Digestive system department focus identification method and system based on object identification

Abstract

The invention discloses a digestive system department focus recognition method and system based on object recognition, and relates to the technical field of digestive system departments, and the technical scheme is characterized in that the method comprises the following steps: setting a cross-modal monitoring unit associated with different modal to-be-fused image sub-regions according to the modal type and imaging parameters of a digestive system department multi-modal image; the abnormal structure monitoring unit is used for extracting focus structure feature data of the to-be-fused image sub-regions according to the cross-modal monitoring unit, and analyzing the focus structure feature data to obtain a structure abnormal coefficient and the cross-modal monitoring unit; the abnormal blood flow monitoring unit is used for extracting focus blood flow characteristic data of the to-be-fused image sub-region according to the cross-modal monitoring unit, and analyzing the focus blood flow characteristic data to obtain a blood flow abnormal coefficient of the to-be-fused image sub-region and an abnormal blood flow monitoring unit of the cross-modal monitoring unit; the method is convenient for obtaining the recognition result of the key focus.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gastroenterology, more specifically, it relates to a gastroenterology lesion identification method and system based on object recognition. BACKGROUND

[0002] Traditional lesion identification methods in the field of gastroenterology lesion identification have many shortcomings. The limitations of single modality images are obvious. For example, relying only on endoscopic images can directly observe the surface morphology of the lesion, but it is difficult to understand the deep structure and blood flow of the lesion. Using only CT images can show the anatomical structure of the lesion, but lacks fine texture information on the surface of the lesion. The limitations of such single modality lead doctors to often combine multiple image data for manual comparison and analysis, which not only consumes time and effort, but also is prone to misdiagnosis or missed diagnosis due to differences in human judgment.

[0003] Traditional methods analyze lesion structure and blood flow characteristics in a rough way, lacking quantitative indicators. For example, there is no systematic threshold comparison and slope analysis to quantify volume abnormalities for changes in lesion volume. For lesion surface roughness, there is no analysis of texture gray deviation and uniformity to accurately assess surface abnormalities. For blood flow characteristics, there is a lack of comprehensive quantitative analysis of blood flow perfusion, velocity, and fluctuation amplitude. This rough analysis method makes it difficult to accurately assess the degree of lesion abnormalities, thereby affecting the accuracy of diagnosis to some extent. SUMMARY

[0004] In view of the deficiencies in the prior art, the purpose of the present application is to provide a gastroenterology lesion identification method and system based on object recognition.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions: A gastroenterology lesion identification method based on object recognition, the method comprising the following steps: According to the modality type and imaging parameter setting of the multi-modality image of gastroenterology, the cross-modality monitoring unit of the different modality sub-regions of the to-be-fused image is associated; According to the cross-modality monitoring unit, the lesion structure feature data of the to-be-fused image sub-region is extracted, and the structure abnormality coefficient and the abnormal structure monitoring unit of the cross-modality monitoring unit are obtained by analyzing the lesion structure feature data; According to the cross-modality monitoring unit, the lesion blood flow feature data of the to-be-fused image sub-region is extracted, and the blood flow abnormality coefficient of the to-be-fused image sub-region and the abnormal blood flow monitoring unit of the cross-modality monitoring unit are obtained by analyzing the lesion blood flow feature data; The abnormal structure monitoring unit and the abnormal blood flow monitoring unit form a lesion suspected monitoring unit; and the lesion risk degree of the lesion suspected monitoring unit is obtained according to the structure abnormality coefficient and the blood flow abnormality coefficient; The lesion risk degree, the lesion suspected monitoring unit and the modality fusion terminal are analyzed to identify the digestive internal medicine lesion.

[0006] Preferably, the lesion risk degree, the lesion suspected monitoring unit and the modality fusion terminal are analyzed to identify the digestive internal medicine lesion, and the method comprises the following steps: The fusion signal strength of the lesion suspected monitoring unit is generated according to the lesion risk degree, the lesion fusion signal of the cross-modality monitoring unit is generated according to the fusion signal strength, the signal fusion time of the lesion suspected monitoring unit is obtained according to the lesion suspected monitoring unit and the cross-modality fusion terminal; The lesion identification priority of the lesion suspected monitoring unit is obtained according to the fusion signal strength and the signal fusion time, and the lesion suspected monitoring unit is identified according to the lesion identification priority and the lesion fusion signal by the cross-modality fusion terminal.

[0007] Preferably, the imaging parameters include the magnification of the endoscopic image, the layer thickness and the resolution of the CT image.

[0008] Preferably, the lesion structure feature data is analyzed to obtain the structure abnormality coefficient and the abnormal structure monitoring unit of the cross-modality monitoring unit, and the method comprises the following steps: The lesion structure feature data includes three-dimensional volume data and surface roughness data of the suspected lesion in the to-be-fused image sub-region. The volume abnormality coefficient of the to-be-fused image sub-region is obtained by threshold comparison according to the three-dimensional volume data. The surface abnormality coefficient of the to-be-fused image sub-region is obtained by texture analysis according to the surface roughness data. The volume abnormality weight and the surface abnormality weight are set, and the structure abnormality coefficient of the to-be-fused image sub-region is obtained according to the volume abnormality weight and the volume abnormality coefficient, and the surface abnormality weight and the surface abnormality coefficient. The cross-modality monitoring unit corresponding to the to-be-fused image sub-region is marked as the abnormal structure monitoring unit.

[0009] Preferably, the volume abnormality coefficient of the to-be-fused image sub-region is obtained by threshold comparison according to the three-dimensional volume data, and the method comprises the following steps: The three-dimensional volume data includes the actual volume value of the suspected lesion in the to-be-fused image sub-region obtained by three-dimensional reconstruction, and the volume change trend curve of the suspected lesion is generated according to the actual volume value. If the actual volume value in the volume change trend curve is greater than a preset normal volume threshold, a first structure abnormality coefficient of the to-be-fused image sub-region is obtained according to the actual volume value and the normal volume threshold. obtaining a volume growth slope of the volume change trend curve, and if the volume growth slope is greater than a preset volume slope threshold, obtaining a second structural abnormality coefficient of the to-be-fused image sub-region according to the volume growth slope and the volume slope threshold; obtaining a volume abnormality coefficient of the to-be-fused image sub-region according to the first structural abnormality coefficient and the second structural abnormality coefficient.

