Intestinal tumor risk assessment method based on image analysis

Abstract

The invention relates to the technical field of tumor risk assessment, in particular to an intestinal tumor risk assessment method based on image analysis, and the method comprises the steps: carrying out the continuous image division and parameter collection; identifying abnormal density image blocks; identifying concerned risk image blocks; judging a response risk image block; adjusting a preset synchronization degree risk threshold value; early warning judgment is carried out, and a serious early warning report is generated. According to the method, a self-adaptive and dynamically-adjusted three-level judgment system is established through real-time collaborative analysis of multi-dimensional parameters of an enteroscope image block, firstly, a suspected focus is locked based on the texture change duration, then, the focus characteristics are verified through vascular abnormality and contrast fluctuation depth, and finally, the accuracy of the focus is improved. And finally, the risk threshold is corrected in combination with intestinal wall peristalsis and the lens shielding degree, so that the problem of low early flat tumor recognition accuracy caused by insufficient flat tumor image feature capture due to excessive dependence on intestinal environment microorganism and metabolite biochemical detection is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of tumor risk assessment, and in particular to an intestinal tumor risk assessment method based on image analysis. Background Art

[0002] With the popularization of colonoscopy screening and the improvement of endoscopic resolution, more and more studies have found that flat early intestinal lesions are difficult to detect under normal vision. Due to the characteristics of unclear surface protrusions, color close to the surrounding mucosa, and blurred boundaries, it is difficult to visually judge their specific location and range with the naked eye during routine examinations. Especially under the interference of complex factors such as intestinal peristalsis, light changes, and liquid obstruction, their existence characteristics are more likely to be concealed, resulting in clinical identification difficulty significantly higher than typical protruding lesions, becoming a key challenge in the early screening of intestinal tumors with a high missed detection rate and high diagnostic dependence.

[0003] Patent document with publication number CN118197643A discloses an intestinal tumor detection system based on intestinal environment data, which includes: a processing and analysis unit, a judgment and evaluation unit, a tumor detection unit, and a personalized program unit; the processing and analysis unit is used to obtain data related to the intestinal environment and patient clinical data, as well as other historical data, and process the acquired data, and then analyze the intestinal microbial combination and metabolites based on the processed intestinal environment data; the judgment and evaluation unit is used to receive the intestinal microbial combination data, analyzed metabolite data and processed patient clinical data analyzed in the processing and analysis unit, and integrate the analyzed intestinal microbial combination data and analyzed metabolite data, and compare the integrated intestinal data with the positive data in the reference library. The intestinal environment data is used to make a judgment, and it is judged that there are abnormalities in the integrated intestinal data, and then the probability of intestinal tumors is evaluated based on the judged intestinal environment abnormalities and the processed patient clinical data; the tumor detection unit is used to receive the intestinal environment abnormalities judged in the judgment and evaluation unit and the evaluated intestinal tumor probability data to detect intestinal tumors; the processing and analysis unit receives the intestinal tumor data detected in the tumor detection unit and other historical data processed in the processing and analysis unit and performs correlation analysis on diabetes factors; the personalized plan unit is used to receive the factor data of the correlation analysis in the processing and analysis unit and the processed patient clinical data, the intestinal environment abnormalities judged in the judgment and evaluation unit, and the intestinal tumor data detected in the tumor detection unit to formulate a personalized plan.

[0004] It can be seen that the intestinal tumor detection system based on intestinal environmental data has the following problems: the system relies too much on intestinal microorganisms and metabolite detection, the process is lengthy and the detection cycle is long; the correlation between images and clinical data is weak, and it is difficult to provide intuitive and visual high-risk area labeling; the personalized program unit focuses on macro-plan formulation and lacks refined risk stratification guidance based on imaging features. Summary of the Invention

[0005] To this end, the present invention provides an intestinal tumor risk assessment method based on image analysis, which is used to overcome the problem of low accuracy in identifying early flat tumors due to excessive reliance on intestinal environmental microorganisms and biochemical detection of metabolites in the existing technology, resulting in insufficient capture of flat tumor image features.

[0006] To achieve the above objectives, the present invention provides a method for intestinal tumor risk assessment based on image analysis, comprising: Each frame of the colonoscopy image acquired continuously is divided into a number of image blocks to be processed, and the texture response value, blood vessel density, local contrast, intestinal wall contraction frequency and liquid adhesion area on the detection lens surface corresponding to the image frame of each image block to be processed are obtained in real time; determining a plurality of abnormal density image blocks according to the blood vessel density and a preset density threshold; Determining a plurality of focus risk image blocks according to the local contrast within each of the abnormal density image blocks, the intestinal wall contraction frequency, and a preset synchronization risk threshold; Adjusting the preset density threshold according to the number and position of all the risk image blocks at each moment in a preset first adjustment period; determining a plurality of response risk image blocks according to the texture response value; Based on the adjusted preset density threshold, adjusting the preset synchronization risk threshold according to all the response risk image blocks, all the focus risk image blocks, and the liquid attachment area within a preset second adjustment period; An early warning judgment is performed based on the texture response values, the blood vessel density and the local contrast of all the risk image blocks of interest in each image frame that are regained after adjusting the preset synchronization risk threshold within a preset evaluation period, and a serious early warning report is generated based on the judgment result of the early warning judgment.

