An integrated analysis method for digestive endoscopy data

Abstract

The present application relates to medical image processing and data analysis technical field, disclose a kind of integrated analysis method of gastrointestinal endoscopy data, including by digital video frame sequence and pressure signal sequence comprising gastrointestinal endoscopy data is synchronously collected using synchronous trigger circuit;Central processing unit uses exposure starting time stamp and air flow transmission delay model to construct synchronous dataset;Discrete fourier transform is executed to pressure signal sequence, and aerodynamic modulation characteristic frequency is identified and synthesized reference signal;Subpixel level morphing tracking is executed to digital video frame sequence, and global motion component is obtained by stripping local motion component.The present application executes frequency domain synchronous decoupling to local motion component by using reference signal, eliminates physiological motion interference and extracts synchronous deformation component modulated by pressure pulsation;According to synchronous deformation component and pressure pulsation amplitude, determine compliance coefficient and map physical characteristic heat map superimposed to digital video frame, generate augmented reality fusion image output.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and data analysis technology, specifically to an integrated analysis method for gastrointestinal endoscopy data. Background Technology

[0002] Current gastrointestinal endoscopy primarily utilizes optical imaging technology to acquire information about the surface morphology of the gastrointestinal tract lumen. The procedure typically involves using an air pump to deliver compressed air into the lumen, maintaining its expansion. Existing endoscopic diagnostic systems focus on extracting features from two-dimensional images, lacking quantitative analysis of the mechanical properties of the gastrointestinal tract lumen. Furthermore, the pressure data and video image data generated during gastrointestinal endoscopy are acquired in independent channels, preventing the formation of a physically correlated structured dataset between visual and pressure information along both the temporal and spatial axes.

[0003] The digital video frame sequence acquired by the image sensor and the pressure signal sequence acquired by the pressure monitoring sensor exhibit differences in sampling frequency and hardware clock skew. The pressure pulsation signal generated by the air injection pump experiences a time difference as it travels to the end of the digestive tract lumen, causing a misalignment in physical phase between the tissue deformation sequence recorded in the image and the pressure signal sequence. Low-frequency displacement interference from respiratory motion and intestinal peristalsis on the digestive tract wall, combined with tissue deformation caused by the high-frequency pressure pulsation from the air injection pump, increases the difficulty of extracting the physical properties of the digestive tract wall tissue from the digital video frame sequence.

[0004] Traditional analytical methods cannot remove motion noise and quantify tissue compliance from non-contact video data. Due to the lack of a unified clock synchronization mechanism and frequency domain decoupling methods, operators cannot obtain real-time information on the mechanical stiffness distribution of the digestive tract wall tissues during endoscopy. The lack of heterogeneous feature fusion between optical images and mechanical parameters results in endoscopic findings being limited by the observer's subjective experience. The absence of an integrated analytical method to transform the subtle tissue responses induced by pressure pulsations into a visualized thermogram of physical characteristics limits the quantitative identification of submucosal sclerotic areas. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an integrated analysis method for gastrointestinal endoscopy data, which solves the problems of time phase inaccuracy between optical and mechanical dimensions, difficulty in separating tissue deformation signals under physiological motion interference, and lack of real-time quantitative compliance assessment methods in existing gastrointestinal endoscopy data.

[0006] To achieve the above objectives, this invention provides an integrated analysis method for gastrointestinal endoscopic examination data, comprising the following steps:

[0007] The synchronous triggering circuit sends clock synchronization pulses to the image sensor and pressure monitoring sensor, and uses the clock synchronization pulses to drive the image sensor and pressure monitoring sensor to synchronously acquire gastrointestinal endoscopy data composed of digital video frame sequences and pressure signal sequences; the central processing unit assigns an exposure start time timestamp to each digital video frame in the digital video frame sequence, and synchronously extracts the instantaneous pressure value at the corresponding time from the pressure signal sequence to construct a synchronous dataset containing multiple sets of heterogeneous data pairs.

[0008] During the construction of the synchronous dataset, the central processing unit uses a multi-threaded circular buffer mechanism to perform unpacking and reassembly operations on parallel video stream data and serial pressure signal data, and calls the airflow transmission delay model to compensate for the time difference generated by pressure pulsation during transmission based on the physical length between the pressure monitoring sensor and the end of the digestive tract lumen, thereby achieving physical consistency between mechanical dimension data and optical dimension data on the time axis.

[0009] The central processing unit (CPU) reads the pressure signal sequence from the synchronous dataset and performs a discrete Fourier transform to convert the time-domain signal into a frequency-domain frequency component distribution. Based on the generated power spectral density sequence, the CPU searches for the peak points of the power spectral density sequence within a preset operating frequency range to determine the aerodynamic modulation characteristic frequency generated by the operation of the air injection pump. Since the frequencies of respiratory motion interference and intestinal peristalsis interference are usually lower than the operating frequency of the air injection pump, the aerodynamic modulation characteristic frequency, which serves as the reference for subsequent carrier waves, can be separated by searching for peaks within the frequency band.

[0010] The central processing unit (CPU) uses the pneumatic modulation characteristic frequency, pressure pulsation amplitude, and the initial phase of the pressure signal sequence to synthesize a continuous reference signal through numerical calculation. Simultaneously, the CPU performs subpixel-level deformation tracking on the digital video frame sequence, resampling adjacent image frames using a bilinear trilinear interpolation algorithm, and using an iterative minimization of the squared error function to solve for the subpixel position offset generated by the pixel coordinates, obtaining the incremental displacement vector. The CPU further combines an affine transformation model to extract the global translation and rotation components generated by the endoscope lens movement from the incremental displacement vector, extracting only the local motion component reflecting the movement of the digestive tract wall tissue.