[0010] Preferably, the surface abnormality coefficient of the to-be-fused image sub-region is obtained according to the texture analysis of the surface roughness data, and specifically includes the following steps: The surface roughness data includes a texture gray deviation value and a texture uniformity of a surface of the suspected lesion in the to-be-fused image sub-region. generating a surface texture curve of the suspected lesion according to the texture gray deviation value and the texture uniformity; if the texture gray deviation value in the surface texture curve is greater than a preset gray deviation threshold, obtaining a first feature abnormality coefficient of the to-be-fused image sub-region according to the texture gray deviation value and the gray deviation threshold; if the texture uniformity in the surface texture curve is less than a preset uniformity threshold, obtaining a second feature abnormality coefficient of the to-be-fused image sub-region according to a difference between the uniformity threshold and the texture uniformity; and obtaining the surface abnormality coefficient of the to-be-fused image sub-region according to the first feature abnormality coefficient and the second feature abnormality coefficient.

[0011] Preferably, the blood flow abnormality coefficient of the to-be-fused image sub-region and the abnormal blood flow monitoring unit of the cross-modality monitoring unit are obtained by analyzing the lesion blood flow feature data, and specifically include the following steps: The lesion blood flow feature data includes a blood flow perfusion and a blood flow velocity of the suspected lesion in the to-be-fused image sub-region. generating a blood flow dynamic curve of the to-be-fused image sub-region according to the blood flow perfusion and the blood flow velocity, and obtaining a blood flow fluctuation amplitude of the blood flow dynamic curve; if the blood flow fluctuation amplitude is greater than a preset blood flow amplitude threshold, obtaining a blood flow abnormality coefficient of the to-be-fused image sub-region according to the blood flow fluctuation amplitude and the blood flow amplitude threshold, and marking the cross-modality monitoring unit corresponding to the to-be-fused image sub-region as an abnormal blood flow monitoring unit.

[0012] Preferably, the signal fusion time consumption of the lesion suspected monitoring unit is obtained according to the lesion suspected monitoring unit and the cross-modality fusion terminal, and specifically includes the following steps: obtaining a start time at which the lesion suspected monitoring unit sends the multi-modality lesion feature data, and obtaining a termination time at which the cross-modality fusion terminal completes multi-modality signal fusion and outputs a fusion result; obtaining an original fusion duration according to the start time and the termination time; Obtain the memory occupation rate of the cross-modal fusion terminal, and correct the original fusion time length according to the memory occupation rate to obtain the signal fusion time consumption of the lesion suspected monitoring unit.

[0013] Preferably, the lesion identification priority of the lesion suspected monitoring unit is obtained according to the fusion signal intensity and the signal fusion time consumption, and specifically includes the following steps: Set the fusion intensity weight and the time consumption correction weight; Obtain the signal contribution coefficient of the lesion suspected monitoring unit according to the fusion signal intensity and the fusion intensity weight; Obtain the time consumption influence coefficient of the lesion suspected monitoring unit according to the signal fusion time consumption and the time consumption correction weight; Obtain the priority score of the lesion suspected monitoring unit according to the signal contribution coefficient and the time consumption influence coefficient, and set the lesion identification priority of the lesion suspected monitoring unit according to the priority score.

[0014] A digestive internal medicine lesion identification system based on object recognition, comprising: A setting module: according to the modal type and imaging parameter of the digestive internal medicine multi-modal image, set the cross-modal monitoring unit of different modal to-be-fused image sub-regions; A first analysis module: according to the cross-modal monitoring unit, extract the lesion structure feature data of the to-be-fused image sub-region, analyze the lesion structure feature data to obtain the structure abnormality coefficient and the abnormal structure monitoring unit of the cross-modal monitoring unit; A second analysis module: according to the cross-modal monitoring unit, extract the lesion blood flow feature data of the to-be-fused image sub-region, analyze the lesion blood flow feature data to obtain the blood flow abnormality coefficient of the to-be-fused image sub-region and the abnormal blood flow monitoring unit of the cross-modal monitoring unit; A processing module: the abnormal structure monitoring unit and the abnormal blood flow monitoring unit constitute the lesion suspected monitoring unit; according to the structure abnormality coefficient and the blood flow abnormality coefficient, obtain the lesion risk degree of the lesion suspected monitoring unit; An identification module: analyze the lesion risk degree, the lesion suspected monitoring unit and the modal fusion terminal to identify the digestive internal medicine lesion.

[0015] Compared with the prior art, the present application has the following beneficial effects: The application realizes the association of different modal image sub-regions through the modal type and imaging parameter setting cross-modal monitoring unit of multi-modal images. When analyzing the lesion structure characteristic data, the structure abnormality is quantified from multiple dimensions such as volume and surface texture by combining three-dimensional volume data and surface roughness data, and the surface abnormality coefficient is calculated through the texture gray deviation value and uniformity, so that the judgment of structure abnormality is more comprehensive and accurate. The blood flow dynamic curve and blood flow fluctuation amplitude generated according to the blood perfusion and velocity are used to determine the blood flow abnormality coefficient, the structure and blood flow abnormality monitoring units are combined into a lesion suspected monitoring unit, and the lesion risk degree is calculated through the structure abnormality coefficient and the blood flow abnormality coefficient, so as to comprehensively evaluate the lesion risk in multiple dimensions, improve the accuracy of lesion identification, and more accurately distinguish between benign and malignant lesions, and reduce misdiagnosis and missed diagnosis. The priority of lesion identification is determined, and the cross-modal fusion terminal processes the lesion suspected monitoring unit according to the priority, so as to realize the priority identification of high-risk lesions. For example, the lesion suspected unit with high risk degree, strong fusion signal and short fusion time is processed preferentially, so that the clinician can obtain the identification result of the key lesion more quickly, and the efficiency of the whole diagnosis process is improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Fig. 1 A step schematic diagram of a digestive internal medicine lesion identification method based on object recognition is provided for the application. Fig. 2 A module schematic diagram of a digestive internal medicine lesion identification system based on object recognition is provided for the application. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0018] In the following description, many specific details are set forth in order to provide a thorough understanding of the application, but the application can also be implemented in other ways different from the description, and those skilled in the art can make similar generalizations without departing from the concept of the application, therefore the application is not limited to the specific embodiments disclosed below.

