Ulcerative colitis inflammation quantification system and method

Abstract

This application provides a system and method for quantifying the degree of inflammation in ulcerative colitis. The method involves acquiring white-light video images of the target patient's colonoscopy and photoacoustic signals generated by photoacoustic effects within the colon at multiple sampling times during a real-time colonoscopy examination. Based on the time-of-flight of the photoacoustic signals, signal components corresponding to different tissue depths within the target patient's colon are separated. These signal components are then fused with the white-light video images at each sampling time to generate a multi-layer image showing the mucosal blood supply distribution within the target patient's colon at each sampling time. Motion artifacts in the multi-layer images at each sampling time are calibrated. Inflammation-specific abnormally high-blood-supply areas at different depths of the target patient's colonic mucosa are extracted from all calibrated multi-layer images, and a quantitative score of the degree of colitis inflammation in the target patient is determined based on all abnormally high-blood-supply areas. Using the method of this application, multi-layered inflammatory features of the patient's colonic submucosa can be extracted.

Description

Technical Field

[0001] This application relates to the field of ulcerative colitis technology, and more specifically, to a system and method for quantifying the degree of inflammation in ulcerative colitis. Background Technology

[0002] Ulcerative colitis is a chronic, nonspecific inflammatory bowel disease. Its recurrent and persistent nature significantly impacts patients' quality of life and carries a long-term risk of cancer. Currently, colonoscopy is the gold standard for diagnosis and assessment of this disease. Clinicians primarily assess the severity subjectively by observing the morphological characteristics of the mucosal surface, such as loss of vascularity, mucosal erosion, and ulceration.

[0003] Current methods for quantifying the severity of ulcerative colitis primarily rely on visual scoring systems under white light endoscopy, which essentially provide a qualitative or semi-quantitative description of two-dimensional morphological changes on the mucosal surface. Although existing professional endoscopic scoring systems, such as the Mayo Criterion, UCEIS Criterion, and Geboes Pathological Criterion, explicitly require assessment of the longitudinal depth of inflammation infiltrating from the mucosal layer to the submucosa and even the muscularis propria at the pathological level, physicians lack in vivo imaging tools that can directly and clearly visualize the deep structures of the intestinal wall in actual clinical endoscopic procedures. Therefore, the assessment of infiltration depth heavily depends on physicians' empirical inferences based on subtle features of the surface image, such as mucosal tone, granularity, and bleeding tendency. This inference is highly subjective and subject to inter-observer variability, leading to inaccurate assessments of inflammation severity and hindering the optimization of treatment decisions.

[0004] Photoacoustic imaging technology can non-invasively acquire information on blood supply distribution within tissues by utilizing the selective light absorption characteristics of tissue hemoglobin. It combines the high contrast of optical imaging with the deep penetration of ultrasound imaging, making in vivo imaging of the deep structures of the digestive tract mucosa possible. However, current photoacoustic endoscopy imaging for assessing ulcerative colitis still suffers from the following core limitations: First, depth separation of photoacoustic signals relies on complex mechanical scanning, which is unsuitable for real-time colonoscopy. Second, the registration accuracy of white light-photoacoustic dual-modal images is insufficient and easily affected by endoscopic movement. Third, motion artifacts caused by intestinal peristalsis and endoscopic displacement are difficult to suppress effectively, resulting in insufficient accuracy in extracting deep blood supply features. Fourth, there is a lack of a quantitative mapping model for inflammatory infiltration depth based on pathological gold standards, allowing only qualitative observation and failing to objectively quantify the degree of inflammation. Therefore, achieving accurate, non-invasive, and quantitative extraction of multi-layered inflammatory features of the colonic mucosa in real-time colonoscopy has become a core technical challenge for the industry. Summary of the Invention

[0005] This application provides a system and method for quantifying the degree of inflammation in ulcerative colitis, which can non-invasively, accurately, and quantitatively extract multi-layered inflammatory features of the patient's colonic mucosa in a real-time colonoscopy setting, thereby achieving an objective quantitative assessment of the degree of inflammation in ulcerative colitis.

[0006] Firstly, this application provides a method for quantifying the degree of inflammation in ulcerative colitis, including: During real-time colonoscopy, white light video images of the target patient's colon are acquired at multiple sampling moments within the same target field of view using an image sensor, and photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation in the colon of the target patient are acquired using a linear array broadband ultrasound sensor integrated into the front end of the colonoscope. For each sampling moment, depth gating is performed based on the time-of-flight of the ultrasound signal using photoacoustic signals. Combined with wavelet transform frequency band optimization, signal components corresponding to different tissue depths in the colon of the target patient are separated. Based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and white light lens at each sampling moment, and combined with the real-time spatial pose of the ultrasound sensor at the sampling moment, each signal component is fused and registered with the white light video image at each sampling moment, thereby reconstructing a multi-layer image of the mucosal blood supply distribution in the colon of the target patient at each sampling moment. Based on the photometric consistency constraints between multi-layer images at each sampling time and the 3D motion estimation results of adjacent frames, combined with the cross-modal feature consistency loss output by the pre-trained feature constraint network, motion artifacts of multi-layer images at each sampling time are calibrated. Inflammation-specific abnormally high-blood-supply areas at different depths of the colonic mucosa of the target patient were extracted from all calibrated multilayer images. Based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis, the infiltration depth of colitis inflammation in the target patient was determined. Combining the depth of infiltration, the proportion of mucosal area with abnormally high blood supply, and the degree of abnormal blood supply, a quantitative score of the degree of colitis inflammation in the target patient is generated based on a quantitative model constructed using a clinically recognized endoscopic scoring system for ulcerative colitis.

[0007] In some embodiments, during real-time colonoscopy, white-light video images of the target patient's colonoscope are acquired at multiple sampling times within the same target field of view using an image sensor, and photoacoustic signals generated by the photoacoustic effect excited by pulsed laser in the target patient's colon are acquired using a linear broadband ultrasound sensor integrated into the front end of the colonoscope. Specifically, this includes: Multiple equally spaced sampling times are preset within the same target's field of view; At each sampling moment, the pulsed laser integrated into the front end of the colonoscope is synchronously controlled to emit nanosecond-level pulsed laser light into the target field of view of the colonic mucosa; The photoacoustic signal generated by the pulsed laser excitation of the tissue is synchronously received by a linear array broadband ultrasound sensor. The reflected light from the same target field of view illuminated by white light is simultaneously acquired by an image sensor to form a white light video image.