[0011] The central processing unit (CPU) performs frequency-domain synchronous decoupling on the local motion components using a reference signal. Utilizing the principle of orthogonal coherent detection, the CPU performs multiplication and accumulation operations on the local motion components at each pixel coordinate with the reference signal, and performs time averaging on the results within the analysis time window. Based on the coherence differences of different frequency components, the low-frequency displacement components caused by frequency-incoherent respiratory movements and intestinal peristalsis cancel each other out during the time integration process, thereby extracting the synchronous deformation components modulated by the pressure pulsation of the air pump from the complex background motion.

[0012] The central processing unit (CPU) determines the compliance coefficient of each pixel location by calculating the quotient of the vector magnitude of the synchronous deformation component and the amplitude of the pressure pulsation, and generates a compliance coefficient distribution matrix. The CPU performs linear scaling based on the compliance coefficient distribution matrix to generate a physical feature heatmap reflecting the biomechanical state of the tissue. The CPU performs weighted pixel fusion calculations on the physical feature heatmap and digital video frames to generate an augmented reality fused image, and uses the Kalman filter principle to perform inter-frame smoothing prediction on the physical feature heatmap.

[0013] The central processing unit outputs augmented reality fusion images and real-time mechanical histogram dashboards containing compliance coefficient statistical indicators in real time through the video output interface; by utilizing the spatial distribution differences of compliance coefficients, it helps to identify hardened or relaxed regions of the digestive tract wall, and completes the integrated analysis and display of optical and physical mechanical characteristics.

[0014] This invention provides an integrated analysis method for gastrointestinal endoscopic examination data. It has the following beneficial effects:

[0015] 1. This invention sends clock synchronization pulses to the image sensor and pressure monitoring sensor through a synchronous trigger circuit, so that the digital video frame sequence and pressure signal sequence are acquired under a unified time reference. The central processing unit uses the timestamp of the exposure start time and the airflow transmission delay model to compensate for the time difference of the pressure pulsation from the pressure monitoring sensor position to the end of the tube, eliminates the phase deviation between optical dimension data and mechanical dimension data on the time axis, and ensures that the heterogeneous data pairs contained in the synchronous dataset have physical correspondence.

[0016] 2. This invention uses a central processing unit to perform a discrete Fourier transform on the pressure signal sequence, identify the pneumatic modulation characteristic frequency generated by the air injection pump, and synthesize a reference signal. The frequency domain synchronous decoupling process utilizes the orthogonal coherent detection principle to correlate local motion components with the reference signal, so that the displacement components generated by frequency-incoherent respiratory motion interference and intestinal peristalsis interference cancel each other out during the time integration process, thereby extracting the synchronous deformation component modulated by pressure pulsation under background physiological motion interference.

[0017] 3. This invention uses a central processing unit to calculate the compliance coefficient based on synchronous deformation components and pressure pulsation amplitude, and maps the compliance coefficient distribution matrix into a physical feature heatmap. Augmented reality fusion images overlay the physical feature heatmap onto digital video frames, and, in conjunction with Kalman filtering principles, perform inter-frame smoothing prediction. This aligns the biomechanical state of the digestive tract wall tissues with the anatomical structure images in spatial coordinates, completing the integrated quantitative assessment and visualization output of digestive tract endoscopy data. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention;

[0019] Figure 2 This is a schematic diagram of the integrated endoscopic diagnostic system of the present invention;

[0020] Figure 3 This is a time-domain waveform diagram of the pressure signal sequence and local displacement vector synchronization of the present invention;

[0021] Figure 4 This is a comparison chart of the lesion recognition performance of the experimental group and the control group of the present invention. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] See attached document Figure 2 This embodiment provides an integrated analysis method for gastrointestinal endoscopy data, which operates within an integrated endoscopic diagnostic system. The integrated endoscopic diagnostic system comprises a gastrointestinal endoscopy imaging unit, a pneumatic control unit, and a data gateway and processing unit.

[0024] The gastrointestinal endoscopic imaging unit comprises an endoscope lens, an image sensor, and a video signal processor. The image sensor uses a charge-coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) element. Deployed at the rear end of the endoscope lens, the image sensor captures reflected light signals from the inner wall of the gastrointestinal tract and converts them into electrical signals. The video signal processor encodes these electrical signals into a sequence of digital video frames. This sequence of digital video frames consists of image frames arranged sequentially over time.

[0025] During the acquisition of data from gastrointestinal endoscopy, the digital video frame sequence not only carries information about the geometric morphology of the tissue surface but also implicitly contains the tissue's absorption and reflection characteristics of spectral light signals. When the image sensor captures reflected light signals from the inner wall of the gastrointestinal tract, the complex and moist environment within the tract results in data containing numerous specular reflection bright spots and mucus interference noise. To ensure the accuracy of integrated analysis, the video signal processor performs piecewise linear stretching and adaptive gain control on each frame of the raw electrical signal before outputting the digital video frame sequence.

[0026] Specifically, the image sensor uses a Bayer array arrangement, and the raw data it acquires includes the raw light intensity of three components: red, green, and blue. Taking advantage of the hemoglobin-rich physical characteristics of the digestive tract lining, the video signal processor specifically enhances the data weight of the green component. This is because green light has a moderate penetration depth in mucosal tissue, allowing for a clearer mapping of microvascular distribution and their displacement due to pressure pulsations. The metadata for each digital video frame also includes the current lens focal length parameters and the brightness level of the illumination source. These focal length parameters, brightness levels, and image pixels together constitute a complete dataset of digestive tract endoscopy examination data.

[0027] Because the digestive tract has a non-linear tubular structure, the edge images acquired by the image sensor exhibit significant radial distortion. The video signal processor incorporates a distortion correction model based on Zhang Zhengyou's calibration method, performing coordinate transformation on each digital video frame to ensure linear consistency between geometric displacement in the image and actual tissue physical deformation. The digital video frame data, after geometric correction and color space conversion, serves as the underlying basis for subsequent local motion vector field calculations. To prevent motion blur during large-scale endoscope movement, the video signal processor dynamically adjusts the image sensor's exposure time based on the motion vector feedback of the current frame, ensuring the clarity of tissue texture within a single frame and thus improving the signal-to-noise ratio of the examination data.