[0019] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is independent of or selectively excludes other embodiments.

[0020] Reference Figs. 1-2 is shown.

[0021] Embodiments further illustrate the digestive internal medicine lesion identification method and system based on object recognition provided by the application.

[0022] A method for identifying gastrointestinal lesions based on object recognition, the method comprising the following steps: Cross-modal monitoring units are associated with different modal sub-regions of images to be fused based on the modal type and imaging parameter settings of multimodal images in the Department of Gastroenterology. Multimodal imaging in gastroenterology includes different types of imaging methods, such as endoscopic imaging and CT imaging, each with its own imaging characteristics. Imaging parameters are key settings when these images are generated. For example, the magnification of endoscopic images determines the degree to which lesion details are presented, while the slice thickness and resolution of CT images affect the accuracy of their depiction of lesion structures.

[0023] Based on these modalities and imaging parameters, a cross-modal monitoring unit is set up. The function of the cross-modal monitoring unit is to associate image sub-regions that need to be fused from different modalities. For example, for a gastric lesion, both endoscopic and CT images are available. The endoscopic image has a magnification of 20x, while the CT image has a slice thickness of 1mm and a resolution of 512×512. Based on this information, it is determined which sub-regions in these two modalities need to be fused for analysis, and these sub-regions are associated with the cross-modal monitoring unit. The purpose of this is to ensure that subsequent extraction of structural and blood flow characteristic data of the lesion can be performed within the corresponding, associated image sub-regions, laying the foundation for accurate lesion identification.

[0024] Based on the lesion structure feature data of the sub-region of the image to be fused extracted by the cross-modal monitoring unit, the structural abnormality coefficient and the abnormal structure monitoring unit of the cross-modal monitoring unit are obtained by analyzing the lesion structure feature data. Based on the blood flow characteristic data of the lesion in the sub-region of the image to be fused extracted by the cross-modal monitoring unit, the abnormal blood flow coefficient of the sub-region of the image to be fused and the abnormal blood flow monitoring unit of the cross-modal monitoring unit are obtained by analyzing the blood flow characteristic data of the lesion. The abnormal structure monitoring unit and the abnormal blood flow monitoring unit are combined to form a lesion suspected monitoring unit; the lesion risk level of the lesion suspected monitoring unit is obtained based on the structural abnormality coefficient and the blood flow abnormality coefficient. The risk level of lesions, the suspected lesion monitoring unit, and the modality fusion terminal are analyzed to identify lesions in the gastroenterology department.

[0025] The identification of gastroenterological lesions involves analyzing lesion risk levels, suspected lesion monitoring units, and modality fusion terminals, specifically including the following steps: The fusion signal intensity of the suspected lesion monitoring unit is generated based on the lesion risk level, and the lesion fusion signal of the cross-modal monitoring unit is generated based on the fusion signal intensity; the signal fusion time of the suspected lesion monitoring unit is obtained based on the suspected lesion monitoring unit and the cross-modal fusion terminal. The lesion identification priority of the suspected lesion monitoring unit is obtained based on the fusion signal strength and signal fusion time; the cross-modal fusion terminal identifies lesions in the suspected lesion monitoring unit based on the lesion identification priority and the lesion fusion signal.

[0026] The fusion signal strength of the suspected lesion monitoring unit is generated based on the lesion risk level. The fusion signal strength is calculated by multiplying the lesion risk level by the signal gain coefficient, adding the modality adaptation coefficient, and then multiplying by the signal stability coefficient. For example, if the signal gain coefficient is 1.3, the modality adaptation coefficient is 0.3, and the signal stability coefficient is 0.92, then if the lesion risk level is 0.8, the fusion signal strength is 0.8 × 1.3 + 0.3, multiplied by 0.92. Based on this fusion signal strength, the lesion fusion signal of the cross-modal monitoring unit is generated. Specifically, the original fusion duration is obtained, which is the difference between the start time of the suspected lesion monitoring unit sending multimodal lesion feature data and the termination time of the cross-modal fusion terminal completing multimodal signal fusion and outputting the fusion result. Then, the original fusion duration is corrected according to the memory occupancy rate of the cross-modal fusion terminal. Assuming the memory influence coefficient is 0.45, the memory occupancy rate is 0.6, and the original fusion duration is 10 seconds, the signal fusion time is 10 × (1 + 0.6 × 0.45).

[0027] The priority of lesion identification in the suspected lesion monitoring unit is determined based on the fusion signal strength and fusion time. Specifically, a fusion strength weight and a time correction weight are set. The fusion strength weight is 0.6, and the time correction weight is 0.4. The signal contribution coefficient is obtained by multiplying the fusion signal strength by the fusion strength weight, and the time influence coefficient is obtained by multiplying the signal fusion time by the time correction weight. Finally, the signal contribution coefficient and the time influence coefficient are combined to obtain a priority score, and the lesion identification priority is set based on the priority score. For example, if the fusion signal strength is 0.9 and the signal fusion time is 12 seconds, the signal contribution coefficient is 0.9 × 0.6 = 0.54, and the time influence coefficient is 12 × 0.4 = 4.8.