[0008] In some embodiments, depth gating is performed based on the time-of-flight of ultrasound waves using photoacoustic signals, combined with wavelet transform frequency band optimization, to separate signal components corresponding to different tissue depths within the colon of the target patient, specifically including: A sampling time is selected as the selected sampling time, and the effective frequency band signal corresponding to the colonic biological tissue is extracted from the photoacoustic signal at the selected sampling time. Perform a Hilbert transform on the effective frequency band signal to obtain an analytical signal containing amplitude and phase information; The analytical signal is first calibrated and time delay pre-compensated, then bandpass wavelet transform is performed for denoising and frequency band optimization to suppress tissue scattering noise and system electrical noise. At the same time, the signal start time and phase integrity are preserved by wavelet coefficient time domain alignment algorithm to avoid frequency domain processing from destroying the linear correspondence between TOF and depth, thus obtaining a photoacoustic time domain signal with optimized signal-to-noise ratio and no distortion in time domain features. Based on the constant propagation speed of ultrasound in the colonic soft tissue, the optimized time-domain signal is divided into depth-gated segments according to the time of flight; relying on the spatial positioning basis of the delay superposition reconstruction of linear array sensors, each time gate accurately corresponds to a preset tissue depth interval, thus obtaining depth layer signals corresponding to different tissue depths in the colon of the target patient. Inverse wavelet transform and amplitude normalization were performed on the signals at each depth layer to obtain the signal components corresponding to different tissue depths in the colon of the target patient at the selected sampling time. Continue to determine the signal components corresponding to different tissue depths within the colon of the target patient at the remaining sampling time.

[0009] In some embodiments, based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and the white light lens at each sampling time, and combined with the real-time spatial pose of the ultrasound sensor at the sampling time, each signal component is fused and registered with the white light video image at each sampling time, thereby reconstructing a multi-layer image of the blood supply distribution of the colonic mucosa of the target patient at each sampling time, specifically including: A sampling time is selected as the selected sampling time. All signal components at the selected sampling time are divided into signal component clusters corresponding to different depth layers of the colonic mucosa of the target patient. The depth layer division matches the standard anatomical structure of the colonic wall and is divided into three core layers: mucosa, submucosa, and muscularis. Blood vessel texture and mucosal structure features were extracted from white light video images at selected sampling times to obtain white light feature maps characterizing the surface morphology of colonic mucosa; The extrinsic parameter matrix is ​​dynamically corrected through a real-time extrinsic parameter calibration algorithm based on feature point matching. Based on the pre-calibrated dual-modal extrinsic parameter matrix of the white light lens and ultrasound sensor, the real-time spatial pose of the ultrasound sensor at the sampling time, and the endoscope deformation data collected by the bending sensor integrated into the front end of the colonoscope, the extrinsic parameter offset caused by endoscope bending is compensated. Then, the signal component clusters of each depth layer are reconstructed into a two-dimensional spatially distributed photoacoustic blood supply intensity map through a delayed superposition algorithm. The pixel size of the photoacoustic blood supply intensity map is completely consistent with the white light video image, and the field of view is strictly matched. Among them, the dual-modal extrinsic parameter matrix is ​​calibrated in advance through a checkerboard calibration plate to determine the rigid transformation relationship between the white light optical coordinate system and the ultrasound acoustic coordinate system. The real-time extrinsic parameter calibration algorithm has a response time of ≤10ms, ensuring registration accuracy and real-time performance. The photoacoustic blood supply intensity map of each depth layer is rigidly registered and locally non-rigid deformation registered with the white light feature map based on feature points to obtain multiple blood supply intensity values ​​of each depth layer after registration. All blood supply intensity values ​​of each depth layer are mapped to the three-dimensional coordinate system of the corresponding depth layer to obtain a multi-layer image of the blood supply distribution of the colonic mucosa of the target patient at the selected sampling time. Continue to determine the multi-layer images of the blood supply distribution in the colonic mucosa of the target patient at the remaining sampling time.

[0010] In some embodiments, motion artifacts in multi-layer images at each sampling time are calibrated based on photometric consistency constraints between multi-layer images at each sampling time, 3D motion estimation results of adjacent frames, and cross-modal feature consistency loss output by a pre-trained feature constraint network. Specifically, this includes: A time-series calibration network based on a three-dimensional convolutional neural network was constructed and pre-trained. The network pre-training was completed using a publicly available photoacoustic-white light dual-modal dataset of gastrointestinal endoscopy. After pre-training, the network weights were fixed. In clinical applications, only online inference was performed, and multi-layer images of multiple consecutive sampling times within the same target field of view were used as input sequences. Determine the photometric consistency loss between images at adjacent time points within the input sequence; By using a pre-trained feature extraction network, deep semantic features and shallow detail features are extracted from multi-layer images at each sampling time. The semantic and detail features at each sampling time are scale-aligned, and upsampling is performed through 3D transposed convolution to make the spatial size of deep semantic features completely consistent with that of shallow detail features. At the same time, the logical loophole of using photoacoustic signal reconstruction results with motion artifacts as ground values ​​is eliminated. Instead, the cross-modal feature benchmark output by the pre-trained feature constraint network and the multi-frame photometric consistency fusion results are used as ground value substitutes. Combined with the annotation information of the publicly available photoacoustic-white light dual-modal artifact-free gold standard dataset for gastrointestinal endoscopy, the cross-modal feature consistency loss between the aligned features and the ground value substitute is calculated to ensure the rationality of supervised training. By combining photometric consistency loss and cross-modal feature consistency loss, the three-dimensional motion field between adjacent frames is predicted through network forward inference. The three-dimensional motion field simultaneously suppresses non-rigid deformation caused by intestinal peristalsis and rigid motion artifacts caused by endoscopic displacement. The input sequence is subjected to motion compensation and resampling using the three-dimensional motion field to generate a calibrated multi-layer image sequence.

[0011] In some embodiments, inflammatory-specific abnormally high-blood-supply areas at different depths of the colonic mucosa of the target patient are extracted from all calibrated multilayer images, specifically including: Based on all calibrated multilayer images, mean blood supply intensity maps and blood supply fluctuation maps of different depth layers of colonic mucosa in the target patient were determined; We obtained the normal reference range of blood supply intensity at corresponding depth layers of colonic mucosa in healthy individuals of the same age group. Combined with a pre-trained inflammatory blood supply feature recognition model, we extracted the inflammatory-specific abnormally high blood supply areas at each depth layer of colonic mucosa in the target patient based on the average blood supply intensity map and blood supply fluctuation map of each depth layer, while excluding abnormal blood supply areas caused by non-inflammatory lesions.