[0007] Furthermore, the process of determining a plurality of focus risk image blocks according to the local contrast in each abnormal density image block, the intestinal wall contraction frequency, and a preset synchronization risk threshold includes: Determine a plurality of local contrast fluctuation values ​​according to all the local contrasts from an initial moment to each moment within a preset synchronization period; Determining a plurality of contraction frequency fluctuation values ​​according to all the intestinal wall contraction frequencies from an initial moment to each moment within the synchronization period; A plurality of risk-of-interest image blocks are determined according to all of the local contrast fluctuation values, all of the contraction frequency fluctuation values, and the preset synchronization risk threshold.

[0008] Further, the process of determining the concerned risk image blocks according to the preset synchronization risk threshold value, the local contrast fluctuation values, and the contraction frequency fluctuation values comprises: determining fluctuation synchronization degrees according to the local contrast fluctuation values and the contraction frequency fluctuation values; determining the concerned risk image blocks according to a comparison result of the fluctuation synchronization degrees and the preset synchronization risk threshold value.

[0009] Further, the process of adjusting the preset density threshold value according to the number and position of the concerned risk image blocks at each time within a preset first adjustment period comprises: determining a risk distribution degree according to the number and position of the concerned risk image blocks within the preset first adjustment period; adjusting the preset density threshold value based on a comparison result of the risk distribution degree and a preset standard distribution degree range.

[0010] Further, the process of adjusting the preset synchronization risk threshold value according to the response risk image blocks, the concerned risk image blocks, and the liquid attachment area within a preset second adjustment period comprises: determining block overlap rate fluctuation values according to the response risk image blocks and the concerned risk image blocks of each image frame within the preset second adjustment period; adjusting the preset synchronization risk threshold value according to the liquid attachment area within the same preset second adjustment period based on a comparison result of the block overlap rate fluctuation values and a preset block overlap rate fluctuation threshold value.

[0011] Further, the process of adjusting the preset synchronization risk threshold value according to the liquid attachment area within the same preset second adjustment period comprises: adjusting the preset synchronization risk threshold value based on a comparison result of an average value of the liquid attachment area and a preset standard attachment area.

[0012] Further, the process of determining the abnormal density image blocks according to the blood vessel density and the preset density threshold value comprises: determining the abnormal density image blocks based on a comparison result of the blood vessel density and the preset density threshold value.

[0013] Further, the process of determining the response risk image blocks according to the texture response value comprises: determining a duration according to a comparison result of the texture response value and a preset texture response threshold value; determining the response risk image blocks based on a comparison result of the duration and a preset duration threshold value.

[0014] Furthermore, the process of performing early warning determination based on the texture response values, the vascular density, and the local contrast of all the risk image blocks of interest in each image frame, which are re-obtained after adjusting the preset synchronization risk threshold within a preset evaluation period, includes: determining a risk index according to the texture response value, the blood vessel density, and the local contrast; The determination result is determined based on a comparison result between the risk index and a preset risk index threshold.

[0015] Furthermore, the process of dividing each frame of the colonoscopy image obtained in continuous frames into a number of image blocks to be processed includes: The colonoscopy image is divided into two-dimensional grids with a preset grid side length to form a plurality of non-overlapping image blocks to be processed.

[0016] Compared with the prior art, the present invention has the beneficial effect of establishing an adaptive, dynamically adjusted three-level judgment system through real-time collaborative analysis of multi-dimensional parameters such as texture response, vascular density, local contrast, intestinal wall contraction frequency, and lens liquid adhesion area of ​​colonoscopic image blocks. First, suspected lesions are identified based on the duration of texture changes. Then, lesion characteristics are verified by vascular abnormalities and the depth of contrast fluctuations. Finally, the risk threshold is corrected based on intestinal wall peristalsis and the degree of lens occlusion, achieving highly sensitive capture of flat early-stage tumors. At the same time, the step-by-step update of the threshold not only follows the spatial statistical laws of image block distribution, but also takes into account the causal relationship between physiological movement and mirror phenomena. This can reduce missed diagnoses caused by strong peristalsis or liquid occlusion, and avoid false alarms caused by occasional noise. Thus, while maintaining high sensitivity, it achieves high specificity and clinical usability, significantly improving the accuracy and stability of early-stage tumor risk assessment, and effectively solving the problem of insufficient capture of flat tumor image features due to over-reliance on biochemical detection of intestinal environmental microorganisms and metabolites, which in turn leads to low accuracy in identifying early-stage flat tumors.

[0017] Furthermore, through standard deviation analysis of the two dimensions of "local contrast fluctuation" and "intestinal wall contraction frequency fluctuation," the synchronous abnormalities of structure and motion in the image blocks are captured. Normal mucosal areas exhibit relatively stable, small fluctuations in color contrast and peristaltic rhythm, while early flat tumors, due to tissue hyperplasia and local stiffness, will produce larger and synchronous discrete changes in both reflected light contrast and wall contraction amplitude. By combining the fluctuation values ​​of the two with the preset synchronous risk threshold, the co-occurrence characteristics of "texture abnormalities" and "physiological dynamic abnormalities" in the lesion are reflected. This is consistent with the optical differences caused by vascular and structural remodeling caused by tumor lesions, and also conforms to the physiological law of increased peristaltic inertia in the lesion area, thereby achieving accurate identification of "hidden" high-risk segments.