[0028] The cross-modal fusion terminal identifies suspected lesions sequentially by combining the fusion signals of the lesions, according to their priority. Higher-priority suspected lesion monitoring units are processed first, and the terminal uses their fusion signals to determine the presence and specific type of lesion. For example, for a high-priority suspected gastric lesion monitoring unit, the terminal combines its fusion signals to identify whether the lesion is a gastric ulcer or a gastric polyp.

[0029] In the gastroenterology lesion identification process, the cross-modal fusion terminal first receives the lesion identification priority and lesion fusion signal from each suspected lesion monitoring unit. The terminal processes the suspected lesion monitoring units sequentially according to their priority from high to low. For each suspected lesion monitoring unit, the terminal analyzes its lesion fusion signal, which is converted from the fusion signal intensity based on the lesion risk level and contains comprehensive feature information of the lesion's structure and blood flow in multimodal imaging.

[0030] Taking several suspected lesion monitoring units in the stomach as an example, units with higher priority are processed first. The terminal will analyze their fused signals, combining them with a database of typical features of gastrointestinal lesions, such as the manifestation patterns of ulcers, polyps, and tumors in multimodal fused signals, to determine whether the suspected unit contains a lesion and the type of lesion. If the fused signal of a high-priority suspected lesion monitoring unit exhibits comprehensive characteristics such as a high coefficient of volume abnormality, large blood flow fluctuation amplitude, and abnormal surface texture, the terminal identifies it as a high-risk lesion, such as a gastric tumor, and provides the corresponding identification result. Units with lower priority will undergo similar analysis and identification in subsequent steps to ensure that all suspected lesion monitoring units can be accurately identified, thereby providing comprehensive lesion information for clinical diagnosis.

[0031] Imaging parameters include the magnification of endoscopic images, slice thickness of CT images, and resolution.

[0032] The analysis of lesion structural feature data yields structural anomaly coefficients and abnormal structural monitoring units for cross-modal monitoring units. This process includes the following steps: The lesion structural feature data includes the three-dimensional volume data and surface roughness data of suspected lesions in the sub-region of the image to be fused; The volume anomaly coefficient of the sub-region of the image to be fused is obtained by threshold comparison based on the three-dimensional volume data. The surface anomaly coefficient of the sub-region of the image to be fused is obtained by texture analysis based on surface roughness data; Set volume anomaly weights and surface anomaly weights, and obtain the structural anomaly coefficients of the sub-regions of the image to be fused based on the volume anomaly weights and volume anomaly coefficients, and the surface anomaly weights and surface anomaly coefficients. The cross-modal monitoring units corresponding to the sub-regions of the images to be fused are marked as abnormal structure monitoring units.

[0033] First, a threshold comparison is performed on the 3D volume data to obtain the volume anomaly coefficient. The 3D volume data is the actual volume value of the suspected lesion in the sub-region of the image to be fused, obtained through 3D reconstruction, generating a volume change trend curve of the suspected lesion. If the actual volume value is greater than the preset normal volume threshold, the actual volume value is subtracted from the normal volume threshold, then divided by the normal volume threshold, and then multiplied by the volume deviation correction coefficient to obtain the first construction anomaly coefficient. At the same time, the volume growth slope of the volume change trend curve is obtained. If the volume growth slope is greater than the preset volume slope threshold, the volume growth slope is subtracted from the volume slope threshold, then divided by the volume slope threshold, and then multiplied by the slope deviation correction coefficient to obtain the second construction anomaly coefficient. The first construction anomaly coefficient is multiplied by the volume baseline weight, plus the second construction anomaly coefficient multiplied by the slope influence weight, and then divided by the sum of the volume baseline weight and the slope influence weight to obtain the volume anomaly coefficient. For example, if the volume baseline weight is 0.5, the slope influence weight is 0.5, the first structural anomaly coefficient is 0.6, and the second structural anomaly coefficient is 0.7, then the volume anomaly coefficient is (0.6×0.5+0.7×0.5)÷(0.5+0.5)=0.65.

[0034] Next, texture analysis is performed based on the surface roughness data to obtain the surface anomaly coefficient. The surface roughness data includes the texture grayscale deviation value and texture uniformity of the suspected lesion surface in the sub-region of the image to be fused, thereby generating the surface texture curve of the suspected lesion. If the texture grayscale deviation value is greater than the preset grayscale deviation threshold, the texture grayscale deviation value is subtracted from the grayscale deviation threshold, then divided by the grayscale deviation threshold, and then multiplied by the grayscale deviation correction coefficient to obtain the first feature anomaly coefficient. If the texture uniformity is less than the preset uniformity threshold, the uniformity threshold is subtracted from the texture uniformity, then divided by the uniformity threshold, and then multiplied by the uniformity deviation correction coefficient to obtain the second feature anomaly coefficient. The first feature anomaly coefficient is multiplied by the grayscale reference weight, plus the second feature anomaly coefficient multiplied by the uniformity influence weight, and then divided by the sum of the grayscale reference weight and the uniformity influence weight to obtain the surface anomaly coefficient. For example, if the grayscale baseline weight is 0.6, the uniformity influence weight is 0.4, the first feature anomaly coefficient is 0.5, and the second feature anomaly coefficient is 0.6, then the surface anomaly coefficient is (0.5×0.6+0.6×0.4)÷(0.6+0.4)=0.54.

[0035] Then, set the volumetric anomaly weights and surface anomaly weights. Multiply the volumetric anomaly coefficient by the volumetric anomaly weight, add the surface anomaly coefficient multiplied by the surface anomaly weight, and then divide by the sum of the volumetric anomaly weights and surface anomaly weights to obtain the structural anomaly coefficient of the sub-region of the image to be fused. Assuming the volumetric anomaly weight is 0.6, the surface anomaly weight is 0.4, the volumetric anomaly coefficient is 0.65, and the surface anomaly coefficient is 0.54, then the structural anomaly coefficient is (0.65×0.6+0.54×0.4)÷(0.6+0.4)=0.606.