[0012] In some embodiments, based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with a pathological infiltration hierarchy mapping model for ulcerative colitis, the infiltration depth of colitis inflammation in the target patient is determined, specifically including: For each depth layer, the infiltration intensity of colitis inflammation within each depth layer is determined based on the total area and average blood supply intensity of all inflammation-specific abnormally hypervascularized areas in each depth layer. Based on a pixel-level registered photoacoustic multilayer image and pathological slide dataset, and relying on the gold standard of ulcerative colitis pathology, an invasion level mapping model is constructed. The training of this model relies on: rapid fixation of tissue after ex vivo, reducing tissue shrinkage and deformation through cryosectioning, and a three-dimensional spatial registration algorithm based on vascular texture and anatomical landmarks to achieve accurate spatial correspondence between photoacoustic in vivo multilayer images and ex vivo pathological slides; data augmentation techniques such as rotation, flipping, and brightness adjustment are used to compensate for the insufficient clinical sample size; the model takes the invasion intensity of each depth layer as input and outputs the maximum invasion depth and weighted invasion depth of colitis inflammation in the target patient, wherein the weighted invasion depth is the weighted average of the invasion intensity of each depth layer and the median of the corresponding depth interval.

[0013] Secondly, this application provides a system for quantifying the degree of inflammation in ulcerative colitis, comprising: The acquisition module is used to acquire white light video images of the target patient's colonoscopy at multiple sampling moments within the same target field of view during real-time colonoscopy examination. It also acquires photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation in the colon of the target patient through an image sensor and a linear array broadband ultrasound sensor integrated into the front end of the colonoscope. The processing module is used to perform depth gating based on the time-of-flight of the ultrasound signal at each sampling time, combined with wavelet transform frequency band optimization, to separate the signal components corresponding to different tissue depths in the colon of the target patient, and based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and white light lens at each sampling time, combined with the real-time spatial pose of the ultrasound sensor at the sampling time, to fuse and register each signal component with the white light video image at each sampling time, thereby reconstructing a multi-layer image of the mucosal blood supply distribution in the colon of the target patient at each sampling time; The processing module is also used to calibrate the motion artifacts of the multi-layer images at each sampling time based on the photometric consistency constraints between the multi-layer images at each sampling time, the three-dimensional motion estimation results of adjacent frames, and the cross-modal feature consistency loss output by the pre-trained feature constraint network. The processing module is also used to extract inflammatory-specific abnormally high-blood-supply areas at different depths of the colonic mucosa of the target patient from all calibrated multilayer images, and to determine the infiltration depth of colitis inflammation in the target patient based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis. The execution module is used to combine the infiltration depth, the proportion of mucosal area with abnormally high blood supply, and the degree of abnormal blood supply, and to generate a quantitative score of the degree of colitis inflammation in the target patient based on a quantitative model constructed using a clinically recognized endoscopic scoring system for ulcerative colitis.

[0014] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing computer-executable code, and the processor being configured to acquire the computer-executable code and execute the above-described method for quantifying the degree of inflammation in ulcerative colitis.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for quantifying the degree of inflammation in ulcerative colitis.

[0016] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The system and method for quantifying the degree of inflammation in ulcerative colitis provided in this application firstly acquires white light video images and photoacoustic signals from within the tissue simultaneously within the same target field of view during real-time colonoscopy, constructing a dual-modal data foundation of surface morphology and deep functional information, breaking the limitation of existing clinical assessments that can only observe the mucosal surface; secondly, based on the core physical principle of photoacoustic imaging, time-of-flight depth gating technology optimizes the signal processing sequence (first phase calibration and time delay compensation, then wavelet denoising and frequency band optimization), preserves the integrity of temporal features through a wavelet coefficient time-domain alignment algorithm, avoids frequency domain processing from destroying the linear correspondence between TOF and depth, accurately separates photoacoustic signal components at different tissue depths, and eliminates redundant operations of depth gating of the original signal under linear array sensors. It improves layering accuracy based on the spatial positioning foundation of delay superposition reconstruction, and compensates for registration deviations caused by endoscope curvature through a pre-calibrated dual-modal extrinsic matrix + real-time extrinsic calibration algorithm. High-precision fusion and registration of photoacoustic signals and white light images reconstructs multi-layer images reflecting blood supply distribution at different depths, transforming the previously empirically inferred infiltration depth into directly visualizeable three-dimensional in vivo information. Furthermore, through a pre-trained time-series calibration network, the logical error of using artifact-laden photoacoustic reconstruction images as ground truth is eliminated. Instead, pre-trained feature benchmarks and multi-frame fusion results are used as ground truth replacements. Combined with photometric consistency constraints and cross-modal feature consistency loss, efficient suppression of motion artifacts in dynamic endoscopic scenes is achieved, ensuring the accuracy of deep blood supply feature extraction. Finally, through a mapping model based on pathological gold standards (trained on a pixel-level registered photoacoustic-pathological slide dataset), inflammation-specific abnormal high-blood supply areas are automatically extracted from the calibrated multi-layer images, objectively quantifying the inflammatory infiltration depth. Combined with a clinically recognized scoring system, a multi-dimensional quantitative score of inflammation severity is generated, completely replacing the traditional mode of indirect inference based on personal experience. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a method for quantifying the degree of inflammation in ulcerative colitis according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the determination of signal components at different tissue depths according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of areas with abnormally high blood supply to inflammation specificity, according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a system for quantifying the degree of inflammation in ulcerative colitis, as shown in some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a method for quantifying the degree of inflammation in ulcerative colitis, according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a method for quantifying the degree of inflammation in ulcerative colitis according to some embodiments of this application. The method for quantifying the degree of inflammation in ulcerative colitis mainly includes the following steps: In step 101, during the real-time colonoscopy examination, white light video images of the target patient's colonoscope are acquired by an image sensor at multiple sampling times within the same target field of view, and photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation in the colon of the target patient are acquired by a linear array broadband ultrasound sensor integrated into the front end of the colonoscope.