[0028] The pneumatic control unit includes an air injection pump, an airflow channel, and a pressure monitoring sensor. The air injection pump is connected to the digestive tract lumen via the airflow channel. The air injection pump is responsible for delivering compressed air into the digestive tract lumen. During operation, the air injection pump generates pressure pulsations with a pneumatically modulated frequency within the airflow channel due to internal mechanical rotation or piston stroke. The pressure monitoring sensor is located at the output interface of the airflow channel. The pressure monitoring sensor is responsible for collecting instantaneous pressure values ​​within the airflow channel.

[0029] The instantaneous pressure values ​​collected by pressure monitoring sensors are a core component of the mechanical dimension of gastrointestinal endoscopy data. In practice, the air injection pump not only provides the basic static pressure to maintain lumen expansion, but its internal mechanical drive module also generates periodic micro-pressure oscillations. These micro-pressure oscillations are transmitted to the gastrointestinal lumen through the airflow channel, and their transfer function is influenced by the length and diameter of the airflow channel, as well as the compliance of the lumen's inner wall.

[0030] To fully characterize this dynamic process, the sampling frequency of the pressure monitoring sensor is set to more than 10 times the image sampling frequency to satisfy the Nyquist sampling theorem and prevent frequency aliasing caused by high-frequency pneumatic characteristics during data conversion. Simultaneously, to ensure that the video image sequence can effectively capture the microscopic tissue deformation induced by the air injection pump, the image sensor's acquisition frequency must meet the Nyquist sampling criterion, i.e., the acquisition frequency should be greater than twice the pneumatic modulation characteristic frequency of the air injection pump; or, by adjusting the speed of the air injection pump drive motor, the pneumatic modulation characteristic frequency can be locked within half the video acquisition bandwidth. Before entering the central processing unit, the acquired pressure signal sequence passes through a hardware-level pre-processing signal chain. This pre-processing signal chain includes a high-impedance differential amplifier to suppress power frequency interference and electromagnetic radiation noise introduced by the long-distance transmission of the endoscope cable.

[0031] The pressure dimension in gastrointestinal endoscopy data also needs to consider the delay effect caused by the compressibility of gases. The central processing unit stores an airflow transmission delay model, which compensates for the time difference in the transmission of pressure pulsations from the sensor position to the end of the lumen based on the physical length of the current airflow channel. This fine-tuning on the time axis ensures absolute synchronization of the pressure signal sequence with the tissue deformation sequence in the image in terms of physical causality.

[0032] Meanwhile, the quality of the pressure signal sequence is directly affected by the lumen's sealing performance. During the inspection, if the endoscope seal ring ages or the valve leaks, a non-stationary baseline drift will appear in the pressure signal. The data gateway and processing unit will monitor the root mean square value of the pressure signal sequence in real time. Once an abnormal pressure drop is detected, the data gateway and processing unit will automatically mark the airway leakage mask in the current inspection data packet and trigger the central processing unit to stop updating the compliance coefficient within the current analysis time window. At the same time, in the augmented reality fusion image of the video output interface, the corresponding area will be marked as a failure state or the heat map distribution of the previous stable frame will be maintained to prevent misdiagnosis caused by abnormal pressure fluctuations. This also reminds the subsequent algorithm modules to weight and weaken or remove the inspection data marked with the airway leakage mask, thereby ensuring the robustness of the integrated analysis results.

[0033] Because the lining of the digestive tract is subject to low-frequency displacement interference from respiratory movements and intestinal peristalsis during the examination, the pressure signal sequence formed by the instantaneous pressure values ​​collected by the pressure monitoring sensor contains both the pneumatic modulation characteristic frequency components generated by the air injection pump and the noise generated by physiological movements.

[0034] The data gateway and processing unit includes a synchronous trigger circuit, a multi-channel data acquisition card, a central processing unit (CPU), and a video output interface. The synchronous trigger circuit is connected to the image sensor and pressure monitoring sensor via a control bus. The synchronous trigger pulses generated by the circuit drive the image sensor and pressure monitoring sensor to sample under a unified time reference. These pulses ensure a clear mapping between the exposure start time of each digital video frame and the pressure sampling point in the pressure signal sequence on the time axis. The multi-channel data acquisition card receives the digital video frame sequence and the pressure signal sequence. The CPU connects the multi-channel data acquisition card to the video output interface. The CPU processes the received digital video frame sequence and pressure signal sequence, internally divided into: a local motion vector field calculation module for extracting tissue displacement; a frequency domain decoupling module for removing physiological noise; a compliance quantification evaluation module for calculating mechanical parameters; and an integrated feature fusion and real-time clinical feedback module for result display. The CPU ultimately outputs visualized results through the video output interface.

[0035] A synchronous trigger circuit coordinates the frequency of digital video frame sequences acquired by the image sensor and the frequency of pressure signal sequences acquired by the pressure monitoring sensor. The central processing unit (CPU) obtains the pneumatic modulation characteristic frequency generated by the air injection pump by performing spectral analysis on the pressure signal sequence. The CPU uses the pneumatic modulation characteristic frequency as a reference for subsequent phase-locked decoupling calculations. Each frame of the digital video frame sequence and the corresponding pressure value in the pressure signal sequence are encapsulated to construct a synchronous dataset for compliance coefficient quantization.

[0036] See attached document Figure 1 This invention provides an integrated analysis method for gastrointestinal endoscopy data, comprising the following steps:

[0037] The synchronization trigger circuit sends clock synchronization pulses to the image sensor and pressure monitoring sensor to ensure that the digital video frame sequence and the pressure signal sequence start recording at the same time. During the reception of the digital video frame sequence, the central processing unit assigns a unique exposure start time timestamp to each frame.

[0038] The central processing unit (CPU) retrieves the sampling point at the same moment in the pressure signal sequence using the exposure start time timestamp and extracts the instantaneous pressure value. If the sampling point time in the pressure signal sequence does not completely coincide with the exposure start time timestamp, the CPU calculates the instantaneous pressure value corresponding to the exposure start time timestamp using a linear interpolation algorithm.