[0018] Furthermore, by performing maximum-minimum normalization on local contrast fluctuations and intestinal wall contraction frequency fluctuations and calculating their Pearson correlation coefficient, we can quantitatively characterize the synchronization anomalies between the image's optical characteristics and physiological motion characteristics. That is, normal mucosal areas exhibit independent tiny fluctuations in color contrast and peristaltic rhythm, while early flat tumors, due to tissue hyperplasia, lead to increased local stiffness, which inevitably causes coupled changes in optical reflection and contraction amplitude. When this coupling degree exceeds the preset threshold, the risk image blocks can be accurately identified, significantly improving the sensitivity and robustness of detecting hidden lesions.

[0019] Furthermore, by counting the number of risky image blocks at each moment and calculating their ratio to the minimum bounding rectangle area, the spatial concentration or dispersion of high-risk areas can be mapped in real time. When the risk distribution exceeds the preset maximum value, increasing the vascular density threshold can fine-tune the judgment range and avoid over-labeling of noise blocks. Conversely, when the distribution is below the minimum value, lowering the threshold can expand the sensitive area to prevent missing small but scattered lesions. The introduction of a threshold adjustment coefficient makes the threshold update amplitude proportional to the deviation, ensuring that the response to changes in the risk distribution is neither hesitant nor excessive, thereby maintaining efficient and stable abnormal block recognition under different distribution patterns.

[0020] Furthermore, by calculating the overlap rate and fluctuation value of the response risk image block and the focus risk image block, the consistency of the initial screening and re-screening results can be dynamically monitored; when the overlap rate fluctuation exceeds the preset threshold, the threshold is adjusted in combination with the lens liquid adhesion area, which can effectively compensate for the positioning deviation caused by occlusion or movement; the judgment conditions can be automatically relaxed in the "unstable-occlusion" situation, and the judgment range can be tightened in the "stable-clear" situation, thereby taking into account the sensitive capture of high-risk areas and the robust suppression of noise blocks, significantly improving the accuracy and reliability of the overall assessment.

[0021] Furthermore, by averaging the liquid adhesion area of ​​the lens during the second adjustment cycle and comparing it with the preset standard adhesion area, the degree of imaging occlusion can be perceived in real time. When the average occlusion exceeds the standard, the synchronous risk threshold is automatically lowered according to the deviation ratio to maintain higher sensitivity to the area obscured by the liquid film, avoiding missed detections due to occlusion. When the occlusion is not sufficient to affect recognition, over-compensation will not be triggered, thereby adaptively balancing the threshold under different occlusion states to ensure that the risk judgment is neither too loose nor too strict, greatly improving the stability and accuracy of the overall recognition.

[0022] Furthermore, by setting a vascular density threshold, the local high-density features caused by "new microvessels" in the lesion area can be quickly captured. That is, when the proportion of vascular pixels in a certain image block exceeds the threshold, it is judged as an abnormal density image block. It can not only accurately reflect the vascular reconstruction phenomenon accompanied by tumor cell proliferation, but also efficiently eliminate the weak vascular structure of normal tissue in massive frames, greatly improving the speed and reliability of abnormal area screening, while providing accurate candidates for subsequent multi-dimensional risk assessment.

[0023] Furthermore, by recording the timestamps of the first time the texture response value exceeds the threshold and returns to below the threshold, and calculating its duration, it is possible to effectively filter out occasional noise and short-term light fluctuations, and only focus on those areas that show significant texture intensity in both space and time. The preset duration threshold and the texture response threshold work together to instantly capture tiny but continuous abnormal fluctuations on the surface, and avoid false alarms caused by momentary jitter or local reflections. This significantly improves the stability and accuracy of preliminary risk block identification while maintaining high sensitivity.

[0024] Furthermore, by normalizing the texture response, vascular density, and local contrast mean of the risky image blocks of interest within each frame during the assessment cycle and integrating them according to preset weights, the overall abnormality level of optical, vascular, and contrast features can be comprehensively measured. The dispersion of the full-cycle temporary index is then calculated, reflecting the stability and consistency of high-risk segments in the form of a risk index. When the risk index exceeds the set threshold, a critical warning report is precisely triggered, thereby achieving high-precision identification and visual annotation of high-risk areas through multi-dimensional feature fusion.

[0025] Furthermore, by dividing the colonoscopy image into a grid according to the preset grid side length, the panoramic field of view can be decomposed into small areas that match the typical size of the lesion, which can not only ensure that each image block contains sufficient pixel details to extract tiny textures and vascular features, but also avoid the computational redundancy caused by whole-frame processing; in addition, block-level division naturally retains the spatial proximity relationship, which facilitates the dynamic adjustment of thresholds based on local statistical characteristics (such as intra-block contrast and vascular density), and parallel processing also greatly improves the real-time feedback speed, thereby achieving a good balance between fine-grained detection and efficient computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of the intestinal tumor risk assessment method based on image analysis in this embodiment; Figure 2 This is a decision logic diagram for determining the risky image blocks in this embodiment; Figure 3 A decision logic diagram for adjusting the preset synchronization risk threshold in this embodiment; Figure 4This is a logic diagram for determining abnormal density image blocks in this embodiment. DETAILED DESCRIPTION