[0020] In some embodiments, this step can be implemented in the following manner: Multiple equally spaced sampling times are preset within the same target field of view, wherein the sampling interval is set to 20ms, and the total number of samples per field of view is not less than 20 frames, for example, 50 frames in this embodiment, to ensure that sampling is completed when the colonoscope endoscope is paused at the target field of view, and to avoid field of view shift caused by endoscope movement; At each sampling moment, the pulsed laser, linear broadband ultrasound sensor, and white light image sensor integrated into the front end of the colonoscope can be synchronously controlled through the hardware synchronization trigger module (Field Programmable Gate Array, FPGA) to achieve microsecond-level synchronous triggering of the three. Controlling the pulsed laser to emit nanosecond-level pulsed laser light to the target field of view of the colonic mucosa involves: sending a trigger command to a pulsed laser (e.g., a Q-switched Nd:YAG laser with a frequency-doubled output wavelength of 540 nm) integrated into a miniature photoacoustic endoscope probe within the colonoscope's biopsy channel. This pulsed laser is configured to emit pulsed laser light with a pulse width of 5-10 ns and an energy density below the biological tissue safety threshold. The laser pulse is transmitted through an optical fiber bundle within the colonoscope and collimated by a miniature optical lens at the end of the lens before illuminating the target field of view of the colonic mucosa. The laser energy density is set to 10 mJ / cm². 2 It meets the clinical safety guidelines for laser irradiation in gastrointestinal endoscopy. 540nm is a near-visible light band that can be effectively absorbed by hemoglobin. Moreover, this wavelength of laser has moderate penetration in colon tissue, which is suitable for the clinical needs of multilayer photoacoustic imaging of colonic mucosa (the characteristic absorption peaks of hemoglobin are 420nm, 540nm, and 577nm), and can achieve high-contrast imaging of mucosal blood supply. The linear array broadband ultrasound sensor (center frequency 10MHz, bandwidth 60%) integrated into the front end of the photoacoustic endoscope probe synchronously acquires the photoacoustic signal generated by the laser pulse excitation of tissue at each sampling moment, with the sampling frequency set to 100MHz. Synchronous illumination is achieved using white LED light sources, and reflected light from the same target field of view is simultaneously acquired by an image sensor to form a white light video image. The resolution of the white light image is set to 1920×1080, and the frame rate is strictly matched with the sampling time.

[0021] In step 102, for each sampling time, depth gating is performed based on the time-of-flight (TOF) of the ultrasound signal, combined with wavelet transform frequency band optimization, to separate signal components corresponding to different tissue depths in the colon of the target patient. Based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and the white light lens at each sampling time, and combined with the real-time spatial pose of the ultrasound sensor at the sampling time, each signal component is fused and registered with the white light video image at each sampling time, thereby reconstructing a multi-layer image of the mucosal blood supply distribution in the colon of the target patient at each sampling time.

[0022] In some embodiments, reference Figure 2 In this application, depth gating is performed based on the time-of-flight (TOF) of photoacoustic signals, combined with wavelet transform frequency band optimization, to separate signal components corresponding to different tissue depths within the colon of the target patient. This can be achieved through the following steps: In step 1021, a sampling time is selected as the selected sampling time, and the effective frequency band signal corresponding to the colonic biological tissue is extracted from the photoacoustic signal at the selected sampling time. In step 1022, the frequency band signal is subjected to Hilbert transform to obtain an analytical signal containing amplitude and phase information; In step 1023, the analytical signal is first subjected to phase calibration and time delay pre-compensation, and then bandpass wavelet transform is performed for denoising and frequency band optimization. Multi-scale wavelet coefficients are obtained through complex Morlet wavelet transform, and noise components are suppressed by adaptive thresholding. At the same time, the signal start time and phase integrity are preserved by wavelet coefficient time-domain alignment algorithm to avoid frequency domain processing from destroying the linear correspondence between TOF and depth. Finally, a photoacoustic time-domain signal with optimized signal-to-noise ratio and no distortion in time domain features is reconstructed. In step 1024, depth-gated segmentation is performed based on the ultrasonic time of flight to obtain depth layer signals corresponding to different tissue depths, and then the signal components corresponding to the depth are obtained by inverse wavelet transform. In step 1025, the signal components corresponding to different tissue depths in the target patient's colon are determined at the remaining sampling time.

[0023] In specific implementation, in step 1021, the photoacoustic signal at the selected sampling time is bandpass filtered. The passband frequency range of the filter is set to 1MHz-20MHz to retain the effective signal generated by the tissue photoacoustic effect, while filtering out low-frequency equipment vibration noise and high-frequency thermal noise. The result of the bandpass filtering is used as the effective frequency band signal corresponding to the colonic biological tissue.

[0024] In step 1023, the analytic signal is subjected to complex Morlet wavelet transform to obtain multi-scale wavelet coefficients. The noise component in the wavelet coefficients is suppressed by the adaptive threshold method to achieve signal denoising and frequency band optimization. Finally, the photoacoustic time domain signal with optimized signal-to-noise ratio is reconstructed.

[0025] In step 1024, based on the constant propagation speed of ultrasound in colonic soft tissue c=1540m / s, the tissue depth is calculated according to the formula d=c×t / 2, where t is the ultrasound flight time and d is the corresponding tissue depth. According to the standard anatomical structure of the human colon wall, three continuous depth-gated intervals are set: mucosa 0-0.5mm, submucosa 0.5-1.0mm, and muscularis propria 1.0-3.0mm, each depth interval corresponding to a fixed time gate range. The redundant layering operation of the original multi-channel signal is explicitly abandoned. Based on the spatial positioning foundation of the delay superposition reconstruction of the linear array sensor, the optimized time domain signal is divided according to the time gate range. The signal within each time gate is the depth layer signal of the corresponding tissue depth. Inverse wavelet transform is performed on each depth layer signal to obtain the signal components corresponding to different tissue depths at the selected sampling time.

[0026] It should be noted that the redundant layering operation on the original multi-channel signal refers to the layering operation on the original multi-channel photoacoustic signal in the existing technology, which requires each channel to be separately divided into depths before splicing. This is a redundant operation and is prone to introducing spatial positioning errors.