[0039] The central processing unit pairs and encapsulates digital video frames with instantaneous pressure values ​​in chronological order to construct a synchronous dataset containing multiple sets of heterogeneous data pairs.

[0040] During the construction of the synchronized dataset, the central processing unit (CPU) employs a multi-threaded circular buffer mechanism. This mechanism unpacks and reassembles parallel video stream data from the video signal processor and serial pressure signal data from the multi-channel data acquisition card. Each heterogeneous data pair is defined as a structure in memory. The structure header contains a 64-bit nanosecond-level system timestamp, the middle contains the image pixel matrix index corresponding to the nanosecond-level system timestamp, and the tail contains the absolute pressure value and pressure change rate at the corresponding moment, obtained through interpolation.

[0041] This structured processing transforms fragmented gastrointestinal endoscopy data into a physically meaningful three-dimensional spatiotemporal force tensor. To address potential frame drops during endoscopy, the central processing unit (CPU) inserts a frame drop compensation flag into the header information of the synchronized dataset. If a missing digital video frame is detected at a certain moment, the CPU uses an optical flow prediction algorithm to synthesize a virtual frame from the preceding and following frames, and labels the confidence level of the virtual frame data to ensure temporal continuity when calculating the compliance coefficient.

[0042] To further optimize the storage and retrieval of massive amounts of endoscopic examination data, the central processing unit also performs online compression on the synchronous dataset. Image data is compressed intra-frame and inter-frame using the H.265 encoding protocol, while pressure signal sequences are compressed losslessly using differential pulse code modulation (DPCM). This efficient data encapsulation method enables the integrated endoscopic diagnostic system to support continuous clinical examination data recording for several hours while maintaining a high sampling rate, without causing data bus congestion.

[0043] In this process, the central processing unit transforms the discrete digital video frame sequence and the continuous pressure signal sequence into a structured data sequence with a physical correspondence, and divides the dataset into multiple continuous analysis time windows according to a preset duration, which serve as the unit for subsequent batch calculations.

[0044] To optimize computational efficiency, the central processing unit (CPU) allocates a contiguous address space for the synchronized dataset. This ensures that during subsequent local motion vector field calculations, the CPU can directly and rapidly access the corresponding digital video frames and instantaneous pressure values ​​based on the exposure start time timestamp. This transforms signals from different channels and sampling rates into time-aligned structured data, laying the foundation for subsequent extraction of tissue physical properties.

[0045] The synchronization dataset is stored in the random access memory connected to the central processing unit (CPU). Each set of data in the synchronization dataset represents the optical and mechanical state of the gastrointestinal tract wall during pressure pulsation. After the CPU completes the construction of the synchronization dataset, it passes the dataset as input parameters to the local motion vector field solution module to perform subsequent compliance coefficient quantization calculations.

[0046] The construction of the synchronization dataset eliminates data inaccuracies caused by hardware clock skew between the image sensor and the pressure monitoring sensor by forcibly combining visual dimension data with pressure dimension data. The structure of the synchronization dataset ensures the consistency of the displacement vector and the pressure carrier in time phase during subsequent phase-locked decoupling calculations.

[0047] The central processing unit (CPU) reads the pressure signal sequence from the synchronous dataset. The pressure signal sequence consists of instantaneous pressure values ​​collected by pressure monitoring sensors arranged in chronological order of sampling time. The CPU performs a discrete Fourier transform on the pressure signal sequence, converting the time-domain pressure signal sequence into a frequency-domain frequency component distribution.

[0048] The central processing unit (CPU) calculates the square of the magnitude of each frequency component, generating a power spectral density sequence of the pressure signal. The CPU pre-stores the operating frequency range of the air pump under normal operating conditions. The lower limit of the operating frequency range is set above the frequencies of respiratory motion interference and intestinal peristalsis interference. The CPU then retrieves the peak points of the power spectral density sequence within the operating frequency range.

[0049] The central processing unit (CPU) identifies and determines the frequency corresponding to the retrieved peak points as the pneumatic modulation characteristic frequency. The pneumatic modulation characteristic frequency represents the dominant frequency of pressure fluctuations generated by the mechanical operation of the air injection pump. The CPU transmits the value of the pneumatic modulation characteristic frequency to a local register for subsequent generation of a reference carrier signal.

[0050] The central processing unit extracts aerodynamic modulation characteristic frequencies to separate aerodynamic features with a fixed period from a pressure signal sequence containing respiratory motion interference, intestinal peristalsis interference, and random interference. The extracted aerodynamic modulation characteristic frequencies determine the distribution location of aerodynamically induced deformation in the frequency domain, providing a physical parameter basis for subsequent decoupling of deformation components from pressure fluctuations at the same frequency.

[0051] The generation of the reference signal is performed after the aerodynamic modulation characteristic frequency is determined. The central processing unit (CPU) retrieves the extracted aerodynamic modulation characteristic frequency from its local register. The CPU processes the pressure signal sequence using a bandpass filter, with the center frequency of the bandpass filter set as the aerodynamic modulation characteristic frequency. The CPU determines the pressure pulsation amplitude by extracting the peak envelope of the signal output from the bandpass filter.

[0052] The central processing unit (CPU) uses analytic signal transformation to calculate the phase evolution information of the pressure signal sequence. The CPU identifies the starting point of the synchronization trigger pulse and determines the initial phase of the pressure signal sequence at that point. The CPU then synthesizes a reference signal through numerical calculation. The expression for the reference signal is as follows:

[0053] ;

[0054] The symbols in the formula are defined as follows: This indicates the reference signal generated by the central processing unit; This represents the amplitude of the corresponding pressure pulsation in the pressure signal sequence; This indicates the extraction of a defined aerodynamic modulation characteristic frequency; Represents a continuous time variable calculated from the start time of the synchronization trigger pulse; This indicates the pressure signal sequence over time. The initial phase at the zero-value moment; Represents the sine function operator; Pi is a constant.

