoustic imaging quality by constructing a complete closed-loop process of acquisition, screening, evaluation, self-learning optimization, and image reconstruction. The spectral texture gradient and imaging uniformity error jointly define the spatial consistency of the image; the background noise energy level and laser energy distribution characterize the stability of the signal quality; the actual clarity index measures the current system performance, while the ideal clarity index uses a self-learning model to extract the potential imaging upper limit under noise interference; and the adjustment module adaptively optimizes through the comparison results, effectively solving the problem of low recognition accuracy and slow response speed caused by over-reliance on fixed parameter models, resulting in poor adaptability to different tissues.

Description

Technical Field

[0001] The present invention relates to the technical field of photoacoustic imaging, and in particular to a photoacoustic imaging system based on self-learning. Background Art

[0002] With the continuous development of medical imaging technology, the demand for non-invasive, high-resolution imaging methods in clinical diagnosis is increasing. As an emerging technology that combines the advantages of optics and ultrasound, photoacoustic imaging has attracted widespread attention due to its high contrast and good penetration in deep tissue imaging. However, faced with the complex and changing human tissue environment and diverse clinical application needs, existing imaging technologies still face many challenges in terms of image quality, tissue resolution, and real-time response. There is an urgent need to achieve technological breakthroughs in system modeling, image reconstruction methods, and intelligent control mechanisms.

[0003] Patent document CN116784798A discloses a photoacoustic imaging system, which includes: a light source, a microring resonator detector array, and a signal processing unit; the microring resonator detector array includes at least two detector arrays, each detector array includes at least two microring resonators; wherein, the light source is used to emit a laser signal, each microring resonator is used to generate a corresponding resonant signal based on the laser signal, and the signal processing unit is used to output an image signal corresponding to each resonant signal.

[0004] It can be seen that the photoacoustic imaging system has the following problems: the system relies on a continuous-wave laser, which poses a risk of high power consumption; the system relies on a fixed microring array structure, making it difficult to achieve flexible three-dimensional imaging configuration and adapt to different tissue structures; the system is highly dependent on the coupling efficiency and stability of the optical fiber and coupler, and the imaging quality is easily affected by external disturbances. Summary of the Invention

[0005] To this end, the present invention provides a self-learning-based photoacoustic imaging system, which is used to overcome the problems in the prior art of poor adaptability to different tissues due to over-reliance on fixed parameter models, resulting in low recognition accuracy and slow response speed through multimodal fusion image reconstruction and dynamic threshold adjustment.

[0006] To achieve the above objectives, the present invention provides a self-learning-based photoacoustic imaging system, comprising:

[0007] An acquisition module is used to collect photoacoustic signals, laser energy distribution, spectrum texture gradient, imaging uniformity error, and background noise energy level in each processing square in real time from the target tissue divided by the preset square length based on the preset pulse repetition frequency;

[0008] a screening module connected to the acquisition module, configured to screen out a plurality of distorted squares from all the processing squares based on the spectral texture gradient, the imaging uniformity error, and a preset distortion threshold, and to screen out a plurality of noise source squares from all the distorted squares based on the background noise energy level and the laser energy distribution in each distorted square;

[0009] an actual generation module, connected to the acquisition module and the screening module respectively, for generating an actual clarity index according to the imaging uniform error of each noise source grid;

[0010] an ideal generation module connected to the acquisition module, configured to input the photoacoustic signals of all the noise source grids into a preset self-learning model to generate an ideal clarity index;

[0011] an adjustment module, connected to the actual generation module and the ideal generation module, respectively, for adjusting the preset pulse repetition frequency and the preset distortion threshold according to the actual clarity index and the ideal clarity index to obtain a first adjustment parameter set, or adjusting the preset pulse repetition frequency and the preset grid side length to obtain a second adjustment parameter set;

[0012] An imaging module is connected to the acquisition module and is used to reconstruct and output a photoacoustic image based on the photoacoustic signal re-acquired based on the first adjustment parameter set or the second adjustment parameter set.

[0013] Furthermore, the acquisition module includes:

[0014] a laser excitation unit, configured to emit nanosecond laser pulses at the preset pulse repetition frequency and irradiate all the processing grids of the target tissue to obtain a plurality of the photoacoustic signals;

[0015] A feature acquisition unit for real-time acquisition of the pulse energy value, center frequency, pixel grayscale average, pixel grayscale standard deviation, and energy value of the acoustic baseline signal during a preset laser-free triggering period within each processing grid;

[0016] A feature calculation unit is connected to the feature acquisition unit, and is used to calculate the energy distribution of each laser according to the pulse energy value, and to calculate the spectrum texture gradient according to the center frequency, and to calculate the imaging uniformity error according to the pixel grayscale average and the pixel grayscale standard deviation, and to calculate the background noise energy level according to the energy value of the acoustic baseline signal.

[0017] Furthermore, the feature calculation unit includes:

[0018] an energy distribution calculation subunit, for calculating an average value of all the pulse energy values ​​to obtain the laser energy distribution;

[0019] a frequency calculation subunit, configured to calculate the maximum difference in center frequencies among adjacent processing squares selected based on a preset neighborhood size with each processing square as the center, to obtain the spectral texture gradient;

[0020] a uniform error calculation subunit, for calculating the ratio of the pixel grayscale standard deviation to the pixel grayscale average value, the imaging uniform error;

[0021] The energy level calculation subunit is used to calculate the noise energy level of the energy value of the acoustic baseline signal to obtain the background noise energy level.

[0022] Furthermore, the screening module includes:

[0023] a screening fluctuation calculation submodule, configured to calculate the standard deviation of the spectral texture gradient at each moment from the initial moment to a preset distortion screening duration to obtain a plurality of gradient fluctuation values, and to calculate the standard deviation of the imaging uniform error at each moment from the initial moment to a preset distortion screening duration to obtain a plurality of error fluctuation values;

[0024] a distortion screening submodule, connected to the screening fluctuation calculation submodule, for screening out a number of distorted squares according to all the gradient fluctuation values ​​and all the error fluctuation values;

[0025] The noise source screening submodule is connected to the distortion screening submodule and is used to screen out a number of the noise source grids from all the distortion grids according to the background noise energy level and the laser energy distribution in each of the distortion grids.

[0026] Furthermore, the distortion screening submodule includes:

[0027] A screening normalization unit is used to normalize all the gradient fluctuation values ​​to obtain a gradient fluctuation normalized set, and to normalize all the error fluctuation values ​​to obtain an error fluctuation normalized set;

[0028] a correlation calculation unit connected to the screening normalization unit, for calculating a correlation coefficient between the gradient fluctuation normalization set and the error fluctuation normalization set to obtain a distortion correlation;

[0029] The distortion screening unit is connected to the correlation calculation unit and is used to determine that the processing grid is the distorted grid when the distortion correlation is greater than the preset distortion threshold, so as to screen out a number of distorted grids from all the processing grids.