[0027] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0028] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] See also Figure 1 As shown, it is a flow chart of the intestinal tumor risk assessment method based on image analysis in this embodiment. This embodiment provides an intestinal tumor risk assessment method based on image analysis, including: Each frame of the colonoscopy image acquired continuously is divided into a number of image blocks to be processed, and the texture response value, blood vessel density, local contrast, intestinal wall contraction frequency and liquid adhesion area on the detection lens surface corresponding to the image frame of each image block to be processed are obtained in real time; determining a plurality of abnormal density image blocks according to the blood vessel density and a preset density threshold; Determining a plurality of focus risk image blocks according to the local contrast within each of the abnormal density image blocks, the intestinal wall contraction frequency, and a preset synchronization risk threshold; Adjusting the preset density threshold according to the number and position of all the risk image blocks at each moment in a preset first adjustment period; determining a plurality of response risk image blocks according to the texture response value; Based on the adjusted preset density threshold, adjusting the preset synchronization risk threshold according to all the response risk image blocks, all the focus risk image blocks, and the liquid attachment area within a preset second adjustment period; An early warning judgment is performed based on the texture response values, the blood vessel density and the local contrast of all the risk image blocks of interest in each image frame that are regained after adjusting the preset synchronization risk threshold within a preset evaluation period, and a serious early warning report is generated based on the judgment result of the early warning judgment.

[0030] The preset density threshold refers to the threshold used to determine whether the vascular density of an image block is abnormal. It depends on the statistical characteristics of the vascular density distribution of healthy tissue and diseased tissue in the training sample. It is usually set between 0.3 and 0.6, and is 0.4 in this embodiment. It can effectively distinguish normal microvessels from lesion hyperplasia areas.

[0031] The preset synchronization risk threshold refers to the threshold used to determine whether the synchronization degree of fluctuations such as local contrast and intestinal wall contraction frequency or the fluctuation of block overlap rate is significant, which depends on the distribution range of the synchronization degree or overlap rate; it is usually set between 0.5 and 0.8, and 0.7 is taken in this embodiment; it can sensitively capture parameter synchronization anomalies and identify potential high-risk areas.

[0032] The preset evaluation period refers to the number of frames or the length of time required to accumulate and evaluate the risk index within a number of consecutive frames, which depends on the endoscopic video frame rate and the frequency of intestinal peristalsis; it is usually set between 1 second and 4 seconds, and is set to 2 seconds in this embodiment; it can balance noise filtering and real-time response to ensure that risk assessment is both stable and timely.

[0033] The acquisition of various parameters is based on image processing and sensor fusion. First, each frame of the colonoscopy image is segmented according to a preset grid, and the texture response value is calculated for each image block (for example, a comprehensive texture index is generated by combining the weighted local binary pattern (LTP), Laplace second derivative, and GLCM contrast features). The vascular density is obtained by applying a multi-scale Frangi filter to the NBI or fluorescence-enhanced image within the block and then binarizing it, and then counting the proportion of vascular pixels. The local contrast is quantified by the difference or standard deviation between the maximum and minimum grayscale within the block. The intestinal wall contraction frequency is estimated by the optical flow method of continuous frames to estimate the block edge motion amplitude, and the number of periodic peaks per unit time is counted. The liquid adhesion area on the lens surface is extracted by using HSV spatial saturation threshold segmentation and fusion with highlight reflection area detection to extract the proportion of the total number of pixels in the panoramic high saturation or highlight area, that is, the liquid adhesion area.

[0034] The severe warning report is a summary document automatically generated for high-risk (severe) tumor areas identified by the system, based on their most abnormal texture response, vascular density, and contrast values. It is used to highlight the "severe" risk segments and provide corresponding frame numbers and heat map overlays.

[0035] Through real-time collaborative analysis of multi-dimensional parameters of colonoscopic image blocks, including texture response, vascular density, local contrast, intestinal wall contraction frequency, and lens fluid adhesion area, an adaptive and dynamically adjusted three-level judgment system was established. First, suspected lesions are identified based on the duration of texture changes. Lesion characteristics are then verified using vascular abnormalities and the depth of contrast fluctuations. Finally, the risk threshold is modified based on intestinal wall peristalsis and the degree of lens occlusion to achieve highly sensitive capture of flat early-stage tumors. At the same time, the step-by-step update of the threshold follows the spatial statistical laws of image block distribution while taking into account the causal relationship between physiological movement and mirror phenomena. This can reduce missed diagnoses caused by strong peristalsis or fluid occlusion, and avoid false positives caused by occasional noise. This system achieves high specificity and clinical usability while maintaining high sensitivity, significantly improving the accuracy and stability of early-stage tumor risk assessment. This effectively addresses the problem of insufficient capture of flat tumor image features due to over-reliance on biochemical detection of intestinal environmental microorganisms and metabolites, which in turn leads to low accuracy in early-stage flat tumor identification.

[0036] Specifically, the process of determining a plurality of focus risk image blocks according to the local contrast in each abnormal density image block, the intestinal wall contraction frequency, and a preset synchronization risk threshold comprises: Calculate the standard deviation of all local contrasts from the initial moment to each moment within the preset synchronization period to obtain a number of local contrast fluctuation values; Calculate the standard deviation of all intestinal wall contraction frequencies from the initial moment to each moment within the preset synchronization period to obtain a number of contraction frequency fluctuation values; A plurality of risk-of-interest image blocks are determined according to all of the local contrast fluctuation values, all of the contraction frequency fluctuation values, and the preset synchronization risk threshold.

[0037] The preset synchronization period refers to the number of frames or time window for continuously collecting and comparing the local contrast and intestinal wall contraction frequency fluctuations, which depends on the endoscopic video frame rate and intestinal peristalsis rhythm; it is usually set between 1 second and 3 seconds, and in this embodiment is set to 2 seconds, which can take into account the timeliness of motion noise filtering and synchronization anomaly capture.