[0055] The reference signal generated by the central processing unit (CPU) has continuous time function characteristics, and the discrete sampling rate of the reference signal is consistent with the sampling frequency of the pressure monitoring sensor. The CPU writes the generated reference signal into a dedicated data buffer in the random access memory (RAM).

[0056] The reference signal provides a clean demodulated carrier for subsequent frequency domain decoupling of the deformation signal based on the phase-locked loop principle. The reference signal contains only the periodic pressure fluctuation characteristics generated by the mechanical operation of the air injection pump, excluding baseline drift components and random noise components present in the pressure signal sequence. By synthesizing the reference signal, the central processing unit establishes a physical excitation benchmark highly correlated with the digital video frame sequence in the time, frequency, and phase dimensions, providing a computational basis for subsequently separating incoherent motion interference from the local displacement vector field.

[0057] The central processing unit reads the first data from the synchronized dataset stored in random access memory. Frame of digital video and adjacent frames in the sequence Frame of digital video. The central processing unit (CPU) processes the first... The digital video frame undergoes gridding processing, dividing the first frame into its gridded form. A digital video frame is divided into multiple independent rectangular pixel blocks.

[0058] The central processing unit identifies the pixel brightness distribution within the rectangular pixel block as a matching feature, and in the first... Block matching calculations are performed within the search region defined in each frame of the digital video. To obtain motion parameters that exceed pixel resolution limitations, the central processing unit resamples the pixel grayscale values ​​of adjacent frames of the digital video using a bilinear trilinear interpolation algorithm and calculates the grayscale gradient information after resampling.

[0059] Based on the resampled grayscale gradient information, the central processing unit searches for and solves the sub-pixel position offset of each pixel coordinate in the digital video frame in the next digital video frame by iteratively minimizing the squared error function, and obtains the incremental displacement vector containing horizontal and vertical components.

[0060] The central processing unit (CPU) performs a global motion component stripping operation on the calculated incremental displacement vector. The CPU establishes an affine transformation model of the endoscope lens by identifying the consistency of pixel motion vectors in static anatomical structures at the edges of digital video frames. The CPU uses this affine transformation model to calculate the global translational and rotational components generated by the endoscope lens movement. The CPU subtracts these components from the incremental displacement vector to obtain the local motion components generated solely by the digestive tract wall tissue.

[0061] The central processing unit (CPU) performs subpixel-level deformation tracking cyclically within the analysis time window contained in the synchronized dataset until all digital video frames within the analysis time window have been traversed. The CPU stores each set of generated incremental displacement vectors containing local motion components into a displacement vector cache sequence in chronological order.

[0062] The generation process of the displacement vector buffer sequence is essentially a secondary mining of microscopic deformation information in gastrointestinal endoscopy data. In the actual gastrointestinal environment, the contrast of the image is low because the cavity is filled with translucent digestive fluid. To improve the calculation accuracy of the incremental displacement vector, the central processing unit performs homomorphic filtering on the digital video frames before performing block matching calculations to eliminate the influence of uneven lighting and enhance the texture features of tissue folds and capillaries.

[0063] The generated incremental displacement vector data includes not only spatial displacement values ​​but also a coherence score for the motion vector. The coherence score measures the matching quality of feature points between adjacent image frames. If the matching quality deteriorates in a region due to bubble occlusion or lens reflection, the coherence score of that region will decrease. The central processing unit automatically reduces the data weights of these low signal-to-noise ratio regions in subsequent ensemble analysis by establishing a dynamic weight mask, preventing erroneous displacements from causing biases in compliance calculations.

[0064] Subpixel-level deformation tracking also requires handling the non-rigid motion of the digestive tract wall. Unlike simple translation or rotation, tissue deformation under pressure exhibits local compression and stretching characteristics. Therefore, when calculating the incremental displacement vector, the central processing unit (CPU) incorporates local curvature analysis based on the Hessian matrix. Local curvature analysis can identify whether there are nonlinear deformation trends at feature points on the tissue surface. This in-depth motion data analysis enables the CPU to capture micron-level tissue vibrations that are imperceptible to the naked eye. These high-frequency vibrations are a direct manifestation of the pneumatic modulation of the air injection pump on the tissue and are a key data source for extracting the physical properties of the digestive tract wall.

[0065] The subpixel-level deformation tracking process enables the quantitative extraction of minute positional changes in the digestive tract wall, providing visual input parameters for subsequent phase-locked decoupling calculations of the time-series displacement signal and the pneumatic modulation reference signal.

[0066] The calculation of local displacement vectors is performed after subpixel-level deformation tracking is completed. The CPU reads all incremental displacement vectors generated within the analysis time window from the displacement vector cache sequence. For each pixel coordinate in the digital video frame sequence, the CPU performs vector superposition of the corresponding incremental displacement vectors generated within the analysis time window in chronological order of the digital video frames.

[0067] Vector superposition transforms discrete time-series incremental displacements into continuous displacements describing the trajectory of pixel coordinates within the analysis time window. The formula for calculating the local displacement vector is as follows:

[0068] ;

[0069] The symbols in the formula are defined as follows: Indicates pixel coordinates At this point, from the start time of the analysis time window to time... The local displacement vector; Indicates the first A digital video frame relative to its adjacent digital video frame in pixel coordinates The incremental displacement vector at that location; Represents the time-series index of a digital video frame; This indicates the total number of incremental displacement vectors contained within the analysis time window.

[0070] The central processing unit (CPU) constructs a discrete-time signal of the movement of the digestive tract wall tissue by calculating local displacement vectors. These local displacement vectors include high-frequency deformation components generated by the air injection pump, low-frequency displacement components generated by respiratory motion interference, and low-frequency displacement components generated by intestinal peristalsis interference. The CPU stores the calculated local displacement vectors in the random access memory connected to the CPU.

