An AI-based gastric image analysis and processing method and system

Abstract

The present application discloses an analysis and processing method and system for gastric images based on AI, and relates to the technical field of image analysis and processing. The method includes: obtaining all gastroscopic images taken by a gastroscope, determining whether the gastroscopic images are valid, and screening to obtain valid images; obtaining gastric abnormalities of valid images, and screening to obtain main images; obtaining the viewing habits of the attending physician, and screening to obtain key images; extracting image information of key images, obtaining associated information of the image information, searching for gastroscopic images corresponding to the associated information and recording them as associated images; obtaining the diagnostic experience of the attending physician, selecting the presentation angles of key images and associated images, and obtaining key images after processing the images; sorting the key images according to the viewing habits of the attending physician, obtaining the image sequence, and packaging the key images according to the image sequence to form a gastroscopic image report. The present application improves the convenience of doctors in diagnosing diseases.

Description

Technical Field

[0001] The present application relates to the technical field of image analysis and processing, and in particular to an AI-based gastric image analysis and processing method and system. Background Art

[0002] Gastroscopic image analysis is an important field in medical imaging. It diagnoses diseases of the upper gastrointestinal tract by carefully observing and analyzing the images taken by the gastroscope. Gastroscopic images play a vital role in doctors' diagnosis and treatment of patients. At present, the gastroscopy operator and the attending physician in the hospital are usually not the same person, which will result in the gastroscopic images analyzed and processed by the operating physician not being suitable for the diagnostic habits of all attending physicians. Similarly, different operating physicians analyze and process gastroscopic images differently. Gastroscopic images that do not fit the diagnostic habits and needs of the attending physician will increase the difficulty of the attending physician's diagnosis and treatment, and bring inconvenience to doctors and patients. Summary of the invention

[0003] The purpose of the present invention is to provide an AI-based gastric image analysis and processing method and system to solve the problems raised in the above-mentioned background technology.

[0004] In the first aspect, the present application provides an AI-based gastric image analysis and processing method, which adopts the following technical solution:

[0005] Obtain all gastroscopic images taken by a gastroscope, determine whether the gastroscopic images are valid, and screen out valid images;

[0006] Obtaining stomach abnormalities of valid images, and screening the valid images to obtain a main image according to the stomach abnormalities;

[0007] Obtain the attending physician's viewing habits of gastroscopy images, and select key images from the main images based on the viewing habits;

[0008] Extracting image information of the key image, obtaining associated information of the image information, searching for gastroscopic images corresponding to the associated information and recording them as associated images;

[0009] Obtain the diagnosis experience of the attending physician, select the presentation angles of the key images and related images based on the diagnosis experience, and obtain the key images after processing the images according to the presentation angles;

[0010] According to the viewing habits of the attending physician, the key images are sorted to obtain the image sequence, and the key images are packaged according to the image sequence to form a gastroscopy image report.

[0011] Preferably, the step of obtaining the attending physician's viewing habits of gastroscopic images and screening key images from main images according to the viewing habits is specifically as follows:

[0012] Obtain the historical records of the attending physician viewing gastroscopic images, and count the browsing time of the attending physician viewing different gastroscopic images based on the historical records;

[0013] Setting a browsing time threshold, filtering gastroscope images that reach the browsing time threshold and recording them as the first images;

[0014] Counting the magnified part images of the gastroscopic images in the historical records and recording them as the second images;

[0015] The first image and the second image are combined to form a historical image set, and the same image features of the historical images are counted and used as image focuses;

[0016] The importance of the main image is obtained, and the key image is screened from the main image in combination with the image focus.

[0017] Preferably, the step of obtaining the importance of the main image and screening the key image from the main image in combination with the image focus is specifically as follows:

[0018] Obtain historical gastroscopic images, and estimate the basic patient condition based on the comparison between the main image and the historical gastroscopic images;

[0019] Determine whether the image focus is useful for basic patient condition judgment. If not, obtain the importance of the main image to the basic patient condition judgment and record it as the first importance, and select the key image according to the first importance;

[0020] If useful, the importance of the image focus to the basic patient condition judgment is obtained and recorded as the second most important level;

[0021] Extract the image focus content ratio in the main image, and multiply the second importance level and the image focus content to obtain the third importance level;

[0022] The first importance level and the third importance level are superimposed to obtain the image importance level, and the gastroscopic images whose image importance levels reach a preset image importance level threshold are selected as key images.