[0030] Furthermore, the noise source screening submodule includes:

[0031] a candidate square determining unit, configured to determine, when the background noise energy level is greater than a preset noise threshold, that the distorted square is the candidate square, so as to determine a plurality of candidate squares;

[0032] a neighborhood mean calculation unit connected to the candidate square determination subunit, configured to calculate the average value of the laser energy distribution of all other candidate squares within the preset neighborhood size centered on the single candidate square, to obtain a neighborhood mean;

[0033] A noise source screening unit is connected to the neighborhood mean calculation unit and is used to screen out a number of noise source squares from all candidate squares according to the laser energy distribution of each candidate square and the corresponding neighborhood mean.

[0034] Furthermore, the noise source screening unit includes:

[0035] a distribution deviation calculation subunit, configured to calculate an absolute value of a relative deviation between the laser energy distribution and the neighborhood mean, to obtain a distribution deviation value;

[0036] The noise source screening subunit is connected to the distribution deviation calculation subunit and is used to determine that the candidate square is the noise source square when the distribution deviation value is greater than a preset distribution deviation threshold, so as to screen out a number of the noise source squares from all the distorted squares.

[0037] Furthermore, the actual generation module includes:

[0038] an error normalization unit, configured to normalize the imaging uniform errors of all the noise source grids to obtain a number of noise source error normalization values;

[0039] The actual index calculation unit is connected to the error normalization unit and is used to perform weighted sum calculation on the error normalization value of each noise source and its corresponding preset position weight to generate the actual clarity index.

[0040] Furthermore, the adjustment module includes:

[0041] an index deviation calculation unit, configured to calculate an absolute value of a relative deviation between the actual clarity index and the ideal clarity index when the actual clarity index is greater than the ideal clarity index, to obtain an index deviation;

[0042] An adjustment unit is connected to the exponential deviation calculation unit and is used to adjust the preset pulse repetition frequency and the preset distortion threshold according to the exponential deviation and the preset absolute deviation range to obtain the first adjustment parameter set, or to adjust the preset pulse repetition frequency and the preset square side length to obtain the second adjustment parameter set.

[0043] Furthermore, the adjustment unit includes:

[0044] a first adjustment subunit, configured to, when the exponential deviation is greater than a maximum value of the preset absolute deviation range, reduce the preset distortion threshold according to a relative deviation between the exponential deviation and the maximum value of the preset absolute deviation range and a preset adjustment coefficient, and increase the preset pulse repetition frequency according to the relative deviation between the exponential deviation and the maximum value of the preset absolute deviation range and the preset adjustment coefficient, so as to obtain the first adjustment parameter set;

[0045] The second adjustment subunit is used to increase the preset square side length according to the minimum value of the preset absolute deviation range, the relative deviation of the exponential deviation, and the preset adjustment coefficient when the exponential deviation is less than the minimum value of the preset absolute deviation range, and to reduce the preset pulse repetition frequency according to the minimum value of the preset absolute deviation range, the relative deviation of the exponential deviation, and the preset adjustment coefficient, so as to obtain the second adjustment parameter set.

[0046] Compared with the existing technology, the present invention has the beneficial effect of achieving dynamic control of photoacoustic imaging quality by constructing a complete closed-loop process of acquisition, screening, evaluation, self-learning optimization, and image reconstruction. The spectral texture gradient and imaging uniform error jointly define the spatial consistency of the image, which is used to identify areas of structural distortion. The background noise energy level and laser energy distribution characterize the stability of signal quality and further screen out "noise source grids" where imaging is severely interfered with. On this basis, the actual clarity index measures the current system performance, while the ideal clarity index uses a self-learning model to extract the potential imaging upper limit under noise interference, forming a clear comparison target. Based on this comparison result, the adjustment module adaptively optimizes the acquisition density (grid side length), time parameters (pulse repetition frequency), and fault tolerance level (distortion threshold), thereby realizing a closed-loop self-optimization mechanism of "imaging parameters-image quality-model feedback". This not only significantly improves image clarity and stability, but also adapts to complex tissue structures and diverse imaging environments through continuous learning iterations, improving the robustness and intelligence of the system, and effectively solving the problem of poor adaptability to different tissues due to over-reliance on fixed parameter models, which leads to low recognition accuracy and slow response speed.

[0047] Furthermore, through the collaborative acquisition and calculation of multiple parameters, the signal quality and image characteristics of each square can be accurately characterized at the microscopic level: the laser energy distribution ensures the uniformity of the input excitation; the spectral texture gradient reveals the unevenness of tissue absorption; the imaging uniformity error reflects the local consistency of the image brightness; the background noise energy level evaluates the overall signal-to-noise ratio; when the energy distribution deviation increases or the noise energy level rises, the texture gradient and uniformity error usually deteriorate synchronously. Based on this, the system can accurately identify imaging weaknesses in the distorted squares and noise source squares, and drive the self-learning module to dynamically optimize the pulse frequency, distortion threshold and spatial sampling density, thereby continuously improving the clarity and stability of photoacoustic images in complex tissue scenarios.

[0048] Furthermore, through a multi-dimensional analysis of laser energy distribution, spectral texture gradient, imaging uniformity error, and background noise energy level, the fundamental factors of imaging quality can be accurately revealed: the stability of laser energy distribution ensures the consistency of photoacoustic signal input; the spectral texture gradient reflects the sudden change in spatial information caused by tissue absorption differences; the imaging uniformity error directly measures the smoothness of the reconstructed grayscale distribution; and the background noise energy level limits the upper limit of the signal-to-noise ratio. When energy distribution is uneven or noise increases, the texture gradient and uniformity error tend to deteriorate simultaneously, providing a precise physical and statistical basis for subsequent distortion grid detection and self-learning parameter optimization, significantly improving the system's adaptive imaging capability and stability for complex tissue structures.

[0049] Furthermore, through in-depth quantification of the "fluctuation" time domain characteristics, accurate identification of imaging stability and distortion risk is achieved: the gradient fluctuation value reflects the drastic changes in the signal structure in the time dimension, and the error fluctuation value reveals the fluctuation trend of imaging uniformity. The two are mutually confirmed to distinguish between occasional noise points and continuous distortion; further combined with the background noise energy level and laser energy distribution, the module can accurately locate the real noise source from the distorted grid.

[0050] Furthermore, by normalizing and correlating gradient and uniform error fluctuations, this module accurately detects synchronization anomalies between signal structure (spectral texture gradient) and image uniformity (imaging uniform error) after eliminating the influence of absolute magnitude differences. When these two fluctuate significantly and synchronously, the image is truly distorted, not just occasional noise. Highly correlated fluctuations indicate areas of persistent distortion, ensuring that the system subsequently optimizes only those truly damaged squares, improving the reliability of distortion detection and the targeted nature of self-learning optimization.