[0071] The local displacement vector serves as the computational input for subsequent frequency-domain decoupling of the deformation signal based on the phase-locked loop principle. The vector superposition calculation from the spatial domain to the time domain enables a complete record of the tissue's dynamic response process, providing the original physical quantities for extracting weak deformation features coherent with the reference signal.

[0072] After performing vector superposition of all pixel coordinates within the analysis time window, the central processing unit (CPU) generates a dynamic displacement field sequence with the same resolution as the digital video frame. Each data point in the dynamic displacement field sequence represents the real-time displacement state of the pixel coordinate position under aerodynamic pressure modulation. The CPU then transmits the dynamic displacement field sequence to the frequency domain decoupling module to eliminate noise generated by respiratory motion interference and intestinal peristalsis interference.

[0073] Frequency-domain synchronous decoupling is performed after the local displacement vector calculation and reference signal synthesis are completed. The central processing unit (CPU) reads the local displacement vector generated within the analysis time window from the random access memory (RAM). Simultaneously, the CPU retrieves the generated reference signal from the data buffer. The computational logic of frequency-domain synchronous decoupling utilizes the principle of orthogonal coherent detection to remove incoherent motion components from the local displacement vector.

[0074] The central processing unit (CPU) uses multiplication and accumulation operations to correlate the local displacement vector at each pixel coordinate with the reference signal at the corresponding time. The CPU then calculates the synchronous deformation component by performing time averaging on the multiplication results within the analysis time window. The formula for calculating the synchronous deformation component is as follows:

[0075] ;

[0076] The symbols in the formula are defined as follows: This represents the synchronous deformation component extracted at the pixel coordinates after frequency domain decoupling; This indicates that at pixel coordinates, the first... Local displacement vector corresponding to a frame of digital video; Indicates the timestamp at the start of the exposure. The corresponding reference signal value; This indicates the total number of digital video frames contained within the analysis time window; Indicates the first The timestamp of the exposure start time corresponding to the frame of the digital video.

[0077] During the frequency-domain synchronous decoupling process, the aerodynamically induced deformation components contained in the local displacement vector have completely consistent aerodynamic modulation characteristic frequencies with the reference signal. The displacement component frequencies caused by respiratory motion interference and intestinal peristalsis interference within the local displacement vector are both in the extremely low frequency range, and these frequencies are incoherent with the aerodynamic modulation characteristic frequencies. The central processing unit (CPU) performs synchronous demodulation operations, causing the frequency-incoherent displacement components generated by respiratory motion interference and intestinal peristalsis interference to cancel each other out during time integration, thereby extracting pure tissue deformation data modulated by the pulsating pressure of the injection pump from the complex background displacement.

[0078] The central processing unit (CPU) constructs a tissue characteristic parameter matrix in random access memory to record the full-field deformation amplitude. The CPU then stores the calculated synchronous deformation components into this matrix. These synchronous deformation components represent the true physical response amplitude of the digestive tract wall tissue under unit-cycle pneumatic excitation. After completing the frequency-domain synchronous decoupling processing of the full-field pixel coordinates, the CPU passes the synchronous deformation components to the subsequent compliance quantification evaluation module. This frequency-domain synchronous decoupling process enables the extraction of weak deformation signals in a non-contact state, eliminating artifact interference caused by biological motion during endoscopic examination.

[0079] Compliance coefficient quantization is performed after frequency domain synchronization decoupling is complete. The CPU reads the synchronization deformation component generated by the decoupling calculation from the organizational characteristic parameter matrix. The CPU retrieves the pressure pulsation amplitude determined when generating the reference signal from the local register. The CPU determines the compliance coefficient at the pixel coordinates by calculating the quotient of the vector magnitude of the synchronization deformation component and the pressure pulsation amplitude. The formula for calculating the compliance coefficient is as follows:

[0080] ;

[0081] The symbols in the formula are defined as follows: This represents the compliance coefficient at pixel coordinates; This represents the vector magnitude of the synchronous deformation component at pixel coordinates; This represents the amplitude of the corresponding pressure pulsation in the pressure signal sequence; Represents the coordinates of pixels in a digital video frame sequence.

[0082] During compliance coefficient quantization, the central processing unit (CPU) iteratively executes calculation logic across all pixel coordinates to generate a compliance coefficient distribution matrix. The compliance coefficient represents the magnitude of deformation displacement of the digestive tract wall tissue under unit pressure fluctuations. The numerical distribution of the compliance coefficient reflects the biomechanical state of different regions of the digestive tract wall tissue. The CPU stores the compliance coefficient distribution matrix in random access memory (RAM).

[0083] The compliance coefficient distribution matrix serves as the raw data input to the integrated feature fusion and real-time clinical feedback module. The compliance coefficient quantization calculation completes the transformation from dynamic displacement signals to tissue physical property parameters. The central processing unit extracts the compliance coefficients, converting the subtle tissue responses induced by the pneumatic carrier wave into clinically significant mechanical parameters, establishing a mapping relationship between optical and physical-mechanical characteristics.

[0084] The mapping of the physical characteristic heatmap is performed after the compliance coefficient quantization is completed. The central processing unit reads the compliance coefficient distribution matrix from the random access memory and traverses it to determine the maximum and minimum compliance values ​​it contains.

[0085] The central processing unit (CPU) performs a linear scaling calculation, mapping the compliance coefficient to a numerical range between zero and one, thereby obtaining a normalized compliance value. The CPU pre-stores a color mapping table, which records the correspondence between the normalized compliance value and the values ​​of the red, green, and blue components, and divides the normalized value range into a first preset interval and a second preset interval.

[0086] The central processing unit retrieves the corresponding red, green, and blue component values ​​from the color map table based on the normalized compliance values ​​of each pixel, and assigns them to the pixels at the corresponding pixel coordinates in the physical feature heatmap.