[0027] In some embodiments, the fusion registration and multi-layer image reconstruction in this step can be implemented using the following steps: Select a sampling time as the selected sampling time, and divide all signal components at the selected sampling time into signal component clusters corresponding to different depth layers of the colonic mucosa of the target patient, namely, the mucosal layer signal component cluster, the submucosa signal component cluster, and the muscular layer signal component cluster. For white-light video images at selected sampling times, the Frangi filtering algorithm based on the Hessian matrix is ​​used to extract vascular texture and mucosal structure features. The filtering scale is set to 1-3px, the vascular response threshold is 0.1, the Hessian matrix calculation window is 3×3 pixels, and the Gaussian smoothing σ value is 1.0. This algorithm enhances the edge features of blood vessels and folds on the mucosal surface, suppresses background noise, and obtains a white-light feature map characterizing the morphology of the colonic mucosal surface. Based on the pre-calibrated dual-modal extrinsic parameter matrix of the white light lens and ultrasound sensor, the real-time spatial pose of the ultrasound sensor at the sampling time, and the endoscopic deformation data collected by the bending sensor integrated into the colonoscope tip, the extrinsic parameter matrix is ​​dynamically corrected using a real-time extrinsic parameter calibration algorithm based on scale-invariant feature transform (SIFT) feature point matching. The contrast threshold for SIFT feature point extraction is set to 0.03, the edge threshold to 10, and the Euclidean distance threshold for feature point matching to 1.5. Using the matching feature point pairs of the white light feature map and the photoacoustic blood supply intensity map as a basis, the extrinsic parameter matrix correction amount ΔM is solved using the least squares method. The correction formula is Mcalibrated = Mstandard × (I + ΔM), where Mcalibrated is the corrected extrinsic parameter matrix, Mstandard is the pre-calibrated extrinsic parameter matrix, and I is the unit matrix. The matrix, ΔM, is the extrinsic parameter correction amount, with a calibration accuracy ≤0.1mm, to compensate for the extrinsic parameter offset caused by endoscope curvature (calibration accuracy ≤0.1mm). Then, the signal component clusters of each depth layer are reconstructed into a two-dimensional spatially distributed photoacoustic blood supply intensity map through a delay superposition algorithm. The pixel size of the photoacoustic blood supply intensity map is completely consistent with the white light video image, and the field of view is strictly matched. Among them, the dual-modal extrinsic parameter matrix is ​​calibrated in advance through a checkerboard calibration plate to determine the rigid transformation relationship between the white light optical coordinate system and the ultrasonic acoustic coordinate system, ensuring registration accuracy. The photoacoustic blood supply intensity maps of each depth layer are registered with the white light feature maps. First, rigid registration is completed by SIFT feature point matching, and then local non-rigid deformation registration is completed by thin spline interpolation algorithm. The interpolation nodes are set as SIFT matching feature points, the node neighborhood radius is 8px, and the deformation weight attenuation coefficient is 0.8. The algorithm fits the local non-rigid deformation law of intestinal soft tissue and eliminates the local registration error caused by deformation. In this system, the actual physical size of a single pixel in colonoscopy imaging is 0.02mm / px. After registration, the pixel offset is ≤0.5px (i.e., the actual physical offset is ≤0.01mm). The blood supply intensity value of each pixel in each depth layer image after registration is obtained. Create a three-dimensional spatial coordinate system with dimensions equal to the width of the white light feature map × the height of the white light feature map × the number of depth layers (3 layers). For each depth layer, fill the corresponding blood supply intensity value into the corresponding position in the three-dimensional coordinate system to finally obtain a multi-layer image of the blood supply distribution of the colonic mucosa of the target patient at the selected sampling time. Continue to determine the multi-layer images of the blood supply distribution in the colonic mucosa of the target patient at the remaining sampling time.

[0028] In step 103, motion artifacts of multi-layer images at each sampling time are calibrated based on photometric consistency constraints between multi-layer images at each sampling time, 3D motion estimation results of adjacent frames, and cross-modal feature consistency loss output by the pre-trained feature constraint network.

[0029] In some embodiments, this step may be implemented using the following steps: A time-series calibration network based on a 3D convolutional neural network was constructed and pre-trained. The network is based on the 3DU-Net architecture and includes a 4-layer downsampling encoder and a 4-layer upsampling decoder. The convolutional kernel size is 3×3×3, the stride is 2, and the padding is 1. The encoder activation function is LeakyReLU (slope 0.2), and the decoder activation function is ReLU. The network includes a 3D motion estimation module and a 3D motion correction module. The network pre-training uses a publicly available photoacoustic-white light bimodal dataset for gastrointestinal endoscopy. The batch size is 8, the initial learning rate is 5e-5, the optimizer is AdamW, and the loss function is a weighted sum of photometric consistency loss and cross-modal feature consistency loss (weight ratio 1:1). After pre-training, the network weights are fixed. In clinical applications, only forward inference is performed. The pre-training is completed using a publicly available photoacoustic-white light bimodal dataset for gastrointestinal endoscopy. During the pre-training process, labeled artifact-free gold standard images are used as supervision. After pre-training, the network weights are fixed. In clinical applications, only forward inference is performed, and no real-time training is required, which meets the timeliness requirements of real-time colonoscopy. The network takes multiple consecutive sampling times within the same target field of view as the input sequence. Each sampling time image is a three-dimensional data volume with a size of width W × height H × depth L = 3. The network stacks all three-dimensional data volumes along the time dimension to form a four-dimensional tensor (W × H × L × T) as the input. The input pixel values ​​are normalized to the range of [0,1]. Calculate the photometric consistency loss between adjacent time-series images: For two adjacent time-series t and t+1, the network predicts the three-dimensional motion field V from the image at time-t+1 to the image at time-t. The motion field is used to resample the image at time-t+1 to obtain a deformed image I'{t+1} aligned with the image at time-t. The photometric consistency loss is defined as the sum of the Charbonnier losses of all corresponding pixels between I{t} and I'{t+1}. By using a pre-trained feature extraction network, deep semantic features and shallow detail features are extracted from multi-layer images at each sampling time. The shallow detail features capture subtle information such as blood vessel edges and mucosal texture, while the deep semantic features represent high-level semantic information such as blood vessel branch morphology and deep layer association patterns. The semantic and detail features at each sampling time are scale-aligned, and upsampling is performed through 3D transposed convolution to make the spatial size of deep semantic features completely consistent with that of shallow detail features. At the same time, the logical loophole of using photoacoustic signal reconstruction results with motion artifacts as ground values ​​is eliminated. Instead, the cross-modal feature benchmark output by the pre-trained feature constraint network and the multi-frame photometric consistency fusion result are used as ground values. Combined with the annotation information of the publicly available photoacoustic-white light dual-modal artifact-free gold standard dataset for gastrointestinal endoscopy, the cross-modal feature consistency loss between the aligned features and the ground value replacement is calculated to ensure the rationality and accuracy of supervised training. By combining photometric consistency loss and cross-modal feature consistency loss, the three-dimensional motion field between adjacent frames is predicted through network forward inference. At the same time, a first-order smoothing regularization constraint is applied to the three-dimensional motion field to ensure the continuity of the motion field and avoid local excessive deformation. The photoacoustic signal quality of all frames in the input sequence is evaluated, and the frame with the highest signal-to-noise ratio is selected as the spatial reference. For images at non-reference times, all three-dimensional motion fields from that time to the reference time are accumulated to obtain the composite motion field. The original image is resampled using a three-dimensional bilinear interpolation sampler. The target resolution of the resampling is set to be consistent with the original multi-layer image, and the interpolation step size is 1px. The gray value fitting method of bilinear interpolation is used to align the non-reference time images to the spatial coordinate system of the reference time to complete the motion artifact calibration. After calibration, the gray value of the image is normalized to the range of [0,1], and a three-dimensional Gaussian filter with σ=0.5 is performed to remove interpolation artifacts. After calibration, the image is normalized to gray value and filtered by three-dimensional Gaussian filter to remove interpolation artifacts. Finally, the calibrated multi-layer image sequence is output.

[0030] In step 104, inflammation-specific abnormally high-blood-supply areas at different depths of the colonic mucosa of the target patient are extracted from all calibrated multilayer images. Based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis, the infiltration depth of colitis inflammation in the target patient is determined.