[0036] The cross-modal monitoring units corresponding to the sub-regions of the images to be fused are marked as abnormal structure monitoring units to identify lesion areas with structural abnormalities. For example, if a sub-region of the stomach to be fused has a high structural abnormality coefficient after the above calculation, its corresponding cross-modal monitoring unit is marked as an abnormal structure monitoring unit, indicating that there may be structural abnormalities in the lesion area.

[0037] The volume anomaly coefficient of the sub-region of the image to be fused is obtained by threshold comparison based on the three-dimensional volume data, specifically including the following steps: The three-dimensional volume data includes the actual volume values ​​of suspected lesions in the sub-region of the image to be fused, obtained through three-dimensional reconstruction, and a volume change trend curve of the suspected lesions is generated based on the actual volume values. If the actual volume value in the volume change trend curve is greater than the preset normal volume threshold, the first construction anomaly coefficient of the sub-region of the image to be fused is obtained based on the actual volume value and the normal volume threshold. Obtain the volume growth slope of the volume change trend curve. If the volume growth slope is greater than the preset volume slope threshold, then obtain the second structural anomaly coefficient of the sub-region of the image to be fused based on the volume growth slope and the volume slope threshold. The volume anomaly coefficient of the sub-region of the image to be fused is obtained based on the first and second structural anomaly coefficients.

[0038] In gastroenterology lesion identification, 3D volume data refers to the actual volume value of a suspected lesion in a sub-region of an image to be fused, obtained through 3D reconstruction. Based on this actual volume value, a volume change trend curve of the suspected lesion is generated. For example, for a suspected intestinal lesion, the curve of its actual volume value changing over time or the number of examinations is obtained through 3D reconstruction.

[0039] If the actual volume value in the volume change trend curve is greater than the preset normal volume threshold, the first structural anomaly coefficient is calculated. The actual volume value is subtracted from the normal volume threshold, then divided by the normal volume threshold, and finally multiplied by the volume deviation correction coefficient. For example, if the normal volume threshold is 5 cubic centimeters, the actual volume value is 7 cubic centimeters, and the volume deviation correction coefficient is 1.2, then the first structural anomaly coefficient is (7-5) ÷ 5 × 1.2 = 0.48.

[0040] Obtain the volume growth slope of the volume change trend curve. If the volume growth slope is greater than a preset volume slope threshold, calculate the second structural anomaly coefficient. The calculation method is to subtract the volume slope threshold from the volume growth slope, divide by the volume slope threshold, and then multiply by the slope deviation correction coefficient. Assuming the volume slope threshold is 0.3 cubic centimeters / week, the volume growth slope is 0.5 cubic centimeters / week, and the slope deviation correction coefficient is 1.1, then the second structural anomaly coefficient is (0.5-0.3)÷0.3×1.1≈0.733.

[0041] The volumetric anomaly coefficient of the sub-region of the image to be fused is obtained based on the first and second structural anomaly coefficients. The calculation method involves multiplying the first structural anomaly coefficient by the volume baseline weight, adding the second structural anomaly coefficient multiplied by the slope influence weight, and then dividing by the sum of the volume baseline weight and the slope influence weight. If the volume baseline weight is 0.5 and the slope influence weight is 0.5, then the volumetric anomaly coefficient is (0.48 × 0.5 + 0.733 × 0.5) ÷ (0.5 + 0.5) = 0.6065. This quantification of the degree of volumetric anomaly of suspected lesions in the sub-region of the image to be fused provides crucial volumetric anomaly information for subsequent lesion identification.

[0042] The surface anomaly coefficients of the sub-regions of the images to be fused are obtained by texture analysis based on surface roughness data, specifically including the following steps: Surface roughness data includes texture grayscale deviation and texture uniformity of the surface of suspected lesions in the sub-region of the image to be fused; Surface texture curves of suspected lesions are generated based on texture grayscale deviation and texture uniformity. If the texture grayscale deviation value in the surface texture curve is greater than the preset grayscale deviation threshold, the first feature anomaly coefficient of the image sub-region to be fused is obtained based on the texture grayscale deviation value and the grayscale deviation threshold. If the texture uniformity in the surface texture curve is less than the preset uniformity threshold, the second feature anomaly coefficient of the image sub-region to be fused is obtained based on the difference between the uniformity threshold and the texture uniformity. The surface anomaly coefficients of the image sub-regions to be fused are obtained based on the first and second feature anomaly coefficients.

[0043] In this gastroenterology lesion identification, surface roughness data encompasses the texture grayscale deviation and texture uniformity of the surface of suspected lesions within the sub-region of the image to be fused. Based on these two data points, surface texture curves for suspected lesions are generated. For example, for a suspected lesion in the stomach, the changes in texture grayscale deviation and texture uniformity are obtained through image analysis.

[0044] If the texture grayscale deviation value in the surface texture curve is greater than the preset grayscale deviation threshold, the first feature anomaly coefficient is calculated. The grayscale deviation value is subtracted from the grayscale deviation threshold, then divided by the grayscale deviation threshold, and finally multiplied by the grayscale deviation correction coefficient. For example, if the grayscale deviation threshold is 10, the texture grayscale deviation value is 15, and the grayscale deviation correction coefficient is 1.2, then the first feature anomaly coefficient is (15-10)÷10×1.2=0.6.

[0045] If the texture uniformity in the surface texture curve is less than a preset uniformity threshold, a second feature anomaly coefficient is calculated. The calculation method is to subtract the texture uniformity from the uniformity threshold, then divide by the uniformity threshold, and finally multiply by the uniformity deviation correction coefficient. Assuming the uniformity threshold is 0.8, the texture uniformity is 0.5, and the uniformity deviation correction coefficient is 1.1, then the second feature anomaly coefficient is (0.8-0.5) ÷ 0.8 × 1.1 = 0.4125.