[0023] Preferably, the steps of extracting image information of the key image, obtaining associated information of the image information, searching for gastroscopic images corresponding to the associated information and recording them as associated images are specifically:

[0024] Extract the image information of the key image, search for the associated information of the image information from the hospital medical record database and record it as the basic associated information;

[0025] Obtain patient information associated with the patient's basic patient condition and determine whether the basic associated information is associated with the patient information;

[0026] If the basic related information is not related to the patient information, the influence of the basic related information on the judgment of the condition is obtained, and the related images are obtained by screening according to the judgment influence;

[0027] If the basic related information is related to the patient information, determine whether the patient information is a sufficient condition for the basic related information;

[0028] The basic associated information corresponding to the sufficient condition of the patient information being the basic associated information is selected and recorded as the primary associated information;

[0029] Find the valid image corresponding to the primary associated information and record it as the associated image.

[0030] Preferably, the step of obtaining the influence of the basic correlation information on the judgment of the disease condition and selecting and obtaining the correlation image according to the judgment of the influence condition is specifically:

[0031] Extract the effective image corresponding to the basic correlation information and use it as the primary image, and estimate the patient's condition again based on the primary image and the main image;

[0032] Determine whether the basic patient condition is consistent with the estimated patient condition, and if so, obtain the degree to which the primary image assists the estimated patient condition;

[0033] Obtain the attending physician's diagnosis experience of the basic patient's condition and record it as condition experience, and find the corresponding auxiliary effect degree according to the preset condition experience-auxiliary effect degree table and record it as the standard effect degree;

[0034] If the auxiliary effect of the elementary image reaches the standard effect, the elementary image is recorded as a related image;

[0035] If the disease conditions are inconsistent, the elementary image is recorded as the associated image.

[0036] Preferably, the step of obtaining the diagnosis experience of the attending physician, selecting the presentation angles of the key image and the associated image according to the diagnosis experience, and obtaining the key image after processing the image according to the presentation angle is specifically as follows:

[0037] Obtaining the attending physician's medical information, including the duration of treatment, misdiagnosis rate, and number of types of conditions treated;

[0038] Obtain key images and related images taken at the same location and record them as location images, determine whether the images need to be processed based on the medical information, and if so, process the images to form a key image;

[0039] If the image does not need to be processed, the presentation angles of the images at different positions are obtained, and the position images are selected as key images according to the presentation angles.

[0040] Preferably, the step of acquiring the key image and the associated image taken at the same location and recording them as location images, determining whether the image needs to be processed according to the medical information, and if the image needs to be processed, forming the key image after processing the image, specifically comprises:

[0041] Obtaining a viewing position according to the position image, and determining an actual viewing difficulty of the viewing position;

[0042] Determine the depth of the attending physician's diagnostic experience based on the medical information, and set the standard viewing difficulty of the attending physician based on the determined experience depth;

[0043] Determine whether the standard viewing difficulty is less than the actual viewing difficulty. If the standard viewing difficulty is less than the actual viewing difficulty, it is determined that the location image needs to be processed;

[0044] The attending physician's familiarity with the basic patient's condition is deeply extracted based on diagnostic experience, and a familiarity threshold is set. If the familiarity does not reach the familiarity threshold, image registration is performed based on the position image to obtain a 3D image as the key image;

[0045] If the familiarity reaches the familiarity threshold, the key information of the location image is selected and locally enhanced as the key image.

[0046] Preferably, the step of obtaining the viewing position according to the position image and determining the actual viewing difficulty of the viewing position is specifically:

[0047] Obtain the curvature of the cavity and the degree of fold obstruction at the viewing position;

[0048] The lesion type corresponding to the location is estimated based on the location image, and the probability of the lesion type being discovered is calculated and recorded as the discovery probability;

[0049] Extracting lesion characteristics in the position image, the lesion characteristics including lesion size, lesion morphology and lesion texture, and determining the lesion visibility based on the lesion characteristics;

[0050] The weight ratios of the curvature of the lumen, the obstruction of the folds, the probability of discovery and the visibility of the lesion are set respectively, and the actual viewing difficulty is calculated according to the curvature of the lumen, the obstruction of the folds, the probability of discovery and the visibility of the lesion and the corresponding weight ratios.