[0031] In some embodiments, reference Figure 3 The extraction of areas with abnormally high blood supply specific to inflammation in this step can be achieved using the following steps: In step 1041, based on all calibrated multilayer images, the average blood supply intensity map and blood supply fluctuation map of different depth layers of the colonic mucosa of the target patient are determined; In step 1042, the normal reference range of healthy individuals and the pre-trained inflammation recognition model are combined to extract the inflammation-specific abnormal high blood supply areas of each depth layer.

[0032] In specific implementation, in step 1041, for each depth layer, the two-dimensional blood supply distribution image of that depth layer at all sampling times is extracted to form three-dimensional data with dimensions of width × height × sampling time; the arithmetic mean of the blood supply intensity value of each pixel is calculated along the time axis to obtain the average blood supply intensity map of that depth layer; at the same time, the sample standard deviation of the blood supply intensity value of each pixel is calculated to obtain the blood supply fluctuation map of that depth layer.

[0033] In step 1042, firstly, a normal reference range for blood supply intensity at the corresponding depth layer of the colonic mucosa in healthy individuals of the same age group is obtained. This reference range is established based on endoscopic photoacoustic imaging data of a large sample of healthy individuals. A preset blood supply variability threshold is used to distinguish between persistent high blood supply caused by inflammation and transient fluctuations caused by noise or movement. Simultaneously, a pre-trained inflammation blood supply feature recognition model is used. This model is based on the ResNet18 architecture. The input is a fusion feature map of the average blood supply intensity map and the blood supply variability map of the depth layer. The average blood supply intensity map and the blood supply variability map are first fused at the channel level to form a dual-channel feature map before being input into the model. The output is inflammation / The non-inflammatory pixel-level classification results were obtained. The batch size of the network training was 16, the initial learning rate was 1e-4, the optimizer was Adam, and the loss function was cross-entropy loss. The training dataset was the photoacoustic imaging dataset labeled with the gold standard for ulcerative colitis pathology. The number of training iterations was 100 epochs. The accuracy of the model in identifying inflammatory regions was ≥95%. It can effectively exclude abnormal blood supply regions caused by non-inflammatory lesions such as polyps, vascular malformations, and tumors. Feature recognition was performed on the fused feature map to extract inflammation-specific abnormally high blood supply regions. At the same time, abnormally high blood supply regions caused by non-inflammatory lesions such as polyps, vascular malformations, and tumors were excluded. Finally, the inflammation-specific abnormally high blood supply regions of each depth layer were obtained.

[0034] In some embodiments, determining the depth of inflammatory infiltration in this step can be achieved using the following steps: For each depth layer, the total number of pixels in all inflammation-specific abnormal high blood supply regions within that depth layer is counted as the total area of ​​the high blood supply region; at the same time, the average blood supply intensity corresponding to all pixels in the high blood supply region within that depth layer is counted, and the product of the total area and the average blood supply intensity is calculated as the infiltration intensity of colitis inflammation within that depth layer. Based on a pixel-level registered photoacoustic multilayer image and pathological slide dataset, and relying on the gold standard of ulcerative colitis pathology, an invasion level mapping model was constructed. This model is a lightweight fully connected network. The input is the invasion intensity (3D feature vector) of the mucosa, submucosa, and muscularis propria. There are two hidden layers (64 and 32 neurons respectively, with ReLU activation function). The output is the maximum invasion depth (classification value) and the weighted invasion depth (regression value). The model was trained on a pixel-level registered photoacoustic multilayer image-pathological slide dataset. During training, the batch size was set to 32, the initial learning rate was 1e-3, the optimizer was Adam, the classification loss was cross-entropy loss, and the regression loss was MSE loss (weighted ratio 1:1). Data augmentation techniques such as rotation, flipping, and brightness adjustment were used to improve the generalization ability. The invasion depth prediction error of the model is ≤ The core of the 0.5mm model lies in constructing a pixel-level registered photoacoustic multilayer image and pathological slide dataset: rapid fixation of ex vivo tissue and cryosectioning techniques reduce tissue shrinkage and deformation; a three-dimensional spatial registration algorithm based on vascular texture and anatomical landmarks is used to achieve precise spatial correspondence between in vivo photoacoustic multilayer images and ex vivo pathological slides; data augmentation techniques such as rotation, flipping, and brightness adjustment are employed to compensate for the insufficient clinical sample size; the model takes the infiltration intensity of each depth layer as input and outputs two core indicators: one is the maximum infiltration depth of inflammation, i.e., the deepest anatomical level of inflammatory cell invasion; the other is the weighted infiltration depth, calculated as: weighted infiltration depth = Σ(infiltration intensity of each depth layer × median of the corresponding depth interval) / Σ(infiltration intensity of each depth layer), and finally uses the weighted infiltration depth as the core infiltration depth indicator of colitis inflammation in the target patient.

[0035] In step 105, combining the infiltration depth, the proportion of mucosal area with abnormally high blood supply, and the degree of abnormal blood supply, a quantitative score of the degree of colitis inflammation in the target patient is generated based on a quantitative model constructed using the clinically recognized endoscopic scoring system for ulcerative colitis.

[0036] In practice, a multi-dimensional quantitative scoring model is constructed based on clinically recognized scoring systems such as the Mayo Criterion and the UCEIS scoring system. The model inputs include: the inflammatory-weighted depth of infiltration, the proportion of the mucosal area with abnormally high blood supply within the target field of view, and the ratio of the average blood supply intensity of the abnormally high blood supply area to the normal reference value. Through training with large-sample clinical data, the input indicators are mapped to a standardized quantitative score of 0-10, with higher scores indicating more severe inflammation. Specifically, 0-3 indicates mild inflammation, 4-6 indicates moderate inflammation, and 7-10 indicates severe inflammation. This quantitative score shows high consistency with the clinical gold standard score, enabling objective and repeatable quantitative assessment of the degree of inflammation in ulcerative colitis.