[0051] Furthermore, by combining the spatial comparison of background noise energy levels with the laser energy distribution, it is possible to eliminate squares that are mislabeled solely due to high overall noise levels and accurately identify key areas affected by both high noise levels and abnormal laser energy distribution. A noise threshold ensures that only areas truly contaminated by noise are focused, while neighborhood mean comparison utilizes spatial consistency to filter out isolated errors. The resulting output of noise source squares represents both the persistent interference in the temporal dimension and the spatial energy heterogeneity, providing a precise positioning basis for subsequent self-learning parameter adjustments, thereby significantly improving the system's denoising and imaging optimization efficiency.

[0052] Furthermore, by strictly comparing the energy distribution of a single square with the average state of its local neighborhood, this module can effectively distinguish between global offsets caused by system or environmental noise and true local energy anomalies. When the distribution deviation is higher than the threshold, it indicates that there is a persistent imbalance in laser excitation or signal reception in the area. By using the absolute relative deviation to the threshold judgment, the noise source square is accurately located in both spatial and temporal dimensions, providing a highly reliable target area for subsequent self-learning parameter adjustment, significantly improving the stability of the imaging system and the efficiency of automated optimization.

[0053] Furthermore, a comprehensive evaluation of both spatial and quality factors is achieved through the "normalized error to weighted sum": the normalization process eliminates the differences in error magnitudes across different grids, making uniformity fluctuations comparable; the position weight allows higher priority to be given to the image center or key structural areas, reflecting the emphasis on key diagnostic information; the weighted sum aggregates the overall error distribution into a single index, which retains the sensitivity to local problems while taking into account global imaging balance, providing intuitive and quantifiable imaging quality feedback for the self-learning model.

[0054] Furthermore, by precisely comparing real-time feedback on imaging quality (actual clarity index) with model expectations (ideal clarity index), this module achieves a closed-loop optimization process from dynamic response to parameter adaptation. Index deviation, acting as a driving metric, directly maps to adjustments to the pulse repetition frequency, distortion threshold, or spatial sampling density. This reduces oversampling overhead when system imaging exceeds expectations and enhances signal acquisition capabilities when imaging is insufficient. A parameter self-learning mechanism based on underlying clarity differences not only balances imaging quality and resource efficiency, but also ensures the system's adaptive robustness under varying tissue structures and imaging requirements.

[0055] Furthermore, by using the relative relationship between exponential deviation and a preset tolerance range as a driving force, the judgment threshold and sampling frequency can be relaxed when image quality is clearly insufficient, effectively improving signal quality and image clarity. Conversely, when quality is excessive, the sampling density can be relaxed and the frequency reduced, saving resources and preventing oversmoothing. By adaptively adjusting the underlying logic based on the deviation ratio, a dynamic balance between image quality and system efficiency is achieved, ensuring both the flexibility of self-learning and the controllability and convergence of parameter adjustments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Schematic diagram of the self-learning-based photoacoustic imaging system of this embodiment;

[0057] Figure 2 This is a decision logic diagram for the distortion screening unit of this embodiment to determine the distortion grid;

[0058] Figure 3 This is a decision logic diagram for the noise source screening subunit of this embodiment to determine the noise source grid;

[0059] Figure 4 This is a decision logic diagram for the index deviation calculation unit of this embodiment to determine the calculated index deviation. DETAILED DESCRIPTION

[0060] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0061] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0062] See also Figure 1 , which is a schematic diagram of the self-learning-based photoacoustic imaging system of this embodiment;

[0063] This embodiment provides a self-learning-based photoacoustic imaging system, including:

[0064] An acquisition module is used to collect photoacoustic signals, laser energy distribution, spectrum texture gradient, imaging uniformity error, and background noise energy level in each processing square in real time from the target tissue divided by the preset square length based on the preset pulse repetition frequency;

[0065] a screening module connected to the acquisition module, configured to screen out a plurality of distorted squares from all the processing squares based on the spectral texture gradient, the imaging uniformity error, and a preset distortion threshold, and to screen out a plurality of noise source squares from all the distorted squares based on the background noise energy level and the laser energy distribution in each distorted square;

[0066] an actual generation module, connected to the acquisition module and the screening module respectively, for generating an actual clarity index according to the imaging uniform error of each noise source grid;

[0067] an ideal generation module connected to the acquisition module, configured to input the photoacoustic signals of all the noise source grids into a preset self-learning model to generate an ideal clarity index;

[0068] an adjustment module, connected to the actual generation module and the ideal generation module, respectively, for adjusting the preset pulse repetition frequency and the preset distortion threshold according to the actual clarity index and the ideal clarity index to obtain a first adjustment parameter set, or adjusting the preset pulse repetition frequency and the preset grid side length to obtain a second adjustment parameter set;

[0069] An imaging module is connected to the acquisition module and is used to reconstruct and output a photoacoustic image based on the photoacoustic signal re-acquired based on the first adjustment parameter set or the second adjustment parameter set.

[0070] Before performing photoacoustic imaging analysis on the target tissue, the system will divide the entire target area into multiple regular two-dimensional square grid units (i.e., processing grids) according to the preset grid side length. The target tissue is the lesion tissue to be identified.

[0071] The acquisition module achieves refined data acquisition and parameter extraction by integrating a laser excitation unit and multiple feature analysis units into the acquisition module: first, the laser excitation unit uses a nanosecond pulse laser to irradiate the target tissue in sections at a set pulse repetition frequency to stimulate the generation of photoacoustic signals; then, the feature acquisition unit uses a high-speed ultrasonic sensor or CMUT (capacitive micromachined ultrasonic transducer) to collect the pulse laser energy value and center frequency in each processing grid in real time, as well as the pixel grayscale distribution information obtained by the photoacoustic image acquisition system, and at the same time collects the background acoustic baseline signal during the window period without laser excitation; then, the feature calculation unit uses a DSP or FPGA chip to quickly process the collected data, obtain the laser energy distribution by calculating the pulse laser energy value, and obtain the spectral texture gradient by extracting the center frequency through Fourier transform. The statistical analysis method is used to calculate the pixel grayscale mean and standard deviation of the image and derive the imaging uniformity error based on this. Finally, the background noise energy level is calculated based on the energy integral of the baseline signal, thereby completing the multi-dimensional extraction of key imaging parameters of each processing grid.

[0072] The preset pulse repetition frequency refers to the time interval frequency of the laser pulses emitted by the laser excitation unit. It depends on the time resolution requirements of the tissue's photoacoustic response and the thermal stability of the equipment. It is usually set between 10 Hz and 10 kHz. In this embodiment, it is set to 500 Hz, which can take into account both imaging speed and tissue thermal safety.