[0051] Preferably, if the image does not need to be processed, the step of obtaining the presentation angles of the images at different positions and selecting the position images as the key images according to the presentation angles is specifically as follows:

[0052] Extracting lesion information from images at different positions, obtaining presentation angles of images at different positions, searching for lesion information corresponding to the presentation angles and recording them as angle information;

[0053] If the presentation angles of the position images are consistent, the position image with the largest angle information content is selected as the primary image;

[0054] Determine whether the angle information of the first-level images is consistent. If the angle information is consistent, select the position image with higher image quality as the key image;

[0055] If the angle information is inconsistent, the position image will not be screened.

[0056] In the second aspect, the present application provides an AI-based gastric image analysis and processing system, which adopts the following technical solution:

[0057] An AI-based gastric image analysis and processing system, comprising:

[0058] The valid image module obtains all gastroscopic images taken by the gastroscope, determines whether the gastroscopic images are valid, and screens out valid images;

[0059] The main image module obtains the stomach abnormality of the effective image, and selects the main image from the effective image according to the stomach abnormality;

[0060] The key image module obtains the attending physician's viewing habits of gastroscopy images and selects key images from the main images based on the viewing habits;

[0061] The associated image module extracts the image information of the key image, obtains the associated information of the image information, searches for the gastroscope image corresponding to the associated information and records it as the associated image;

[0062] The key image module obtains the diagnosis experience of the attending physician, selects the presentation angles of the key image and the related images according to the diagnosis experience, and obtains the key image after processing the image according to the presentation angle;

[0063] The image report module sorts the key images according to the viewing habits of the attending physician, obtains the image sequence, and packages the key images according to the image sequence to form a gastroscopy image report.

[0064] In summary, the present application includes at least one of the following beneficial technical effects:

[0065] 1. Valid images are obtained by screening based on whether the gastroscopic images are valid, and main images are obtained by screening based on the stomach abnormalities corresponding to the valid images. Key images are screened based on the attending physician's viewing habits of gastroscopic images, and related images of the key images are found. The presentation angles of key images and related images are selected based on diagnostic experience, and the key images are obtained after the images are processed based on the presentation angles. Finally, the key images are sorted and packaged to form a gastroscopic image report. Without affecting the diagnosis, gastroscopic images that are suitable for the doctor's viewing habits and diagnostic experience are selected as much as possible, and a gastroscopic image report is formed according to the doctor's viewing habits, which can effectively improve the diagnostic efficiency and the convenience of the doctor in diagnosing the condition.

[0066] 2. Based on the historical records of gastroscopic images viewed by the attending physician, the image focus of the attending physician is extracted, and the gastroscopic images are screened according to the importance of the image focus and the information in the gastroscopic image to the judgment of the patient's condition. This not only satisfies the screening of important information in gastroscopic images, but also meets the viewing habits of doctors, effectively improves the diagnosis efficiency of doctors, and improves the intelligence of the analysis and processing of gastric images based on AI.

[0067] 3. The depth of the attending physician's diagnostic experience can be determined by the duration of diagnosis and treatment, the misdiagnosis rate, and the number of types of conditions treated. The depth of diagnostic experience can be used to determine whether the image needs to be processed. If image processing is not required, the gastroscopic images that meet the physician's experience are screened based on the presentation angle of the gastroscopic image and the depth of diagnostic experience. Selecting gastroscopic images to be processed according to the actual situation of the attending physician and selecting gastroscopic images at different angles can help the physician make a faster diagnosis and improve the convenience of the physician's diagnosis. The analysis and processing of gastric images based on AI can be tailored to the individual, which reflects the high adaptability of the analysis and processing of gastric images based on AI. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the specific steps of an embodiment of an AI-based gastric image analysis and processing method of the present invention.

[0069] Figure 2 It is a module connection diagram of an embodiment of an AI-based gastric image analysis and processing system of the present invention. DETAILED DESCRIPTION

[0070] Below is a combination of the embodiments and Figure 1-Figure 2 The present invention is further described in detail, but the embodiments of the present invention are not limited thereto.

[0071] The present invention discloses an analysis and processing method of stomach images based on AI, which specifically comprises the following steps:

[0072] Step S1, obtaining all gastroscopic images taken by a gastroscope, determining whether the gastroscopic images are valid, and screening to obtain valid images.

[0073] Some gastroscopy images are considered invalid because they are of low quality and cannot play an effective role in diagnosis. For example, when a gastroscopy image is overexposed, the entire field of view becomes too bright, resulting in the inability to clearly display the mucosal details. Therefore, screening out some blurred and unclear images is helpful to reduce the interference of invalid images on doctors.