[0046] The surface anomaly coefficient of the sub-region of the image to be fused is obtained based on the first and second feature anomaly coefficients. The first feature anomaly coefficient is multiplied by the grayscale reference weight, and then the second feature anomaly coefficient is multiplied by the uniformity influence weight. This result is then divided by the sum of the grayscale reference weight and the uniformity influence weight. If the grayscale reference weight is 0.6 and the uniformity influence weight is 0.4, then the surface anomaly coefficient is (0.6 × 0.6 + 0.4125 × 0.4) ÷ (0.6 + 0.4) = 0.525. This quantification of the degree of surface texture abnormality of suspected lesions in the sub-region of the image to be fused provides important surface anomaly information for subsequent lesion identification.

[0047] Analyzing the lesion blood flow characteristic data yields the blood flow abnormality coefficient of the sub-region of the image to be fused and the abnormal blood flow monitoring unit of the cross-modal monitoring unit. This process includes the following steps: Lesion blood flow characteristic data include blood perfusion volume and blood flow velocity of suspected lesions in the sub-region of the image to be fused; The blood flow dynamic curve of the sub-region of the image to be fused is generated based on the blood perfusion volume and blood flow velocity, and the blood flow fluctuation amplitude of the blood flow dynamic curve is obtained. If the blood flow fluctuation amplitude is greater than the preset blood flow amplitude threshold, the blood flow abnormality coefficient of the sub-region of the image to be fused is obtained based on the blood flow fluctuation amplitude and the blood flow amplitude threshold, and the cross-modal monitoring unit corresponding to the sub-region of the image to be fused is marked as an abnormal blood flow monitoring unit.

[0048] In this gastroenterology lesion identification, the lesion blood flow characteristic data includes the blood perfusion volume and blood flow velocity of the suspected lesion in the sub-region of the image to be fused. Based on these two data, a blood flow dynamic curve of the sub-region of the image to be fused is generated. For example, for a suspected intestinal lesion, the blood perfusion volume and blood flow velocity change curves over time are obtained by judging the blood flow in its image, and then the blood flow fluctuation amplitude of the curve is obtained, which is the difference between the maximum and minimum values ​​of the blood flow values ​​of adjacent sampling points in the curve.

[0049] If the blood flow fluctuation amplitude exceeds a preset blood flow amplitude threshold, a blood flow abnormality coefficient is calculated. The calculation method is to subtract the blood flow amplitude threshold from the blood flow fluctuation amplitude, divide by the blood flow amplitude threshold, and then multiply by a blood flow fluctuation correction coefficient. For example, if the blood flow amplitude threshold is 20, the blood flow fluctuation amplitude is 30, and the blood flow fluctuation correction coefficient is 1.2, then the basic blood flow abnormality coefficient is (30-20) ÷ 20 × 1.2 = 0.6. The blood flow abnormality coefficient is obtained by multiplying the basic blood flow abnormality coefficient by the perfusion influence weight corresponding to the blood flow perfusion volume, adding the basic blood flow abnormality coefficient multiplied by the velocity influence weight corresponding to the blood flow velocity, and then dividing by the sum of the perfusion influence weight and the velocity influence weight. Assuming the perfusion influence weight is 0.6 and the velocity influence weight is 0.4, then the blood flow abnormality coefficient is (0.6 × 0.6 + 0.6 × 0.4) ÷ (0.6 + 0.4) = 0.6.

[0050] The cross-modal monitoring units corresponding to the sub-regions of the images to be fused are marked as abnormal blood flow monitoring units. By quantifying the degree of blood flow abnormalities in the suspected lesions within these sub-regions, lesion areas with abnormal blood flow are identified, providing crucial blood flow abnormality information for subsequent lesion identification. For example, if a sub-region of the stomach to be fused shows a high blood flow abnormality coefficient after the above calculation, its corresponding cross-modal monitoring unit will be marked as an abnormal blood flow monitoring unit, indicating that there may be abnormal blood flow in that area.

[0051] The signal fusion time of the suspected lesion monitoring unit is obtained based on the suspected lesion monitoring unit and the cross-modal fusion terminal, specifically including the following steps: The start time of the suspected lesion monitoring unit sending multimodal lesion feature data is obtained, and the termination time of the cross-modal fusion terminal completing multimodal signal fusion and outputting fusion results is obtained. The original fusion duration is obtained based on the start and end times; The memory usage rate of the cross-modal fusion terminal is obtained, and the original fusion time is corrected based on the memory usage rate to obtain the signal fusion time of the suspected lesion monitoring unit.

[0052] First, obtain the start time of the suspected lesion monitoring unit sending multimodal lesion feature data, and the end time of the cross-modal fusion terminal completing multimodal signal fusion and outputting the fusion result. For example, if the suspected lesion monitoring unit starts sending data at 9:00:00 AM, and the cross-modal fusion terminal completes fusion and outputs the result at 9:00:10 AM, then the original fusion duration is obtained by subtracting the start time from the end time. Here, the original fusion duration is 10 seconds.

[0053] It is necessary to obtain the memory occupancy rate of the cross-modal fusion terminal, and then correct the original fusion time based on the memory occupancy rate to obtain the signal fusion time of the suspected lesion monitoring unit. Assuming the memory influence coefficient is 0.4, the memory occupancy rate is 0.5, and the original fusion time is 10 seconds, then the signal fusion time is 10 × (1 + 0.5 × 0.4) = 12 seconds. This method can more accurately reflect the actual time spent on multimodal signal fusion under different memory occupancy conditions, providing accurate time consumption data to support the subsequent determination of lesion identification priority.

[0054] The lesion identification priority of the suspected lesion monitoring unit is obtained based on the fusion signal strength and signal fusion time, specifically including the following steps: Set the fusion strength weight and the time correction weight; The signal contribution coefficient of the suspected lesion monitoring unit is obtained based on the fusion signal strength and the fusion strength weight; The time-consuming impact coefficient of the suspected lesion monitoring unit is obtained based on the signal fusion time and time-consuming correction weight; The priority score of the suspected lesion monitoring unit is obtained based on the signal contribution coefficient and the time consumption impact coefficient, and the lesion identification priority of the suspected lesion monitoring unit is set according to the priority score.