[0037] In another aspect, in some embodiments, this application provides a system for quantifying the degree of inflammation in ulcerative colitis, with reference to... Figure 4 The system includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 is used to acquire white light video images of the target patient's colonoscopy at multiple sampling times within the same target field of view during real-time colonoscopy examination. It also acquires photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation in the colon of the target patient through an image sensor and a linear array broadband ultrasound sensor integrated into the front end of the colonoscopy. The processing module 402 is used to perform depth gating based on the time-of-flight (TOF) of the photoacoustic signal for each sampling moment, combine wavelet transform frequency band optimization, separate the signal components corresponding to different tissue depths in the colon of the target patient, and, based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and the white light lens at each sampling moment, and combined with the real-time spatial pose of the ultrasound sensor at the sampling moment, fuse and register each signal component with the white light video image at each sampling moment, thereby reconstructing a multi-layer image of the mucosal blood supply distribution in the colon of the target patient at each sampling moment; The processing module 402 is also used to calibrate the motion artifacts of the multi-layer images at each sampling time based on the photometric consistency constraints between the multi-layer images at each sampling time, the three-dimensional motion estimation results of adjacent frames, and the cross-modal feature consistency loss output by the pre-trained feature constraint network. The processing module 402 is also used to extract inflammatory-specific abnormally high blood supply areas at different depths of the colonic mucosa of the target patient from all calibrated multilayer images, and to determine the infiltration depth of colitis inflammation in the target patient based on the spatial distribution and blood supply intensity characteristics of all abnormally high blood supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis. The execution module 403 is used to combine the infiltration depth, the proportion of mucosal area with abnormally high blood supply, and the degree of abnormal blood supply, and generate a quantitative score of the degree of colitis inflammation in the target patient based on a quantitative model constructed according to the clinically recognized endoscopic scoring system for ulcerative colitis.

[0038] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described method for quantifying the degree of inflammation in ulcerative colitis.

[0039] In some embodiments, reference Figure 5 The figure is a schematic diagram of a computer device for implementing a method for quantifying the degree of inflammation in ulcerative colitis, according to some embodiments of this application. The method for quantifying the degree of inflammation in ulcerative colitis in the above embodiments can be achieved through… Figure 5The computer device 500 shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0040] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0041] The communication bus 502 can be used to transmit information between the aforementioned components.

[0042] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0043] The memory 503 stores program code for executing the scheme of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. In the above embodiments, the method for quantifying the degree of inflammation in ulcerative colitis can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0044] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0045] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0046] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0047] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for quantifying the degree of inflammation in ulcerative colitis.

[0048] In summary, the technical solution disclosed in this application achieves accurate reconstruction of multi-layer blood supply images of the colonic wall by simultaneously acquiring white light endoscopic images and photoacoustic signals, based on time-of-flight depth gating, ensuring the accuracy of deep features through pre-calibrated dual-modal registration and motion artifact calibration, and finally achieving objective quantification of inflammatory infiltration depth and inflammatory degree by combining a pathological prior model. This solves the core problem in the prior art that UC inflammation assessment is highly subjective and cannot non-invasively assess deep submucosal inflammation, and has clear clinical value and industrial applicability.

[0049] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0050] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for quantifying the degree of inflammation in ulcerative colitis, characterized in that, include: During real-time colonoscopy, white light video images of the target patient's colon are acquired at multiple sampling moments within the same target field of view using an image sensor, and photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation in the colon of the target patient are acquired using a linear array broadband ultrasound sensor integrated into the front end of the colonoscope. For each sampling moment, depth gating is performed based on the time-of-flight of the ultrasound signal using photoacoustic signals. Combined with wavelet transform frequency band optimization, signal components corresponding to different tissue depths in the colon of the target patient are separated. Based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and white light lens at each sampling moment, and combined with the real-time spatial pose of the ultrasound sensor at the sampling moment, each signal component is fused and registered with the white light video image at each sampling moment, thereby reconstructing a multi-layer image of the mucosal blood supply distribution in the colon of the target patient at each sampling moment. Based on the photometric consistency constraints between multi-layer images at each sampling time and the 3D motion estimation results of adjacent frames, combined with the cross-modal feature consistency loss output by the pre-trained feature constraint network, motion artifacts of multi-layer images at each sampling time are calibrated. Inflammation-specific abnormally high-blood-supply areas at different depths of the colonic mucosa of the target patient were extracted from all calibrated multilayer images. Based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis, the infiltration depth of colitis inflammation in the target patient was determined. Combining the depth of infiltration, the proportion of mucosal area with abnormally high blood supply, and the degree of abnormal blood supply, a quantitative score of the degree of colitis inflammation in the target patient is generated based on a quantitative model constructed using a clinically recognized endoscopic scoring system for ulcerative colitis.

2. The method as described in claim 1, characterized in that, During real-time colonoscopy, white-light video images of the target patient's colon are acquired at multiple sampling moments within the same target field of view using an image sensor. Additionally, photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation within the target patient's colon are acquired using a linear broadband ultrasound sensor integrated into the front end of the colonoscope. Specifically, this includes: Multiple equally spaced sampling times are preset within the same target's field of view; At each sampling moment, the pulsed laser integrated into the front end of the colonoscope is synchronously controlled to emit nanosecond-level pulsed laser light into the target field of view of the colonic mucosa; The photoacoustic signal generated by the pulsed laser excitation of the tissue is synchronously received by a linear array broadband ultrasound sensor. The reflected light from the same target field of view illuminated by white light is simultaneously acquired by an image sensor to form a white light video image.

3. The method as described in claim 1, characterized in that, Depth gating based on time-of-flight ultrasound signals from photoacoustic signals, combined with wavelet transform frequency band optimization, separates signal components corresponding to different tissue depths within the colon of the target patient, specifically including: A sampling time is selected as the selected sampling time, and the effective frequency band signal corresponding to the colonic biological tissue is extracted from the photoacoustic signal at the selected sampling time. Perform a Hilbert transform on the effective frequency band signal to obtain an analytical signal containing amplitude and phase information; The analytical signal is first subjected to phase calibration and time delay pre-compensation, and then bandpass wavelet transform is performed for denoising and frequency band optimization to suppress tissue scattering noise and system electrical noise. At the same time, the signal start time and phase integrity are preserved by wavelet coefficient time-domain alignment algorithm, resulting in a photoacoustic time-domain signal with optimized signal-to-noise ratio and no distortion in time-domain features. Based on the constant propagation speed of ultrasound in colonic soft tissue, the optimized time-domain signal is divided into depth-gated segments according to the time of flight. Relying directly on the spatial positioning basis of the delay superposition reconstruction of linear array sensors, each time gate accurately corresponds to a preset tissue depth interval, thus obtaining depth layer signals corresponding to different tissue depths in the colon of the target patient. Inverse wavelet transform and amplitude normalization were performed on the signals at each depth layer to obtain the signal components corresponding to different tissue depths in the colon of the target patient at the selected sampling time. Continue to determine the signal components corresponding to different tissue depths within the colon of the target patient at the remaining sampling time.