[0038] By analyzing the standard deviation of the two dimensions of "local contrast fluctuation" and "intestinal wall contraction frequency fluctuation," the system captures synchronous abnormalities in structure and motion within image blocks. Normal mucosal areas exhibit relatively stable, small fluctuations in color contrast and peristaltic rhythm, while early-stage flat tumors, due to tissue hyperplasia and localized stiffness, produce larger, synchronized, discrete changes in both reflected light contrast and wall contraction amplitude. By combining the fluctuation values ​​of these two factors with a preset synchronous risk threshold, the system reflects the co-occurrence of "texture abnormalities" and "physiological dynamic abnormalities" at the lesion site. This is consistent with both the optical differences caused by vascular and structural remodeling due to tumor lesions and the physiological law of increased peristaltic inertia in the lesion area, thereby enabling accurate identification of "hidden" high-risk segments.

[0039] See also Figure 2 As shown in FIG, which is a determination logic diagram for determining a risk image block of interest in this embodiment, in this embodiment, the process of determining a number of risk image blocks of interest based on all the local contrast fluctuation values, all the contraction frequency fluctuation values, and the preset synchronization risk threshold includes: Perform maximum and minimum normalization on all local comparison fluctuation values ​​to obtain several local comparison standard values; Perform maximum and minimum normalization on all contraction frequency fluctuation values ​​to obtain several contraction frequency standard values; The Pearson correlation coefficient of all local contrast standard values ​​and all contraction frequency standard values ​​was calculated to obtain the degree of fluctuation synchronization; When the fluctuation synchronization degree is greater than a preset synchronization risk threshold, the abnormal density image block is determined to be a focus risk image block, so as to determine a number of focus risk image blocks.

[0040] By performing maximum-minimum normalization on local contrast fluctuations and intestinal wall contraction frequency fluctuations and calculating the Pearson correlation coefficient between the two, it is possible to quantitatively characterize the synchronization abnormalities between the image's optical characteristics and physiological motion characteristics. That is, normal mucosal areas exhibit independent tiny fluctuations in color contrast and peristaltic rhythm, while early flat tumors, due to tissue hyperplasia, lead to increased local stiffness, which inevitably causes coupled changes in optical reflection and contraction amplitude. When this coupling degree exceeds the preset threshold, the risk image blocks of concern can be accurately identified, significantly improving the sensitivity and robustness of detection of hidden lesions.

[0041] Specifically, the process of adjusting the preset density threshold according to the number and position of all the risk image blocks at each moment in the preset first adjustment period includes: Count the number and location of risky image blocks at each moment; Calculate the minimum bounding rectangle area of ​​all risk image blocks; Calculate the ratio of the number of risky image blocks to the area of ​​the minimum circumscribed rectangle to obtain the risk distribution degree; When the risk distribution degree is greater than the maximum value of the preset standard distribution degree range, the preset density threshold is increased according to the relative deviation between the risk distribution degree and the maximum value of the preset standard distribution degree range and the preset first adjustment coefficient, Q'=Q×[1+k1×(R-Rmax) / Rmax], where Q' is the preset density threshold after the increase, Q is the preset density threshold before the increase, k1 is the preset first adjustment coefficient, R is the risk distribution degree, and Rmax is the maximum value of the preset standard distribution degree range; When the risk distribution degree is less than the minimum value of the preset standard distribution degree range, the preset density threshold is reduced according to the relative deviation between the minimum value of the preset standard distribution degree range and the risk distribution degree and the preset first adjustment coefficient, Q'=Q×[1-k1×(Rmin-R) / R], where Q' is the preset density threshold after reduction, Q is the preset density threshold before reduction, and Rmin is the minimum value of the preset standard distribution degree range.

[0042] The preset standard distribution range refers to the reasonable upper and lower limits of the risk distribution, which is used to judge the degree of spatial aggregation or dispersion of high-risk areas, depending on the size and distribution characteristics of the lesions; it is usually set between 0.5 and 2.0. In this embodiment, [0.8, 1.5] is taken to balance the sensitivity to dense and dispersed risk blocks.

[0043] The first adjustment coefficient refers to the proportional factor used to control the deviation between the update amplitude of the vascular density threshold and the risk distribution degree, which depends on the system response speed and noise tolerance. It is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.2, which enables the threshold adjustment to quickly follow the distribution changes while avoiding excessive fluctuations.

[0044] By counting the number of risky image blocks at each moment and calculating their ratio to the minimum bounding rectangle area, the spatial concentration or dispersion of high-risk areas can be mapped in real time. When the risk distribution exceeds the preset maximum value, increasing the vascular density threshold can fine-tune the judgment range and avoid over-labeling of noise blocks. Conversely, when the distribution is below the minimum value, lowering the threshold can expand the sensitive area to prevent missing small but scattered lesions. The introduction of a threshold adjustment coefficient makes the threshold update amplitude proportional to the deviation, ensuring that the response to changes in the risk distribution is neither hesitant nor excessive, thereby maintaining efficient and stable abnormal block recognition under different distribution patterns.