[0074] Step S2, obtaining stomach abnormalities of valid images, and screening the valid images to obtain a main image according to the stomach abnormalities.

[0075] Abnormality of the stomach in the effective image refers to the difference between the patient's stomach and the normal standard stomach shown in the gastroscopy image. When the patient's stomach is different from the standard stomach in terms of morphology, structure and other parameters, it indicates that there may be a lesion in the stomach, which requires special attention during the diagnosis process. The abnormal image can be obtained by comparing the effective image with the standard gastroscopy image.

[0076] Step S3, obtaining the attending physician's viewing habits of gastroscopic images, and filtering key images from the main images according to the viewing habits.

[0077] Step S4, extracting the image information of the key image, obtaining the associated information of the image information, searching for the gastroscope image corresponding to the associated information and recording it as the associated image.

[0078] Step S5, obtaining the diagnosis experience of the attending physician, selecting the presentation angles of the key image and the associated images according to the diagnosis experience, and obtaining the key image after processing the images according to the presentation angles.

[0079] Step S6, sorting the key images according to the viewing habits of the attending physician to obtain the image sequence, and packaging the key images according to the image sequence to form a gastroscopy image report.

[0080] Different doctors have different habits of viewing gastroscopy image reports. For example, if a patient is suspected of having gastric cancer, the doctor will give priority to carefully observing the prone areas such as the gastric angle and gastric antrum. Some doctors are accustomed to observing from the location of the lesion, that is, viewing smaller local gastroscopy images. Some doctors are more accustomed to viewing global-related images first.

[0081] In actual use, during the gastroscopy process, multiple gastroscopy images are usually taken. These images need to be screened and processed to form a gastroscopy report and sent to the attending physician. Because sending all gastroscopy images to the attending physician will not only increase the diagnosis time, but also increase the workload of the attending physician and interfere with the diagnosis. Different doctors have different diagnostic experience and diagnostic habits. Providing gastroscopy images that are more in line with the doctor's habits will help doctors diagnose the patient's condition more quickly and accurately. Therefore, according to the image information of the gastroscopy image, the viewing habits of the attending physician, and the diagnostic experience, the gastroscopy image report formed after analyzing and processing the gastroscopy image improves the convenience of diagnosis.

[0082] The steps of obtaining the viewing habits of the attending physician for gastroscopic images and filtering key images from the main images according to the viewing habits are as follows:

[0083] Step S31, obtaining the historical records of the attending physician viewing the gastroscopic images, and counting the browsing time of the attending physician viewing different gastroscopic images according to the historical records.

[0084] Step S32, setting a browsing time threshold, screening gastroscope images that reach the browsing time threshold and recording them as first images.

[0085] Step S33, collecting statistics of the magnified part images of the gastroscope images in the historical records and recording them as the second images.

[0086] Step S34, merging the first image and the second image into a historical image set, counting the same image features of the historical images and using them as image focuses.

[0087] Step S35, obtaining the importance of the main image, and filtering the key image from the main image in combination with the image focus.

[0088] In actual application, different doctors have different habits of viewing gastroscopic images. Some doctors may focus more on observing subtle changes in the color, morphology and texture of the gastric mucosa to infer the condition. Other doctors may pay more attention to observing the overall structure of the stomach, the size, shape and location of the lesions and other macroscopic features. Therefore, different doctors have different focuses on gastroscopic images. In the historical records of the attending physician viewing gastroscopic images, the longer the browsing time, the higher the attention the attending physician pays to the image. In the viewing of gastroscopic images, some key contents will be magnified and carefully observed, that is, the images that the doctor magnifies and observes are also the focus of the doctor's attention. The common features of the historical images that the doctor focuses on are extracted to obtain the doctor's focus, so as to select gastroscopic images that are more suitable for the attending physician. For example, the doctor is accustomed to magnifying the gastroscopic images of the gastric mucosa. The magnified gastric mucosa images are more conducive to observing the texture of the gastric mucosa, so the doctor's focus is on the gastric mucosa texture.

[0089] The steps of obtaining the importance of the main image and filtering the key image from the main image in combination with the image focus are as follows:

[0090] Step S351, obtaining historical gastroscopic images, and estimating the basic patient condition based on the comparison between the main image and the historical gastroscopic images.