[0073] The preset grid side length refers to the side length of each square when the target tissue is divided into processing grids. It depends on the required spatial resolution and system sampling capability and is usually set between 0.1 mm and 2 mm. In this embodiment, it is set to 0.5 mm, which can achieve detailed spatial feature extraction and distortion positioning.

[0074] The preset distortion threshold refers to the error limit used to determine whether the processed grid imaging is distorted. It depends on the critical judgment criteria of image uniformity error and texture gradient. It is usually set between 0.05 and 0.5 (dimensionless). In this embodiment, it is set to 0.2, which can effectively identify abnormal grids to trigger the self-learning optimization process.

[0075] The preset self-learning model uses an end-to-end deep regression network to map the original photoacoustic time-domain waveform of the noise source grid to an ideal clarity index.

[0076] Model structure:

[0077] Input layer: Receives the photoacoustic time domain waveform of each noise source square (one-dimensional sequence length 2048 points) and appends the geometric position code of the square (length 16).

[0078] Feature extraction subnetwork: It consists of three layers of 1D-CNN, with kernel sizes of 7, 5, and 3, and the number of channels of 64, 128, and 256, respectively. Each layer is followed by batch normalization and ReLU activation.

[0079] Fusion and dimensionality reduction: Global average pooling is performed on the output of the last convolution layer to obtain a 256-dimensional feature vector. The position codes are concatenated and further compressed through a 128→64 fully connected layer.

[0080] Regression output subnetwork: It consists of two fully connected layers (64→32→1) and finally outputs a scalar, which is the predicted "ideal clarity index".

[0081] Training process:

[0082] Data preparation: Multiple batches of high-quality photoacoustic simulation data and experimental calibration datasets are used. Each sample contains: the original time domain waveform of the noise source grid and the clarity index of the corresponding "gold standard" image (such as the SSIM or PSNR calculated in the area).

[0083] Loss function: Mean square error (MSE) loss is used;

[0084] Optimizer and hyperparameters: Use Adam optimizer, and the learning rate is initially set to , batch size 32, train for 100 epochs, and monitor MSE convergence on the validation set.

[0085] Generate an ideal sharpness index:

[0086] Inference stage: The time domain waveform and position code of each noise source grid collected in real time are sent to the trained self-learning model, and the model outputs the corresponding ideal clarity index after forward propagation.

[0087] Index meaning: Indicates the theoretically achievable maximum image clarity level under the current photoacoustic signal conditions, without any system distortion or noise interference. This index is compared with the actual clarity index generated by the actual generation module and serves as the basis for adaptive parameter adjustment.

[0088] After receiving the photoacoustic time-domain signal re-acquired according to the first or second adjustment parameter set, the imaging module first performs time delay correction according to the newly set pulse repetition frequency and grid side length. It then uses Delay-And-Sum beamforming to focus and superimpose the signals from each channel. It then performs envelope detection and logarithmic compression on the superimposed time-domain signal. Finally, it maps the discrete grid data onto a continuous two-dimensional grid through bilinear interpolation. Finally, it outputs a grayscale or pseudo-color photoacoustic image, ensuring both spatial resolution and signal-to-noise ratio optimization, thereby generating intuitive and clear imaging results.

[0089] First, the acquisition module divides the target tissue area at a preset pulse repetition frequency, and collects the photoacoustic signal, laser energy distribution, spectral texture gradient, imaging uniformity error, and background noise energy level in each processing grid in real time. Subsequently, the screening module preliminarily screens out the distorted grids based on the spectral texture gradient, imaging uniformity error, and distortion threshold, and further extracts the noise source grids based on the background noise energy level and laser energy distribution. The actual generation module calculates the current actual imaging clarity based on this. At the same time, the ideal generation module inputs the photoacoustic signal of the noise source grid into the self-learning model to generate an ideal clarity index. By comparing the actual and ideal clarity indices, the adjustment module determines whether to optimize the acquisition settings through the first parameter set (adjusting the pulse repetition frequency and distortion threshold) or the second parameter set (adjusting the pulse repetition frequency and grid side length). Finally, the imaging module uses the optimized parameters to re-acquire the photoacoustic signal and reconstruct the image, achieving dynamic adaptive high-quality photoacoustic imaging.

[0090] By constructing a complete closed-loop process of acquisition, screening, evaluation, self-learning optimization, and image reconstruction, dynamic control of photoacoustic imaging quality is achieved. Spectral texture gradient and imaging uniformity error jointly define the spatial consistency of the image, which is used to identify areas of structural distortion. The background noise energy level and laser energy distribution characterize the stability of signal quality and further screen out "noise source grids" where imaging is severely interfered with. Furthermore, the actual clarity index measures current system performance, while the ideal clarity index uses a self-learning model to extract the potential upper limit of imaging under noise interference, forming a clear comparison target. Based on this comparison result, the adjustment module adaptively optimizes the acquisition density (grid side length), time parameters (pulse repetition frequency), and fault tolerance level (distortion threshold), thus implementing a closed-loop self-optimization mechanism of "imaging parameters-image quality-model feedback". This not only significantly improves image clarity and stability, but also, through continuous learning iterations, adapts to complex tissue structures and diverse imaging environments, improving the system's robustness and intelligence. This effectively addresses the problems of low recognition accuracy and slow response speed caused by over-reliance on fixed parameter models, which are poorly adaptable to different tissues.

[0091] Specifically, the acquisition module includes:

[0092] a laser excitation unit, configured to emit nanosecond laser pulses at the preset pulse repetition frequency and irradiate all the processing grids of the target tissue to obtain a plurality of the photoacoustic signals;

[0093] A feature acquisition unit for real-time acquisition of the pulse energy value, center frequency, pixel grayscale average, pixel grayscale standard deviation, and energy value of the acoustic baseline signal during a preset laser-free triggering period within each processing grid;

[0094] A feature calculation unit is connected to the feature acquisition unit, and is used to calculate the energy distribution of each laser according to the pulse energy value, and to calculate the spectrum texture gradient according to the center frequency, and to calculate the imaging uniformity error according to the pixel grayscale average and the pixel grayscale standard deviation, and to calculate the background noise energy level according to the energy value of the acoustic baseline signal.

[0095] The preset laser-free triggering period refers to the time window used for acoustic baseline sampling, which depends on the maximum acoustic wave propagation delay of the photoacoustic signal in the tissue and the stability of the environmental noise. It is usually set between 0.5ms and 5ms. In this embodiment, it is set to 2ms, which can fully collect background noise data under the premise of eliminating laser excitation response.