[0087] The spatial resolution of the physical feature heatmap is consistent with that of the digital video frames. Linear scaling calculations transform the numerical values ​​representing tissue mechanical information in the compliance coefficient distribution matrix into visual signals with color characteristics. In the physical feature heatmap, pixels with normalized compliance values ​​within a first preset range are assigned colors representing hard tissue, and pixels with normalized compliance values ​​within a second preset range are assigned colors representing soft tissue. The central processing unit (CPU) generates the physical feature heatmap to prepare the image input source for subsequent fusion and display of tissue physical property information with the digital video frame sequence.

[0088] Augmented reality overlay is executed after the physical feature heatmap mapping is completed. The CPU retrieves the physical feature heatmap from random access memory. The CPU retrieves digital video frames with the same exposure start time timestamp as the physical feature heatmap from the synchronization dataset. The CPU performs weighted pixel fusion calculations to generate the augmented reality fused image. The calculation formula for the augmented reality fused image is as follows:

[0089] ;

[0090] The symbols in the formula are defined as follows: This represents the pixel value at pixel coordinates in the augmented reality fused image; This represents the fusion coefficient used to adjust the transparency of the physical feature heatmap; This represents the pixel value at pixel coordinates in the physical feature heatmap; This represents the pixel value at pixel coordinates of a digital video frame; Represents the coordinates of pixels in a digital video frame sequence; Indicates the first The timestamp of the exposure start time corresponding to the frame of the digital video.

[0091] The central processing unit (CPU) controls the visual proportion of the physical feature heatmap above the digital video frame by adjusting the fusion coefficient values. The CPU then transmits the generated augmented reality fused image to the video output interface of the integrated endoscopic diagnostic system in real time. The overlay display transforms the abstract compliance coefficient distribution into an intuitive color distribution and aligns the color distribution with the spatial coordinates of the actual anatomical structure of the digestive tract lumen.

[0092] To enhance the practicality of clinical feedback, the integrated analysis method also involves multi-dimensional visualization post-processing of the outcome data. The augmented reality fusion images generated by the central processing unit (CPU) are not simply image overlays, but rather data filtered and graded according to clinical diagnostic needs. The CPU incorporates a pathological feature probability mapping table, which automatically identifies suspected sclerotic areas (such as tumor infiltration areas) or excessively relaxed areas (such as the periphery of diverticula) by analyzing abnormal gradients in compliance coefficients.

[0093] The continuity of the augmented reality display is ensured by an inter-frame smoothing filter. This filter utilizes the Kalman filter principle to predict and update the heatmap of physical features over time. Even when rapid endoscope movement causes instability in instantaneous data acquisition, the filter can use historical examination data for inertial compensation, ensuring smooth color transitions in the heatmap and avoiding visual fatigue caused by visual flicker.

[0094] The visualization output from the video output interface also includes a real-time mechanics histogram dashboard. This dashboard dynamically displays the statistical distribution of the compliance coefficient across the entire field of view, including the mean, variance, and skewness. These statistical indicators, such as mean, variance, and skewness, serve as derived components of the gastrointestinal endoscopy data, providing physicians with a quantitative reference for the overall pathological state of the tissue. For example, when the histogram's centroid shifts towards a low compliance region, the integrated endoscopic diagnostic system will issue a warning signal, indicating potential submucosal induration. This integrated feedback mechanism transforms raw physical detection signals into medical intelligence data with decision-making support capabilities.

[0095] Augmented reality overlay display enables the fusion of heterogeneous features between tissue biomechanical characteristics and optical image information. The fused augmented reality image, by overlaying a semi-transparent color layer onto the original image, assists physicians in identifying areas of abnormal tissue compliance in real time without altering their endoscopic examination procedures. After completing the fusion processing of the current digital video frame, the central processing unit continues to read the next digital video frame from the synchronized dataset, cyclically performing feature fusion processing to ensure the continuity of the augmented reality display effect.

[0096] To verify the effectiveness of an integrated analysis method for gastrointestinal endoscopic examination data, a comparative trial was conducted in a clinical setting. The subjects were two groups of patients with similar pathological features of submucosal gastric masses, designated as the experimental group and the control group. The experimental group underwent diagnosis using this invention, while the control group used a combination of traditional high-definition white light endoscopy and manual observation.

[0097] See attached document Figure 3 In the figure, the solid line represents the pressure signal sequence (kPa) collected by the pressure monitoring sensor and processed through the pre-processing link, while the dashed line represents the local displacement vector (micrometers) extracted by the local motion vector field calculation module within the analysis time window. The horizontal axis represents the continuous time variable (ms) calculated from the start of the synchronization trigger pulse.

[0098] Within the time interval of 40ms to 120ms, the pressure signal sequence exhibited periodic micro-pressure oscillations generated by the mechanical operation of the air injection pump, with the pressure pulsation amplitude stabilizing at approximately 0.5 kPa. The experimental group ensured time alignment between image sampling and pressure sampling through a synchronous trigger circuit. The displacement vector buffer sequence clearly recorded the sub-pixel positional offset of the tissue surface under micro-pressure oscillation excitation. The local displacement vector exhibited a fluctuation frequency highly consistent with the pressure signal sequence on the time axis, demonstrating that the frequency domain decoupling module can accurately capture the subtle tissue deformation affected by pneumatic modulation.

[0099] Comparative data shows that the control group, lacking a pressure dimension reference, was unable to distinguish between low-frequency displacements caused by respiratory motion interference and tissue compression deformation, resulting in displacement identification results containing a large amount of incoherent noise.

[0100] See attached document Figure 4The bar chart in the figure shows the comparison results between the experimental group and the control group in terms of lesion identification accuracy. The horizontal axis represents the evaluation indicators (diagnostic sensitivity, examination efficiency, deformation extraction accuracy), and the vertical axis represents the percentage values ​​(%). The experimental group used the compliance coefficient distribution matrix generated by the compliance coefficient quantification evaluation module to transform the abstract mechanical response into a physical feature heatmap. In the identification of sclerosing tumor infiltration areas, the experimental group achieved accurate delineation of lesion edges through abnormal gradient detection of the compliance coefficient. Augmented reality overlay display aligned the physical feature heatmap with digital video frames in spatial coordinates, assisting doctors in discovering two early submucosal micro-indurations missed by the control group.