[0045] Specifically, the process of adjusting the preset synchronization risk threshold according to all the response risk image blocks, all the focus risk image blocks, and the liquid attachment area within the preset second adjustment period includes: Calculate the image block overlap ratio of all response risk image blocks and all focus risk image blocks in each image frame within the preset second adjustment period to obtain several image block overlap ratios, where Y=|F1∪F2| / |F1∩F2|, Y is the image block overlap ratio, F1 is the set formed by all response risk image blocks in each image frame, and F2 is the set formed by all focus risk image blocks in each image frame; Calculate the standard deviation of the overlap rate of all image blocks to obtain the block overlap rate fluctuation value; When the overlap rate fluctuation value is greater than the preset block overlap rate fluctuation threshold, the preset synchronization risk threshold is adjusted according to all liquid attachment areas within the same preset second adjustment period.

[0046] The preset second adjustment period refers to the number of consecutive frames or time windows used to accumulate the overlap of the initial screening and re-screening risk image blocks and perform threshold correction, which depends on the endoscope frame rate and intestinal movement frequency; it is usually set between 1 second and 3 seconds, and is set to 2 seconds in this embodiment; it can achieve sensitive and timely threshold adjustment while ensuring sufficient data volume.

[0047] The preset block overlap rate fluctuation threshold refers to the critical value of the fluctuation amplitude used to determine the stability of the overlap rate of the risk image blocks in the initial screening and re-screening, which depends on the image noise level and the degree of motion jitter; it is usually set between 0.05 and 0.2, and is 0.1 in this embodiment, which can distinguish between occasional jitter and real positioning instability, thereby accurately triggering threshold correction.

[0048] By calculating the overlap rate and fluctuation value of the response risk image block and the focus risk image block, the consistency of the initial screening and re-screening results can be dynamically monitored; when the overlap rate fluctuation exceeds the preset threshold, the threshold is adjusted in combination with the lens liquid adhesion area, which can effectively compensate for the positioning deviation caused by occlusion or movement; it can automatically relax the judgment conditions in the "unstable-occlusion" situation, and tighten the judgment range in the "stable-clear" situation, thereby taking into account the sensitive capture of high-risk areas and the robust suppression of noise blocks, significantly improving the accuracy and reliability of the overall assessment.

[0049] See also Figure 3 As shown in FIG. , which is a determination logic diagram for determining and adjusting the preset synchronization risk threshold in this embodiment, in this embodiment, the process of adjusting the preset synchronization risk threshold according to all the liquid adhesion areas within the same preset second adjustment period includes: Calculating an average value of all liquid attachment areas within the same preset second adjustment period; When the average value of all liquid attachment areas is greater than the preset standard attachment area, the preset synchronization risk threshold is reduced based on the relative deviation between the average value of all liquid attachment areas and the preset standard attachment area and the preset second adjustment coefficient: E'=E×[1-k2×(P-P0) / P0], where E' is the preset synchronization risk threshold after reduction, E is the preset synchronization risk threshold before reduction, k2 is the preset second adjustment coefficient, P is the average value of all liquid attachment areas, and P0 is the preset standard attachment area.

[0050] The preset standard adhesion area refers to a reference value used to judge the degree of liquid adhesion on the lens surface, which depends on the cleanliness of the colonoscope and the mirror reflection characteristics. It is usually set between 5% and 20%, and is set to 10% in this embodiment. It can accurately distinguish between slight residue and significant occlusion, thereby reasonably triggering the threshold to be lowered.

[0051] The preset second adjustment coefficient refers to the proportional factor that controls the liquid adhesion deviation to the reduction of the synchronization risk threshold, which depends on the system's requirements for occlusion compensation speed and stability. It is usually set between 0.1 and 0.5, and is set to 0.3 in this embodiment. It can avoid over-adjustment while improving sensitivity, ensuring that the evaluation is robust and reliable.

[0052] By averaging the liquid adhesion area of ​​the lens during the second adjustment cycle and comparing it with the preset standard adhesion area, the degree of imaging occlusion can be perceived in real time. When the average occlusion exceeds the standard, the synchronous risk threshold is automatically lowered according to the deviation ratio to maintain higher sensitivity to the area obscured by the liquid film, avoiding missed detections due to occlusion. When the occlusion is not sufficient to affect recognition, over-compensation will not be triggered, thereby adaptively balancing the threshold under different occlusion states to ensure that the risk judgment is neither too loose nor too strict, greatly improving the stability and accuracy of the overall recognition.

[0053] See also Figure 4 As shown in FIG. , which is a determination logic diagram for determining abnormal density image blocks in this embodiment, in this embodiment, the process of determining a number of abnormal density image blocks according to the blood vessel density and the preset density threshold includes: When the blood vessel density is greater than a preset density threshold, the image block to be processed is determined to be an abnormal density image block, so as to determine a number of abnormal density image blocks.

[0054] By setting a vascular density threshold, the local high-density features caused by "new microvessels" in the lesion area can be quickly captured. That is, when the proportion of vascular pixels in a certain image block exceeds the threshold, it is judged as an abnormal density image block. It can not only accurately reflect the vascular reconstruction phenomenon accompanied by tumor cell proliferation, but also efficiently eliminate the weak vascular structure of normal tissue in massive frames, greatly improving the speed and reliability of abnormal area screening, while providing accurate candidates for subsequent multi-dimensional risk assessment.

[0055] Specifically, the process of determining a plurality of response risk image blocks according to the texture response value includes: Recording a timestamp when the texture response value is greater than a preset texture response threshold, and recording a timestamp when the texture response value is less than or equal to the preset texture response threshold, to obtain a duration; When the duration is greater than a preset duration threshold, the image block to be processed is determined to be a response risk image block, so as to determine a number of response risk image blocks.