[0096] The laser excitation unit first applies nanosecond pulse irradiation to the entire pre-divided grid area at a preset pulse repetition frequency to stimulate a photoacoustic response; the feature acquisition unit synchronously records the pulse energy value (reflecting the excitation consistency), spectral center frequency (reflecting the local absorption characteristics), pixel grayscale average and standard deviation (characterizing the imaging brightness and texture distribution), and acoustic baseline signal energy collected during the laser interval (characterizing the environment and system noise level) of each square through a high-speed ultrasonic detector and an image acquisition system; then, the feature calculation unit performs statistics and transformation on these raw data, summarizes the energy value into laser energy distribution, maps the frequency difference into spectral texture gradient, converts the grayscale statistics into imaging uniform error, and calculates the background noise energy based on the baseline signal, providing refined multi-dimensional input for subsequent distortion screening, self-learning optimization and image reconstruction.

[0097] Through the coordinated acquisition and calculation of multiple parameters, the signal quality and image characteristics of each square can be accurately characterized at the microscopic level: the laser energy distribution ensures the uniformity of the input excitation; the spectral texture gradient reveals the unevenness of tissue absorption; the imaging uniformity error reflects the local consistency of the image brightness; the background noise energy level evaluates the overall signal-to-noise ratio; when the energy distribution deviation increases or the noise energy level rises, the texture gradient and uniformity error usually deteriorate synchronously. Based on this, the system can accurately identify imaging weaknesses in the distorted squares and noise source squares, and drive the self-learning module to dynamically optimize the pulse frequency, distortion threshold and spatial sampling density, thereby continuously improving the clarity and stability of photoacoustic images in complex tissue scenarios.

[0098] Specifically, the feature calculation unit includes:

[0099] an energy distribution calculation subunit, for calculating an average value of all the pulse energy values ​​to obtain the laser energy distribution;

[0100] a frequency calculation subunit, configured to calculate the maximum difference in center frequencies among adjacent processing squares selected based on a preset neighborhood size with each processing square as the center, to obtain the spectral texture gradient;

[0101] a uniform error calculation subunit, for calculating the ratio of the pixel grayscale standard deviation to the pixel grayscale average value, the imaging uniform error;

[0102] The energy level calculation subunit is used to calculate the noise energy level of the energy value of the acoustic baseline signal to obtain the background noise energy level. , where M is the background noise energy level, T' is the corrected no-laser trigger period after excluding the highest and lowest 10% of the data in the preset no-laser trigger period, t1 is the initial moment of the corrected no-laser trigger period, t2 is the end moment of the corrected no-laser trigger period, and S(t) is the energy value of the acoustic baseline signal.

[0103] The preset neighborhood size refers to the range of adjacent grids selected for spectral texture gradient calculation, which depends on the spatial resolution and signal stability requirements and is usually set between [1 grid, 3 grids]. In this embodiment, it is set to 1 grid (i.e., 3×3 neighborhood), which can suppress the influence of isolated noise while ensuring local contrast.

[0104] In the feature calculation unit, the energy distribution calculation subunit first averages the multiple pulse energy values ​​collected in each processing grid to obtain the laser energy distribution; the frequency calculation subunit calculates the maximum difference in the center frequency among adjacent grids selected according to the preset neighborhood size with the target grid as the center to obtain the spectrum texture gradient; the uniform error calculation subunit quantifies the imaging uniform error through the ratio of the pixel grayscale standard deviation to the average grayscale value; finally, the energy level calculation subunit calculates the background noise energy level based on the statistical results of the acoustic baseline signal energy value.

[0105] Through a multi-dimensional analysis of laser energy distribution, spectral texture gradient, imaging uniformity error, and background noise energy level, the fundamental factors of imaging quality can be accurately revealed: the stability of the laser energy distribution ensures the consistency of the photoacoustic signal input; the spectral texture gradient reflects the sudden change in spatial information caused by differences in tissue absorption; the imaging uniformity error directly measures the smoothness of the reconstructed grayscale distribution; and the background noise energy level limits the upper limit of the signal-to-noise ratio. When the energy distribution is uneven or the noise increases, the texture gradient and uniformity error tend to deteriorate simultaneously, providing a precise physical and statistical basis for subsequent distortion grid detection and self-learning parameter optimization, significantly improving the system's adaptive imaging capability and stability for complex tissue structures.

[0106] Specifically, the screening module includes:

[0107] a screening fluctuation calculation submodule, configured to calculate the standard deviation of the spectral texture gradient at each moment from the initial moment to a preset distortion screening duration to obtain a plurality of gradient fluctuation values, and to calculate the standard deviation of the imaging uniform error at each moment from the initial moment to a preset distortion screening duration to obtain a plurality of error fluctuation values;

[0108] a distortion screening submodule, connected to the screening fluctuation calculation submodule, for screening out a number of distorted squares according to all the gradient fluctuation values ​​and all the error fluctuation values;

[0109] The noise source screening submodule is connected to the distortion screening submodule and is used to screen out a number of the noise source grids from all the distortion grids according to the background noise energy level and the laser energy distribution in each of the distortion grids.

[0110] The preset distortion screening time refers to the length of the time window used for fluctuation calculation, which depends on the pulse repetition frequency and the stabilization time of the tissue acoustic response. It is usually set between 50ms and 500ms. In this embodiment, it is set to 100ms, which can capture imaging quality fluctuations in a timely manner while avoiding excessive delay in optimization feedback.

[0111] The screening fluctuation calculation submodule first calculates the standard deviation of the spectral texture gradient sequence and imaging uniform error sequence of each processing grid within the time window from the initial moment to the preset distortion screening duration, obtaining a series of gradient fluctuation values ​​and error fluctuation values; the distortion screening submodule then jointly compares these two sets of fluctuation indicators, and marks the grids with high fluctuation as distorted grids based on the preset threshold; the noise source screening submodule then compares the background noise energy level in these distorted grids with the laser energy distribution, eliminates low-noise or normal-energy items, and finally screens out the noise source grids that are truly affected by noise and need to be optimized.

[0112] By deeply quantifying the time-domain characteristics of "fluctuation", the module achieves precise identification of imaging stability and distortion risk: the gradient fluctuation value reflects the dramatic changes in the signal structure in the time dimension, and the error fluctuation value reveals the fluctuation trend of imaging uniformity. The two complement each other and can distinguish between occasional noise points and continuous distortion; further combined with the background noise energy level and laser energy distribution, the module can accurately locate the true noise source from the distorted grid.

[0113] Please continue reading Figure 2 As shown, it is a judgment logic diagram of the distortion screening unit judging the distortion grid in this embodiment;

[0114] The distortion screening submodule includes:

[0115] A screening normalization unit is used to normalize all the gradient fluctuation values ​​to obtain a gradient fluctuation normalized set, and to normalize all the error fluctuation values ​​to obtain an error fluctuation normalized set;

[0116] a correlation calculation unit connected to the screening normalization unit, for calculating a correlation coefficient between the gradient fluctuation normalization set and the error fluctuation normalization set to obtain a distortion correlation;

[0117] The distortion screening unit is connected to the correlation calculation unit and is used to determine that the processing grid is the distorted grid when the distortion correlation is greater than the preset distortion threshold, so as to screen out a number of distorted grids from all the processing grids.