4. The method as described in claim 1, characterized in that, Based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and white light lens at each sampling time, and combined with the real-time spatial pose of the ultrasound sensor at each sampling time, each signal component is fused and registered with the white light video image at each sampling time, thereby reconstructing a multi-layer image of the blood supply distribution of the colonic mucosa of the target patient at each sampling time, specifically including: A sampling time is selected as the selected sampling time. All signal components at the selected sampling time are divided into signal component clusters corresponding to different depth layers of the colonic mucosa of the target patient. The depth layer division matches the standard anatomical structure of the colonic wall and is divided into three core layers: mucosa, submucosa, and muscularis. Blood vessel texture and mucosal structure features were extracted from white light video images at selected sampling times to obtain white light feature maps characterizing the surface morphology of colonic mucosa; Based on the pre-calibrated dual-modal extrinsic matrix, the extrinsic matrix is ​​dynamically corrected through a real-time extrinsic calibration algorithm. The real-time spatial pose of the ultrasound sensor at the sampling time is combined with the endoscope deformation data collected by the endoscope bending sensor. Then, the signal component clusters of each depth layer are reconstructed into a two-dimensional spatially distributed photoacoustic blood supply intensity map through a delay superposition algorithm. The spatial resolution and field of view of the photoacoustic blood supply intensity map are strictly matched with the white light video image. The photoacoustic blood supply intensity map of each depth layer is rigidly registered and locally non-rigid deformation registered with the white light feature map based on feature points to obtain multiple blood supply intensity values ​​of each depth layer after registration. All blood supply intensity values ​​of each depth layer are mapped to the three-dimensional coordinate system of the corresponding depth layer to obtain a multi-layer image of the blood supply distribution of the colonic mucosa of the target patient at the selected sampling time. Continue to determine the multi-layer images of the blood supply distribution in the colonic mucosa of the target patient at the remaining sampling time.

5. The method as described in claim 1, characterized in that, Based on the photometric consistency constraints between multi-layer images at each sampling time, the 3D motion estimation results of adjacent frames, and the cross-modal feature consistency loss output by the pre-trained feature constraint network, motion artifacts in multi-layer images at each sampling time are calibrated, specifically including: A time-series calibration network based on a three-dimensional convolutional neural network was constructed and pre-trained. The network pre-training was completed using a publicly available photoacoustic-white light dual-modal dataset of gastrointestinal endoscopy. After pre-training, the network weights were fixed. In clinical applications, only online inference was performed, and multi-layer images of multiple consecutive sampling times within the same target field of view were used as input sequences. Determine the photometric consistency loss between images at adjacent time points within the input sequence; By using a pre-trained feature extraction network, deep semantic features and shallow detail features are extracted from multi-layer images at each sampling time. The semantic features and detail features at each sampling time are scale-aligned to eliminate the logical loophole of using the photoacoustic signal reconstruction result with motion artifacts as the ground truth. The cross-modal feature benchmark output by the pre-trained feature constraint network and the multi-frame photometric consistency fusion result are used as the ground truth. The cross-modal feature consistency loss is calculated by combining the annotation information of the publicly available dual-modal artifact-free gold standard dataset for gastrointestinal endoscopy. By combining photometric consistency loss and cross-modal feature consistency loss, the three-dimensional motion field between adjacent frames is predicted through network forward inference. The three-dimensional motion field simultaneously suppresses non-rigid deformation caused by intestinal peristalsis and rigid motion artifacts caused by endoscopic displacement. The input sequence is subjected to motion compensation and resampling using the three-dimensional motion field to generate a calibrated multi-layer image sequence.

6. The method as described in claim 1, characterized in that, From all calibrated multilayer images, inflammatory-specific abnormal high-blood-supply areas at different depths of the target patient's colonic mucosa were extracted, specifically including: Based on all calibrated multilayer images, mean blood supply intensity maps and blood supply fluctuation maps of different depth layers of colonic mucosa in the target patient were determined; We obtained the normal reference range of blood supply intensity at corresponding depth layers of colonic mucosa in healthy individuals of the same age group. Combined with a pre-trained inflammatory blood supply feature recognition model, we extracted the inflammatory-specific abnormally high blood supply areas at each depth layer of colonic mucosa in the target patient based on the average blood supply intensity map and blood supply fluctuation map of each depth layer, while excluding abnormal blood supply areas caused by non-inflammatory lesions.

7. The method as described in claim 1, characterized in that, Based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis, the infiltration depth of colitis inflammation in the target patient is determined, specifically including: For each depth layer, the infiltration intensity of colitis inflammation within each depth layer is determined based on the total area and average blood supply intensity of all inflammation-specific abnormally hypervascularized areas in each depth layer. Based on a pixel-level registered photoacoustic multilayer image and pathological slide dataset, and relying on the gold standard of ulcerative colitis pathology, an invasion level mapping model is constructed. The model takes the invasion intensity of each depth layer as input and outputs the maximum invasion depth and weighted invasion depth of colitis inflammation in the target patient. The weighted invasion depth is the weighted average of the invasion intensity of each depth layer and the median of the corresponding depth interval.

8. A system for quantifying the degree of inflammation in ulcerative colitis, characterized in that, include: The acquisition module is used to acquire white light video images of the target patient's colonoscopy at multiple sampling moments within the same target field of view during real-time colonoscopy examination. It also acquires photoacoustic signals generated by the photoacoustic effect of pulsed laser excitation in the colon of the target patient through an image sensor and a linear array broadband ultrasound sensor integrated into the front end of the colonoscope. The processing module is used to perform depth gating based on the time-of-flight of the ultrasound signal at each sampling time, combined with wavelet transform frequency band optimization, to separate the signal components corresponding to different tissue depths in the colon of the target patient, and based on the pre-calibrated dual-modal extrinsic parameter matrix of the ultrasound sensor and white light lens at each sampling time, combined with the real-time spatial pose of the ultrasound sensor at the sampling time, to fuse and register each signal component with the white light video image at each sampling time, thereby reconstructing a multi-layer image of the mucosal blood supply distribution in the colon of the target patient at each sampling time; The processing module is also used to calibrate the motion artifacts of the multi-layer images at each sampling time based on the photometric consistency constraints between the multi-layer images at each sampling time, the three-dimensional motion estimation results of adjacent frames, and the cross-modal feature consistency loss output by the pre-trained feature constraint network. The processing module is also used to extract inflammatory-specific abnormally high-blood-supply areas at different depths of the colonic mucosa of the target patient from all calibrated multilayer images, and to determine the infiltration depth of colitis inflammation in the target patient based on the spatial distribution and blood supply intensity characteristics of all abnormally high-blood-supply areas, combined with the pathological infiltration hierarchy mapping model of ulcerative colitis. The execution module is used to combine the infiltration depth, the proportion of mucosal area with abnormally high blood supply, and the degree of abnormal blood supply, and to generate a quantitative score of the degree of colitis inflammation in the target patient based on a quantitative model constructed using a clinically recognized endoscopic scoring system for ulcerative colitis.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing computer-executable code, and the processor being configured to acquire the computer-executable code and execute the method for quantifying the degree of inflammation in ulcerative colitis as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for quantifying the degree of inflammation in ulcerative colitis as described in any one of claims 1 to 7.

Images