[0101] Effect Comparison Statistics Table

[0102] Evaluation indicators control group experimental group Optimization effect Tissue deformation extraction accuracy ±15.2 micrometers ±2.1 micrometers Accuracy improved by 86.18% Motion noise suppression ratio 12dB 35dB Signal-to-noise ratio enhancement Lesion boundary identification error 3.5mm 0.8mm Positioning accuracy improved by 77.14% Diagnostic sensitivity 78.5% 94.2% The detection rate increased by 15.7%. Average inspection time 15.5min 11.2min Clinical efficiency improved by 27.7%.

[0103] The comparative statistics show that the video signal processor, by enhancing the weight of the green component data and combining it with a distortion correction model, provides high-quality underlying data for subsequent calculations. The central processing unit (CPU) uses the generated reference signal to perform frequency domain synchronization decoupling, successfully separating the pure synchronous deformation component from complex motions containing respiratory and intestinal peristalsis interference. The compliance coefficient quantification evaluation module establishes a mapping relationship between optical and physical / mechanical characteristics by calculating the quotient of the vector magnitude of the synchronous deformation component and the amplitude of pressure pulsations. Compared to traditional methods, the integrated analysis method for gastrointestinal endoscopy data demonstrates significant advantages in non-contact mechanical parameter measurement, lesion identification accuracy, and clinical diagnostic efficiency.

Claims

1. An integrated analysis method for gastrointestinal endoscopic examination data, characterized in that, Includes the following steps: The synchronous triggering circuit sends clock synchronization pulses to the image sensor and pressure monitoring sensor to synchronously acquire gastrointestinal endoscopy data consisting of digital video frame sequences and pressure signal sequences. The central processing unit assigns an exposure start time timestamp to the digital video frame sequence and extracts the instantaneous pressure value at the corresponding time from the pressure signal sequence to construct a synchronous dataset; The central processing unit performs a discrete Fourier transform on the pressure signal sequence in the synchronous dataset, and identifies the pneumatic modulation characteristic frequency generated by the air injection pump within a preset operating frequency range based on the generated power spectral density sequence. Using the aerodynamic modulation characteristic frequency to synthesize a reference signal, subpixel-level deformation tracking and global motion component stripping are simultaneously performed on the digital video frame sequence to obtain local motion components. The local motion component is frequency-domain synchronously decoupled using the reference signal to eliminate physiological motion interference and extract the synchronous deformation component modulated by the pressure pulsation of the injection pump. The compliance coefficient is determined based on the synchronous deformation component and the pressure pulsation amplitude. The physical feature heat map is then mapped and superimposed onto the digital video frame to generate an augmented reality fusion image for visualization.

2. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, During the construction of the synchronous dataset, the central processing unit uses a multi-threaded circular buffer mechanism to unpack and reassemble parallel video stream data and serial pressure signal data; Each heterogeneous data pair is defined as a structure in memory, and the central processing unit compensates for the time difference in the transmission of the pressure signal sequence from the pressure monitoring sensor location to the end of the lumen according to the airflow transmission delay model.

3. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, When acquiring the digital video frame sequence, the video signal processor enhances the weight of the green component data in the raw data acquired by the image sensor, taking into account the physical characteristics of the digestive tract wall being rich in hemoglobin, and performs coordinate transformation on each frame using a distortion correction model based on Zhang Zhengyou calibration method.

4. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, The central processing unit monitors the root mean square value of the pressure signal sequence in real time. Once an abnormal pressure drop is detected, it marks the airway leak mask in the current inspection data packet and performs weighted weakening or elimination on the inspection data marked with the airway leak mask.

5. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, The process of synthesizing the reference signal includes: The pressure signal sequence is processed using a bandpass filter to determine the pressure pulsation amplitude. The initial phase is determined by calculating the phase evolution information of the pressure signal sequence through analytical signal transformation. The reference signal is then synthesized based on the pressure pulsation amplitude, the aerodynamic modulation characteristic frequency, and the initial phase.

6. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, The sub-pixel level deformation tracking includes: The adjacent digital video frames are resampled using a bitrilinear interpolation algorithm. The squared error function is iteratively minimized based on the grayscale gradient information after resampling. The subpixel position offset of the pixel coordinates in the next frame is calculated and the incremental displacement vector is obtained. Simultaneously, local curvature analysis based on the Hessian matrix is ​​introduced to identify the nonlinear deformation trend of the tissue surface.

7. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 6, characterized in that, The process of acquiring the local motion components includes: Identify the consistency of pixel motion vectors in the static anatomical structure region at the edge of the digital video frame, and establish an affine transformation model for the endoscope lens; The global translation component and the global rotation component are calculated using the affine transformation model, and then the global translation component and the global rotation component are subtracted from the incremental displacement vector.

8. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, The frequency domain synchronization decoupling adopts the orthogonal coherent detection principle. Through multiplication and accumulation, the local motion component at each pixel coordinate is associated with the reference signal at the corresponding time. The calculation results within the analysis time window are processed by time averaging, so that the displacement components caused by incoherent respiratory motion interference and intestinal peristalsis interference cancel each other out during the time integration process.

9. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, The mapping of the physical feature heatmap includes: The compliance coefficient is linearly scaled to obtain a normalized compliance value. The normalized compliance value is then mapped to the corresponding red, green, and blue component values ​​according to a color mapping table. Suspected sclerotic or excessively relaxed areas are then marked using a pathological feature probability mapping table.

10. The integrated analysis method for gastrointestinal endoscopic examination data according to claim 1, characterized in that, In the process of generating the augmented reality fused image, the Kalman filter principle is used to predict and update the physical feature heatmap in the time series to achieve inter-frame smoothing; The visualization includes a real-time mechanics histogram dashboard, which dynamically displays the mean, variance, and skewness of the compliance coefficients across the entire field of view in the current field of view.

Images