[0055] In this gastroenterology lesion identification process, it is necessary to set a fusion intensity weight and a time-consuming correction weight. These two weights are used to balance the impact of fusion signal intensity and signal fusion time on the priority of lesion identification. For example, the fusion intensity weight can be set to 0.6, and the time-consuming correction weight can be set to 0.4.

[0056] The signal contribution coefficient of the suspected lesion monitoring unit is calculated based on the fused signal strength and the fused strength weight. Assuming the fused signal strength is 0.8, the signal contribution coefficient is 0.8 multiplied by 0.6, which equals 0.48.

[0057] The time-consuming impact coefficient of the suspected lesion monitoring unit is calculated based on the signal fusion time and the time-consuming correction weight. For example, if the signal fusion time is 12 seconds, then the time-consuming impact coefficient is 12 multiplied by 0.4, which equals 4.8.

[0058] The priority score of the suspected lesion monitoring unit is calculated based on the signal contribution coefficient and the time-consuming impact coefficient, and the lesion identification priority is set according to the priority score. The priority score can be calculated by dividing the signal contribution coefficient by the time-consuming impact coefficient. If the signal contribution coefficient is 0.48 and the time-consuming impact coefficient is 4.8, then the priority score is 0.48 divided by 4.8, which equals 0.1. The higher the score, the higher the priority of lesion identification, and the cross-modal fusion terminal will prioritize the identification of lesion suspected monitoring units with higher priority. This ensures identification accuracy while optimizing the identification process to improve the efficiency of lesion identification.

[0059] A gastroenterology lesion identification system based on object recognition, comprising: Setting module: Based on the modal type and imaging parameters of multimodal images in the Department of Gastroenterology, set cross-modal monitoring units that associate different modal sub-regions of images to be fused; First analysis module: Extract lesion structure feature data of the sub-region of the image to be fused based on the cross-modal monitoring unit, analyze the lesion structure feature data to obtain the structural abnormality coefficient and the abnormal structure monitoring unit of the cross-modal monitoring unit; The second analysis module extracts lesion blood flow characteristic data of the sub-region of the image to be fused based on the cross-modal monitoring unit, and analyzes the lesion blood flow characteristic data to obtain the blood flow abnormality coefficient of the sub-region of the image to be fused and the abnormal blood flow monitoring unit of the cross-modal monitoring unit; Processing module: The abnormal structure monitoring unit and the abnormal blood flow monitoring unit are combined to form a lesion suspected monitoring unit; the lesion risk level of the lesion suspected monitoring unit is obtained based on the structural abnormality coefficient and the blood flow abnormality coefficient; Identification module: After analyzing the lesion risk level, suspected lesion monitoring unit and modality fusion terminal, it identifies lesions in the gastroenterology department.