[0118] First, the screening and normalization unit performs minimum-maximum normalization on the gradient fluctuation values ​​and error fluctuation values ​​obtained for each processing grid within a preset time length (existing technology, no further description is given here), forming a normalized set of gradient fluctuations and a normalized set of error fluctuations; then, the correlation calculation unit calculates the correlation coefficient of these two sets of normalized data to obtain the distortion correlation; finally, the distortion screening unit compares this correlation with a preset distortion threshold. If the correlation exceeds the threshold, it is determined that the corresponding grid continuously exhibits synchronous fluctuations in spectrum and uniformity, and is then marked as a distorted grid.

[0119] By normalizing and correlating gradient and uniform error fluctuations, this module accurately detects synchronization anomalies between signal structure (spectral texture gradient) and image uniformity (imaging uniform error) after eliminating the influence of absolute magnitude differences. When these two fluctuate significantly and synchronously, the image is truly distorted, not just occasional noise. Highly correlated fluctuations indicate areas of persistent distortion, ensuring that the system subsequently optimizes only those truly damaged squares, improving the reliability of distortion detection and the targeted nature of self-learning optimization.

[0120] Specifically, the noise source screening submodule includes:

[0121] a candidate square determining unit, configured to determine, when the background noise energy level is greater than a preset noise threshold, that the distorted square is the candidate square, so as to determine a plurality of candidate squares;

[0122] a neighborhood mean calculation unit connected to the candidate square determination subunit, configured to calculate the average value of the laser energy distribution of all other candidate squares within the preset neighborhood size centered on the single candidate square, to obtain a neighborhood mean;

[0123] A noise source screening unit is connected to the neighborhood mean calculation unit and is used to screen out a number of noise source squares from all candidate squares according to the laser energy distribution of each candidate square and the corresponding neighborhood mean.

[0124] The candidate square determination unit first determines the background noise energy level of all distorted squares and marks any square whose energy level exceeds the preset noise threshold as a candidate square. Then, the neighborhood mean calculation unit selects other candidate squares around each candidate square, centered on the square and according to the preset neighborhood size, and calculates the average laser energy distribution of these squares. Finally, the noise source screening unit compares the laser energy distribution of each candidate square with the corresponding neighborhood mean. If the deviation exceeds the set range, the square is determined to be a noise source, completing the precise elimination from the candidate set to the noise source set.

[0125] By combining the spatial comparison of background noise energy levels with the laser energy distribution, we can eliminate squares that are mislabeled solely due to high overall noise levels and accurately identify key areas affected by both high noise levels and abnormal laser energy distribution. A noise threshold ensures that only areas truly contaminated by noise are focused, while neighborhood mean comparison utilizes spatial consistency to filter out isolated errors. The resulting noise source squares represent both the persistent interference in the temporal dimension and the spatial energy heterogeneity, providing a precise positioning basis for subsequent self-learning parameter adjustments, significantly improving the system's denoising and imaging optimization efficiency.

[0126] Please continue reading Figure 3 As shown, it is a decision logic diagram of the noise source screening subunit in this embodiment for determining the noise source grid;

[0127] Specifically, the noise source screening unit includes:

[0128] a distribution deviation calculation subunit, configured to calculate an absolute value of a relative deviation between the laser energy distribution and the neighborhood mean, to obtain a distribution deviation value;

[0129] The noise source screening subunit is connected to the distribution deviation calculation subunit and is used to determine that the candidate square is the noise source square when the distribution deviation value is greater than a preset distribution deviation threshold, so as to screen out a number of the noise source squares from all the distorted squares.

[0130] The preset distribution deviation threshold refers to the relative deviation limit used to determine abnormal laser energy distribution. It depends on the system energy output stability and tissue absorption uniformity requirements. It is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.2, which can effectively distinguish normal energy fluctuations from true local abnormal distribution.

[0131] The distribution deviation calculation subunit first calculates the absolute relative deviation between the laser energy distribution of each candidate square and the mean of its neighborhood to obtain the distribution deviation value; then, the noise source screening subunit compares the distribution deviation value with the preset distribution deviation threshold. Any candidate square whose deviation exceeds the threshold is determined to be a noise source square, accurately screening out the noise source area with the most serious impact from all distorted squares.

[0132] By strictly comparing the energy distribution of a single square with the average state of its local neighborhood, this module can effectively distinguish between global offsets caused by system or environmental noise and true local energy anomalies. When the distribution deviation is higher than the threshold, it indicates that there is a persistent imbalance in laser excitation or signal reception in the area. By using absolute relative deviation to threshold judgment, it can accurately locate the noise source square in both spatial and temporal dimensions, providing a highly reliable target area for subsequent self-learning parameter adjustment, significantly improving the stability of the imaging system and the efficiency of automated optimization.

[0133] Specifically, the actual generation module includes:

[0134] an error normalization unit, configured to normalize the imaging uniform errors of all the noise source grids to obtain a number of noise source error normalization values;

[0135] an actual index calculation unit connected to the error normalization unit, for performing weighted sum calculation on the error normalization value of each noise source and its corresponding preset position weight to generate the actual clarity index; , where C is the actual clarity index, wi is the preset position weight, N is the total number of noise source grids, and Pi is the normalized value of the noise source error of the i-th noise source grid.

[0136] The preset position weight refers to the spatial priority coefficient assigned to different noise source grids in the actual clarity index calculation. It depends on the diagnostic importance and geometric centrality of the grid in the imaging field of view and is usually set between 0.5 and 2.0. In this embodiment, the core grid is set to 1.5 and the edge grid is set to 1.0. This can highlight the clarity contribution of key areas and suppress the noise impact of secondary areas.

[0137] The error normalization unit first normalizes the imaging uniform errors of all noise source grids using the minimum-maximum or mean-variance method to generate a set of dimensionless error normalization values. Subsequently, the actual index calculation unit multiplies each normalized error value by the corresponding preset spatial position weight and sums them up, outputting the comprehensive weighted result as the actual clarity index.

[0138] A comprehensive evaluation of both spatial and quality factors is achieved through the "normalized error to weighted sum" process: normalization eliminates the differences in error magnitudes across grids, making uniformity fluctuations comparable; position weighting allows higher priority to be given to the image center or key structural areas, reflecting the emphasis on key diagnostic information; and weighted summation aggregates the overall error distribution into a single index, which retains the sensitivity to local problems while taking into account global imaging balance, providing intuitive and quantifiable imaging quality feedback for the self-learning model.