[0056] The preset texture response threshold refers to the critical value used to determine whether the texture response intensity is significant, which depends on the statistical distribution of texture indicators of normal mucosa and lesion areas; it is usually set between 0.2 and 0.4, and is set to 0.3 in this embodiment, which can accurately capture continuous small changes in concave and convexity and exclude instantaneous noise.

[0057] The preset duration threshold refers to the time window used to filter short-term fluctuations and confirm the persistent existence of abnormal textures, which depends on the endoscope frame rate and intestinal movement rhythm; it is usually set between 0.5 seconds and 1.5 seconds, and in this embodiment is set to 1 second, which can balance sensitivity and stability to ensure continuous recognition of the initial screening block.

[0058] By recording the timestamps of the first time the texture response value exceeds the threshold and returns to below the threshold, and calculating its duration, it is possible to effectively filter out occasional noise and short-term light fluctuations, and only focus on those areas that show significant texture intensity in both space and time. The preset duration threshold and texture response threshold work together to instantly capture tiny but continuous abnormal fluctuations on the surface, and avoid false alarms caused by momentary jitter or local reflections. This significantly improves the stability and accuracy of preliminary risk block identification while maintaining high sensitivity.

[0059] Specifically, the process of performing early warning determination based on the texture response values, the vascular density, and the local contrast of all the risk image blocks of interest in each image frame, which are re-obtained after adjusting the preset synchronization risk threshold within a preset evaluation period, includes: Calculating the average value of all texture response values ​​in each image frame of a preset evaluation period to obtain a plurality of texture response means; Calculating the average value of all blood vessel densities in each image frame of a preset evaluation period to obtain a number of blood vessel density means; Calculating the average value of all local contrasts in each image frame of a preset evaluation period and several local contrast means; Normalize the mean values ​​of all texture responses to obtain several texture normalization values; Normalize the mean values ​​of all blood vessel densities to obtain several density normalization values; Normalize all local contrast means to obtain several contrast normalization values; The texture normalization value, the preset texture weight, the density normalization value, the preset density weight, the contrast normalization value and the preset contrast weight in each image frame are weighted and summed to obtain a plurality of temporary indexes, U=a×w+b×m+c×d, wherein U is the temporary index, a is the preset texture weight, w is the texture normalization value, b is the preset density weight, m is the density normalization value, c is the preset contrast weight, and d is the contrast normalization value; The standard deviation of all temporary indexes is calculated to obtain a risk index; When the risk index is greater than a preset risk index threshold, it is determined that a serious risk occurs, and a serious warning report is generated.

[0060] The preset texture weight refers to a proportional coefficient for the influence of the texture response in the risk index calculation, which depends on the importance of the texture feature to tumor detection, and is usually set to be between 0.2 and 0.6, and is set to 0.4 in the embodiment, which can ensure that the texture details have a proper contribution to the comprehensive evaluation.

[0061] The preset density weight refers to a proportional coefficient for the influence of the blood vessel density in the risk index calculation, which depends on the importance of the neovascularization feature; it is usually set to be between 0.2 and 0.6, and is set to 0.4 in the embodiment, which can reasonably amplify the effect of microvascular proliferation on risk determination.

[0062] The preset contrast weight refers to a proportional coefficient for the influence of the local contrast in the risk index calculation, which depends on the role of the contrast feature in identifying flat lesions; it is usually set to be between 0.1 and 0.4, and is set to 0.2 in the embodiment, which can take into account the auxiliary distinguishing ability of texture and blood vessel features in the comprehensive evaluation.

[0063] The preset risk index threshold refers to a risk index determination standard for distinguishing high-risk and non-high-risk, which depends on the balance requirement of sensitivity and specificity in clinical practice; it is usually set to be between 0.5 and 0.8, and is set to 0.7 in the embodiment, which can effectively trigger the generation of a serious warning report and avoid false positives.

[0064] By normalizing the texture response, blood vessel density and local contrast of each frame of the attention risk image block in the evaluation period and integrating them according to the preset weight, the overall abnormal level of optical features, blood vessel features and contrast features can be comprehensively measured. Then the dispersion of the temporary indexes in the whole cycle is calculated to reflect the stability and consistency of the high-risk segments in the form of risk index; when the risk index exceeds the set threshold, the generation of a serious warning report is triggered, so that the high-precision locking and visual marking of the high-risk area are realized under the fusion of multiple features.

[0065] Specifically, the process of dividing each frame in the colonoscopy images obtained by the continuous frames into a plurality of to-be-processed image blocks includes: The colonoscopy image is divided into two-dimensional grids with a preset grid side length to form a plurality of non-overlapping image blocks to be processed.

[0066] The preset grid side length is the width of the image block designed to take into account both lesion size and imaging resolution, and depends on the endoscopic resolution and the typical diameter of a flat tumor. It is usually set between 32 pixels and 128 pixels, and in this embodiment is set to 64 pixels, which can achieve the best balance between fine-grained feature extraction and real-time computing performance.

[0067] By dividing the colonoscopy image into a grid according to the preset grid side length, the panoramic field of view can be decomposed into small areas that match the typical size of the lesion. This can ensure that each image block contains sufficient pixel details to extract tiny textures and vascular features, while avoiding computational redundancy caused by whole-frame processing. In addition, block-level division naturally retains spatial proximity, facilitating dynamic adjustment of thresholds based on local statistical characteristics (such as intra-block contrast and vascular density). Parallel processing also greatly improves the real-time feedback speed, thereby achieving a good balance between fine-grained detection and efficient computing.