[0060] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0061] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying a digestive disease lesion based on object recognition, characterized by, The method comprises the following steps: According to the modality type and imaging parameter setting of the digestive internal medicine multi-modal image, the cross-modality monitoring unit of the sub-region of the image to be fused is associated; According to the cross-modality monitoring unit, the lesion structure feature data of the sub-region of the image to be fused is extracted, and the structure abnormality coefficient and the abnormal structure monitoring unit of the cross-modality monitoring unit are obtained by analyzing the lesion structure feature data; According to the cross-modality monitoring unit, the lesion blood flow feature data of the sub-region of the image to be fused is extracted, and the blood flow abnormality coefficient of the sub-region of the image to be fused and the abnormal blood flow monitoring unit of the cross-modality monitoring unit are obtained by analyzing the lesion blood flow feature data; The abnormal structure monitoring unit and the abnormal blood flow monitoring unit constitute a lesion suspected monitoring unit; the lesion risk degree of the lesion suspected monitoring unit is obtained according to the structure abnormality coefficient and the blood flow abnormality coefficient; The lesion risk degree, the lesion suspected monitoring unit and the modality fusion terminal are analyzed to identify the lesion of the digestive internal medicine. 2.The method of claim 1, wherein, The lesion risk degree, the lesion suspected monitoring unit and the modality fusion terminal are analyzed to identify the lesion of the digestive internal medicine, which comprises the following steps: According to the lesion risk degree, the fusion signal strength of the lesion suspected monitoring unit is generated, the lesion fusion signal of the cross-modality monitoring unit is generated according to the fusion signal strength, and the signal fusion time consumption of the lesion suspected monitoring unit is obtained according to the lesion suspected monitoring unit and the cross-modality fusion terminal; According to the fusion signal strength and the signal fusion time consumption, the lesion identification priority of the lesion suspected monitoring unit is obtained; the lesion suspected monitoring unit is identified according to the lesion identification priority and the lesion fusion signal by the cross-modality fusion terminal. 3.The method of claim 1, wherein the method further comprises: determining a location of the lesion based on the identified object. The imaging parameters include the magnification of the endoscopic image, the layer thickness of the CT image and the resolution. 4.The method of claim 2, wherein, The structure abnormality coefficient and the abnormal structure monitoring unit of the cross-modality monitoring unit are obtained by analyzing the lesion structure feature data, which comprises the following steps: The lesion structure feature data includes three-dimensional volume data and surface roughness data of the suspected lesion in the sub-region of the image to be fused; The volume abnormality coefficient of the sub-region of the image to be fused is obtained by threshold comparison according to the three-dimensional volume data; The surface abnormality coefficient of the sub-region of the image to be fused is obtained by texture analysis according to the surface roughness data; The volume abnormality weight and the surface abnormality weight are set, and the structure abnormality coefficient of the sub-region of the image to be fused is obtained according to the volume abnormality weight and the volume abnormality coefficient, the surface abnormality weight and the surface abnormality coefficient; The cross-modality monitoring unit corresponding to the sub-region of the image to be fused is marked as the abnormal structure monitoring unit. 5.The method of claim 4, wherein the method further comprises: determining a location of the lesion based on the identified object. The volume abnormality coefficient of the sub-region of the image to be fused is obtained by threshold comparison according to the three-dimensional volume data, which comprises the following steps: The three-dimensional volume data includes the actual volume value of the suspected lesion in the sub-region of the image to be fused obtained by three-dimensional reconstruction, and the volume change trend curve of the suspected lesion is generated according to the actual volume value; If the actual volume value in the volume change trend curve is greater than the preset normal volume threshold, the first structure abnormality coefficient of the sub-region of the image to be fused is obtained according to the actual volume value and the normal volume threshold; Obtaining a volume growth slope of the volume change trend curve, and if the volume growth slope is greater than a preset volume slope threshold, obtaining a second structural anomaly coefficient of the to-be-fused image sub-region according to the volume growth slope and the volume slope threshold; Obtaining a volume anomaly coefficient of the to-be-fused image sub-region according to the first structural anomaly coefficient and the second structural anomaly coefficient. 6.The method of claim 5, wherein, Obtaining a surface anomaly coefficient of the to-be-fused image sub-region according to the surface roughness data, specifically including the following steps: The surface roughness data includes a texture gray deviation value and a texture uniformity of a surface of the suspected lesion in the to-be-fused image sub-region; Generating a surface texture curve of the suspected lesion according to the texture gray deviation value and the texture uniformity; If the texture gray deviation value in the surface texture curve is greater than a preset gray deviation threshold, obtaining a first feature anomaly coefficient of the to-be-fused image sub-region according to the texture gray deviation value and the gray deviation threshold; If the texture uniformity in the surface texture curve is less than a preset uniformity threshold, obtaining a second feature anomaly coefficient of the to-be-fused image sub-region according to a difference between the uniformity threshold and the texture uniformity; And obtaining the surface anomaly coefficient of the to-be-fused image sub-region according to the first feature anomaly coefficient and the second feature anomaly coefficient. 7.The method of claim 6, wherein, Analyzing the lesion blood flow characteristic data to obtain a blood flow anomaly coefficient of the to-be-fused image sub-region and an abnormal blood flow monitoring unit of the cross-modality monitoring unit, specifically including the following steps: The lesion blood flow characteristic data includes a blood flow perfusion and a blood flow velocity of the suspected lesion in the to-be-fused image sub-region; Generating a blood flow dynamic curve of the to-be-fused image sub-region according to the blood flow perfusion and the blood flow velocity, and obtaining a blood flow fluctuation amplitude of the blood flow dynamic curve; If the blood flow fluctuation amplitude is greater than a preset blood flow amplitude threshold, obtaining the blood flow anomaly coefficient of the to-be-fused image sub-region according to the blood flow fluctuation amplitude and the blood flow amplitude threshold, and marking the cross-modality monitoring unit corresponding to the to-be-fused image sub-region as an abnormal blood flow monitoring unit. 8.The method of claim 1, wherein the method further comprises: determining a lesion region of interest in the endoscopy image based on the object recognition result. Obtaining a signal fusion time consumption of the lesion suspected monitoring unit according to the lesion suspected monitoring unit and the cross-modality fusion terminal, specifically including the following steps: Obtaining a start time of the lesion suspected monitoring unit sending the multi-modality lesion characteristic data, and obtaining a termination time of the cross-modality fusion terminal completing multi-modality signal fusion and outputting a fusion result; Obtaining an original fusion time length according to the start time and the termination time; Obtaining a memory occupation rate of the cross-modality fusion terminal, and correcting the original fusion time length according to the memory occupation rate to obtain the signal fusion time consumption of the lesion suspected monitoring unit. 9.The method of claim 2, wherein the method further comprises: determining a lesion region of interest in the endoscopy image based on the object recognition result. Obtaining a lesion identification priority of the lesion suspected monitoring unit according to the fusion signal intensity and the signal fusion time consumption, specifically including the following steps: Setting a fusion intensity weight and a time consumption correction weight; Obtaining a signal contribution coefficient of the lesion suspected monitoring unit according to the fusion signal intensity and the fusion intensity weight; Obtaining a time consumption influence coefficient of the lesion suspected monitoring unit according to the signal fusion time consumption and the time consumption correction weight; Obtaining a priority score of the lesion suspected monitoring unit according to the signal contribution coefficient and the time consumption influence coefficient, and setting the lesion identification priority of the lesion suspected monitoring unit according to the priority score.

10. A system for identifying lesions in a digestive system based on object recognition, applied to the method for identifying lesions in a digestive system based on object recognition according to any one of claims 1 to 9, characterized in that, It includes: The setting module: according to the modality type and imaging parameter of the multi-modal image of the digestive department, the cross-modal monitoring unit of the different modalities to be fused image sub-region is set; The first analysis module: according to the cross-modal monitoring unit, the lesion structure feature data of the to-be-fused image sub-region is extracted, the lesion structure feature data is analyzed, the structure abnormality coefficient and the abnormal structure monitoring unit of the cross-modal monitoring unit are obtained; The second analysis module: according to the cross-modal monitoring unit, the lesion blood flow feature data of the to-be-fused image sub-region is extracted, the lesion blood flow feature data is analyzed, the blood flow abnormality coefficient of the to-be-fused image sub-region and the abnormal blood flow monitoring unit of the cross-modal monitoring unit are obtained; The processing module: the abnormal structure monitoring unit and the abnormal blood flow monitoring unit constitute the lesion suspected monitoring unit; according to the structure abnormality coefficient and the blood flow abnormality coefficient, the lesion risk degree of the lesion suspected monitoring unit is obtained; The identification module: after analyzing the lesion risk degree, the lesion suspected monitoring unit and the modality fusion terminal, the lesion of the digestive department is identified.

Images