[0139] Please continue reading Figure 4 As shown, it is a decision logic diagram of the index deviation calculation unit of this embodiment for determining the calculated index deviation;

[0140] The adjustment module includes:

[0141] an index deviation calculation unit, configured to calculate an absolute value of a relative deviation between the actual clarity index and the ideal clarity index when the actual clarity index is greater than the ideal clarity index, to obtain an index deviation;

[0142] An adjustment unit is connected to the exponential deviation calculation unit and is used to adjust the preset pulse repetition frequency and the preset distortion threshold according to the exponential deviation and the preset absolute deviation range to obtain the first adjustment parameter set, or to adjust the preset pulse repetition frequency and the preset square side length to obtain the second adjustment parameter set.

[0143] The preset absolute deviation range refers to the upper and lower limits of the relative deviation used to trigger different adjustment strategies. It is determined by the system's tolerance for imaging quality and the self-learning convergence speed requirements. It is usually set between 0.05 and 0.3. In this embodiment, it is set to [0.1, 0.2]. It can ensure image clarity while avoiding frequent or delayed parameter adjustments due to excessively small or large deviations.

[0144] The index deviation calculation unit first determines whether the actual clarity index is higher than the ideal clarity index, calculates the relative deviation between the two, and outputs the index deviation; the adjustment unit compares the deviation value with the preset absolute deviation range. If the deviation exceeds the upper limit, the preset pulse repetition frequency and the preset distortion threshold are adjusted to generate a first adjustment parameter set to enhance imaging sensitivity; if the deviation is lower than the lower limit, the preset pulse repetition frequency and the preset grid side length are adjusted to generate a second adjustment parameter set to improve sampling efficiency; the system then re-acquires and feeds back the imaging with the new parameter set.

[0145] By precisely comparing real-time image quality feedback (actual clarity index) with model expectations (ideal clarity index), this module achieves a closed-loop optimization process from dynamic response to parameter adaptation. Index deviation, the driving metric, directly maps to adjustments to the pulse repetition frequency, distortion threshold, or spatial sampling density. This reduces oversampling overhead when system imaging exceeds expectations and enhances signal acquisition when imaging is insufficient. A parameter self-learning mechanism based on underlying clarity differences not only balances imaging quality and resource efficiency, but also ensures the system's adaptive robustness under varying tissue structures and imaging requirements.

[0146] Specifically, the adjustment unit includes:

[0147] a first adjustment subunit, configured to, when the exponential deviation is greater than a maximum value of the preset absolute deviation range, reduce the preset distortion threshold according to a relative deviation between the exponential deviation and the maximum value of the preset absolute deviation range and a preset adjustment coefficient, and increase the preset pulse repetition frequency according to the relative deviation between the exponential deviation and the maximum value of the preset absolute deviation range and the preset adjustment coefficient, so as to obtain the first adjustment parameter set, wherein: , is the preset distortion threshold after reduction, G is the preset distortion threshold before reduction, k is the preset adjustment coefficient, U is the exponential deviation, Umax is the maximum value of the preset absolute deviation range, , is the preset pulse repetition frequency after the increase, and L is the preset pulse repetition frequency before the increase;

[0148] The second adjustment subunit is configured to, when the exponential deviation is less than the minimum value of the preset absolute deviation range, increase the preset square side length according to the minimum value of the preset absolute deviation range, the relative deviation of the exponential deviation, and the preset adjustment coefficient, and reduce the preset pulse repetition frequency according to the minimum value of the preset absolute deviation range, the relative deviation of the exponential deviation, and the preset adjustment coefficient, so as to obtain the second adjustment parameter set, wherein: , is the preset square length after enlargement, J is the preset square length before enlargement, k is the preset adjustment coefficient, Umin is the minimum value of the preset absolute deviation range, , is the preset pulse repetition frequency after reduction, and L is the preset pulse repetition frequency before reduction.

[0149] The preset adjustment coefficient refers to the proportional factor used to control the adaptive adjustment speed of the parameters. It depends on the system's requirements for the balance between adjustment sensitivity and convergence stability. It is usually set between 0.1 and 1.0. In this embodiment, it is set to 0.5, which can ensure significant adjustment effect while avoiding excessive parameter oscillation.

[0150] When the exponential deviation exceeds the upper limit of the preset absolute deviation range, the system first calculates the ratio of the deviation to the upper limit, and then reduces the distortion threshold according to the ratio and the preset adjustment coefficient to relax the distortion judgment conditions, and at the same time increases the pulse repetition frequency according to the same ratio and adjustment coefficient to enhance signal acquisition; in the second adjustment subunit, when the exponential deviation is lower than the lower limit of the preset absolute deviation range, the system increases the grid side length according to the lower limit deviation ratio and the adjustment coefficient to reduce the spatial sampling density, and reduces the pulse repetition frequency according to the same ratio and adjustment coefficient to reduce power consumption, and outputs the corresponding first or second adjustment parameter set respectively.

[0151] By using the relative relationship between exponential deviation and a preset tolerance range as a driving force, the system can relax the judgment threshold and increase the sampling frequency when image quality is clearly insufficient, effectively improving signal quality and image clarity. Conversely, when quality is excessive, the sampling density can be relaxed and the frequency reduced to conserve resources and prevent oversmoothing. Through adaptive adjustment of the underlying logic based on the deviation ratio, a dynamic balance between image quality and system efficiency is achieved, ensuring both the flexibility of self-learning and the controllability and convergence of parameter adjustments.