[0068] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for intestinal tumor risk assessment based on image analysis, characterized in that: include: Each frame of the colonoscopy image acquired continuously is divided into a number of image blocks to be processed, and the texture response value, blood vessel density, local contrast, intestinal wall contraction frequency and liquid adhesion area on the detection lens surface corresponding to the image frame of each image block to be processed are obtained in real time; determining a plurality of abnormal density image blocks according to the blood vessel density and a preset density threshold; Determining a plurality of focus risk image blocks according to the local contrast within each of the abnormal density image blocks, the intestinal wall contraction frequency, and a preset synchronization risk threshold; Adjusting the preset density threshold according to the number and position of all the risk image blocks at each moment in a preset first adjustment period; determining a plurality of response risk image blocks according to the texture response value; Based on the adjusted preset density threshold, adjusting the preset synchronization risk threshold according to all the response risk image blocks, all the focus risk image blocks, and the liquid attachment area within a preset second adjustment period; An early warning judgment is performed based on the texture response values, the blood vessel density and the local contrast of all the risk image blocks of interest in each image frame that are regained after adjusting the preset synchronization risk threshold within a preset evaluation period, and a serious early warning report is generated based on the judgment result of the early warning judgment.

2. The intestinal tumor risk assessment method based on image analysis according to claim 1, characterized in that: The process of determining a plurality of focus risk image blocks according to the local contrast in each abnormal density image block, the intestinal wall contraction frequency, and a preset synchronization risk threshold comprises: Determine a plurality of local contrast fluctuation values ​​according to all the local contrasts from an initial moment to each moment within a preset synchronization period; Determining a plurality of contraction frequency fluctuation values ​​according to all the intestinal wall contraction frequencies from an initial moment to each moment within the synchronization period; A plurality of risk-of-interest image blocks are determined according to all of the local contrast fluctuation values, all of the contraction frequency fluctuation values, and the preset synchronization risk threshold.

3. The intestinal tumor risk assessment method based on image analysis according to claim 2, characterized in that: The process of determining a plurality of focus risk image blocks according to all the local contrast fluctuation values, all the contraction frequency fluctuation values ​​and the preset synchronization risk threshold comprises: determining a degree of fluctuation synchronization based on all of the local contrast fluctuation values ​​and all of the contraction frequency fluctuation values; A plurality of risk-of-concern image blocks are determined based on a comparison result of the fluctuation synchronization degree and the preset synchronization risk threshold.

4. The intestinal tumor risk assessment method based on image analysis according to claim 3, characterized in that: The process of adjusting the preset density threshold according to the number and position of all the risk image blocks at each moment in the preset first adjustment period includes: Determining the risk distribution degree according to the number and position of all the risk-related image blocks within the preset first adjustment period; The preset density threshold is adjusted based on a comparison result between the risk distribution degree and a preset standard distribution degree range.

5. The intestinal tumor risk assessment method based on image analysis according to claim 4, characterized in that: The process of adjusting the preset synchronization risk threshold according to all the response risk image blocks, all the focus risk image blocks, and the liquid attachment area within the preset second adjustment period includes: determining a block overlap rate fluctuation value according to all the response risk image blocks and all the focus risk image blocks of each image frame within the preset second adjustment period; Based on a comparison result between the block overlap rate fluctuation value and a preset block overlap rate fluctuation threshold, the preset synchronization risk threshold is adjusted according to all the liquid attachment areas within the same preset second adjustment period.

6. The method for intestinal tumor risk assessment based on image analysis according to claim 5, characterized in that: The process of adjusting the preset synchronization risk threshold according to all the liquid attachment areas within the same preset second adjustment period includes: The preset synchronization risk threshold is adjusted based on a comparison result of an average value of all the liquid attachment areas and a preset standard attachment area.

7. The method for intestinal tumor risk assessment based on image analysis according to claim 6, characterized in that: The process of determining a plurality of abnormal density image blocks according to the blood vessel density and a preset density threshold comprises: A plurality of abnormal density image blocks are determined based on a comparison result between the blood vessel density and a preset density threshold.

8. The method for intestinal tumor risk assessment based on image analysis according to claim 7, characterized in that: The process of determining a plurality of response risk image blocks according to the texture response value includes: Determining a duration based on a comparison result of the texture response value and a preset texture response threshold; A plurality of response risk image blocks are determined based on a comparison result of the duration and a preset duration threshold.

9. The intestinal tumor risk assessment method based on image analysis according to claim 1, characterized in that: The process of performing early warning determination based on the texture response values, the blood vessel density, and the local contrast of all the risk image blocks of interest in each image frame, which are re-obtained after adjusting the preset synchronous risk threshold within a preset evaluation period, includes: determining a risk index according to the texture response value, the blood vessel density, and the local contrast; The determination result is determined based on a comparison result between the risk index and a preset risk index threshold.

10. The intestinal tumor risk assessment method based on image analysis according to claim 1, characterized in that: The process of dividing each frame of the colonoscopy image acquired in continuous frames into a number of image blocks to be processed includes: The colonoscopy image is divided into two-dimensional grids with a preset grid side length to form a plurality of non-overlapping image blocks to be processed.

Images