[0152] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A self-learning-based photoacoustic imaging system, characterized in that: include: An acquisition module is used to collect photoacoustic signals, laser energy distribution, spectrum texture gradient, imaging uniformity error, and background noise energy level in each processing square in real time from the target tissue divided by the preset square length based on the preset pulse repetition frequency; a screening module connected to the acquisition module, configured to screen out a plurality of distorted squares from all the processing squares based on the spectral texture gradient, the imaging uniformity error, and a preset distortion threshold, and to screen out a plurality of noise source squares from all the distorted squares based on the background noise energy level and the laser energy distribution in each distorted square; an actual generation module, connected to the acquisition module and the screening module respectively, for generating an actual clarity index according to the imaging uniform error of each noise source grid; an ideal generation module connected to the acquisition module, configured to input the photoacoustic signals of all the noise source grids into a preset self-learning model to generate an ideal clarity index; an adjustment module, connected to the actual generation module and the ideal generation module, respectively, for adjusting the preset pulse repetition frequency and the preset distortion threshold according to the actual clarity index and the ideal clarity index to obtain a first adjustment parameter set, or adjusting the preset pulse repetition frequency and the preset grid side length to obtain a second adjustment parameter set; an imaging module connected to the acquisition module, configured to reconstruct and output a photoacoustic image based on the photoacoustic signal reacquired based on the first adjustment parameter set or the second adjustment parameter set; The adjustment module includes: an index deviation calculation unit, configured to calculate an absolute value of a relative deviation between the actual clarity index and the ideal clarity index when the actual clarity index is greater than the ideal clarity index, to obtain an index deviation; an adjustment unit connected to the exponential deviation calculation unit, configured to adjust the preset pulse repetition frequency and the preset distortion threshold according to the exponential deviation and the preset absolute deviation range to obtain the first adjustment parameter set, or to adjust the preset pulse repetition frequency and the preset grid side length to obtain the second adjustment parameter set; The adjustment unit includes: a first adjustment subunit, configured to, when the exponential deviation is greater than a maximum value of the preset absolute deviation range, reduce the preset distortion threshold according to a relative deviation between the exponential deviation and the maximum value of the preset absolute deviation range and a preset adjustment coefficient, and increase the preset pulse repetition frequency according to the relative deviation between the exponential deviation and the maximum value of the preset absolute deviation range and the preset adjustment coefficient, to obtain a first adjustment parameter set, wherein G'=G×[1-k×(U-Umax) / Umax], G' is the preset distortion threshold after reduction, G is the preset distortion threshold before reduction, k is the preset adjustment coefficient, U is the exponential deviation, Umax is the maximum value of the preset absolute deviation range, and L'=L×[1+k×(U-Umax) / Umax], L' is the preset pulse repetition frequency after increase, and L is the preset pulse repetition frequency before increase; a second adjustment subunit, configured to, when the exponential deviation is less than a minimum value of the preset absolute deviation range, increase the preset square side length according to the minimum value of the preset absolute deviation range, the relative deviation of the exponential deviation, and the preset adjustment coefficient, and reduce the preset pulse repetition frequency according to the minimum value of the preset absolute deviation range, the relative deviation of the exponential deviation, and the preset adjustment coefficient, to obtain a second adjustment parameter set, wherein J'=J×[1+k×(Umin-U) / U], J' is the preset square side length after increase, J is the preset square side length before increase, k is the preset adjustment coefficient, Umin is the minimum value of the preset absolute deviation range, and L'=L×[1-k×(Umin-U) / U], L' is the preset pulse repetition frequency after decrease, and L is the preset pulse repetition frequency before decrease; The preset self-learning model adopts an end-to-end deep regression network to map the original photoacoustic time domain waveform of the noise source grid to an ideal clarity index.

2. The self-learning based photoacoustic imaging system according to claim 1, characterized in that: The acquisition module includes: a laser excitation unit, configured to emit nanosecond laser pulses at the preset pulse repetition frequency and irradiate all the processing grids of the target tissue to obtain a plurality of the photoacoustic signals; A feature acquisition unit for real-time acquisition of the pulse energy value, center frequency, pixel grayscale average, pixel grayscale standard deviation, and energy value of the acoustic baseline signal during a preset laser-free triggering period within each processing grid; A feature calculation unit is connected to the feature acquisition unit, and is used to calculate the energy distribution of each laser according to the pulse energy value, and to calculate the spectrum texture gradient according to the center frequency, and to calculate the imaging uniformity error according to the pixel grayscale average and the pixel grayscale standard deviation, and to calculate the background noise energy level according to the energy value of the acoustic baseline signal.

3. The self-learning based photoacoustic imaging system according to claim 2, characterized in that: The feature calculation unit includes: an energy distribution calculation subunit, for calculating an average value of all the pulse energy values ​​to obtain the laser energy distribution; a frequency calculation subunit, configured to calculate the maximum difference in center frequencies among adjacent processing squares selected based on a preset neighborhood size with each processing square as the center, to obtain the spectral texture gradient; a uniform error calculation subunit, for calculating the ratio of the pixel grayscale standard deviation to the pixel grayscale average value, the imaging uniform error; The energy level calculation subunit is used to calculate the noise energy level of the energy value of the acoustic baseline signal to obtain the background noise energy level.

4. The self-learning based photoacoustic imaging system according to claim 3, characterized in that: The screening module includes: a screening fluctuation calculation submodule, configured to calculate the standard deviation of the spectral texture gradient at each moment from the initial moment to a preset distortion screening duration to obtain a plurality of gradient fluctuation values, and to calculate the standard deviation of the imaging uniform error at each moment from the initial moment to a preset distortion screening duration to obtain a plurality of error fluctuation values; a distortion screening submodule, connected to the screening fluctuation calculation submodule, for screening out a number of distorted squares according to all the gradient fluctuation values ​​and all the error fluctuation values; The noise source screening submodule is connected to the distortion screening submodule and is used to screen out a number of the noise source grids from all the distortion grids according to the background noise energy level and the laser energy distribution in each of the distortion grids.

5. The self-learning based photoacoustic imaging system according to claim 4, characterized in that: The distortion screening submodule includes: A screening normalization unit is used to normalize all the gradient fluctuation values ​​to obtain a gradient fluctuation normalized set, and to normalize all the error fluctuation values ​​to obtain an error fluctuation normalized set; a correlation calculation unit connected to the screening normalization unit, for calculating a correlation coefficient between the gradient fluctuation normalization set and the error fluctuation normalization set to obtain a distortion correlation; The distortion screening unit is connected to the correlation calculation unit and is used to determine that the processing grid is the distorted grid when the distortion correlation is greater than the preset distortion threshold, so as to screen out a number of distorted grids from all the processing grids.

6. The self-learning based photoacoustic imaging system according to claim 5, characterized in that: The noise source screening submodule includes: a candidate square determining unit, configured to determine, when the background noise energy level is greater than a preset noise threshold, that the distorted square is the candidate square, so as to determine a plurality of candidate squares; a neighborhood mean calculation unit connected to the candidate square determination subunit, configured to calculate the average value of the laser energy distribution of all other candidate squares within the preset neighborhood size centered on the single candidate square, to obtain a neighborhood mean; A noise source screening unit is connected to the neighborhood mean calculation unit and is used to screen out a number of noise source squares from all candidate squares according to the laser energy distribution of each candidate square and the corresponding neighborhood mean.

7. The self-learning based photoacoustic imaging system according to claim 6, characterized in that: The noise source screening unit includes: a distribution deviation calculation subunit, configured to calculate an absolute value of a relative deviation between the laser energy distribution and the neighborhood mean, to obtain a distribution deviation value; The noise source screening subunit is connected to the distribution deviation calculation subunit and is used to determine that the candidate square is the noise source square when the distribution deviation value is greater than a preset distribution deviation threshold, so as to screen out a number of the noise source squares from all the distorted squares.

8. The self-learning based photoacoustic imaging system according to claim 7, characterized in that: The actual generation module includes: an error normalization unit, configured to normalize the imaging uniform errors of all the noise source grids to obtain a number of noise source error normalization values; The actual index calculation unit is connected to the error normalization unit and is used to perform weighted sum calculation on the error normalization value of each noise source and its corresponding preset position weight to generate the actual clarity index.

Images