[0091] Step S352, determine whether the image focus is useful for basic patient condition judgment. If not, obtain the importance of the main image to the basic patient condition judgment and record it as the first importance, and select the key image according to the first importance.

[0092] The importance is evaluated based on the necessity of the image information in the main image for judging the patient's condition. For example, if the basic patient's condition cannot be judged without image A, then the importance is 100%.

[0093] Step S353: If it is useful, the importance of the image focus to the basic patient condition judgment is obtained and recorded as the second importance.

[0094] Step S354, extracting the image focus content ratio in the main image, and multiplying the second importance level and the image focus content to obtain the third importance level.

[0095] Step S355: superimpose the first importance level and the third importance level to obtain the image importance level, and select the gastroscope images whose image importance levels reach a preset image importance level threshold as key images.

[0096] In actual application, when screening the main images, we should not only consider the doctor's focus, but also the patient's condition presented by the gastroscopic image. According to the comparison between the gastroscopic image and the historical images in the hospital, the patient's condition can be preliminarily judged. If the doctor's focus is not related to the patient's condition, it is obviously inappropriate to select the gastroscopic image based on the focus. The gastroscopic image that can reflect the patient's condition should be selected. When the doctor's focus is useful for the diagnosis of the patient's condition, the importance of the main image is obtained based on the image focus content and the importance of the image focus. For example, the image focus is the stomach morphology, but in fact the probability that the patient's condition affects the stomach morphology is small, and the display content of the stomach morphology in the gastroscopic image is also small, so the second importance is small. The first importance of the main image for the basic patient's condition judgment is superimposed to obtain the importance of the main image, and the screening is performed according to the threshold. According to the importance of the image to the judgment of the condition and the doctor's habitual focus, the image is comprehensively screened, and the image that is more suitable for the doctor is selected under the condition of meeting the gastroscopic image diagnosis.

[0097] The steps of extracting image information of the key image, obtaining associated information of the image information, searching for gastroscopic images corresponding to the associated information and recording them as associated images are specifically as follows:

[0098] Step S41, extracting the image information of the key image, searching the hospital medical record database to obtain the associated information of the image information and recording it as the basic associated information.

[0099] Step S42, obtaining patient information associated with the basic patient condition, and determining whether the basic associated information is associated with the patient information.

[0100] Step S43: If the basic associated information is not associated with the patient information, the influence of the basic associated information on the judgment of the disease condition is obtained, and the associated image is screened according to the judgment influence.

[0101] If the basic associated information is not associated with the patient information, the basic associated information is directly screened based on the impact of the basic associated information on the judgment of the condition. For example, when the gastroscopy image shows an ulcer or tumor in the gastric antrum, the doctor usually checks the mucosal images of the gastric body and fundus. Although the mucosa in these parts may look normal, their morphology, color, and arrangement of folds are important for evaluating the health of the entire stomach and the progression of the disease. It has nothing to do with the patient's personal information, and all patients need to view this associated information.

[0102] Step S44: If the basic associated information is associated with the patient information, it is determined whether the patient information is a sufficient condition for the basic associated information.

[0103] For example, after diagnosing gastric ulcer based on gastroscopy images, doctors may also pay attention to whether there are precancerous lesions such as intestinal metaplasia and atypical hyperplasia in other parts of the stomach mucosa for elderly patients or patients with a long history of gastric ulcer. This related information is related to patient information, that is, if the patient is younger and has no history of gastric ulcer, there is no need to check the gastric mucosa in so many parts.

[0104] Step S45, screening basic related information corresponding to the sufficient condition for the patient information to be the basic related information and recording it as primary related information.

[0105] When basic related information is related to patient information, it is necessary to retrieve it based on the patient information. For example, if a patient has a history of gastric ulcer, it is necessary to view gastroscopy images of other parts of the gastric mucosa. If the patient has a history of gastric ulcer, it is necessary to view it. If not, it is not necessary.

[0106] Step S47, searching for a valid image corresponding to the primary associated information and recording it as an associated image.

[0107] In actual use, in gastroscopic images, abnormal gastroscopic images are generally checked to calculate the lesion. However, images with good stomach conditions do not mean that they are not needed. Sometimes these images can also assist in the diagnosis of the disease. However, it is obvious that all normal stomach images are sent to doctors, which wastes the doctor's viewing time and affects the diagnosis time. Therefore, according to the patient's condition, relevant images are selected and presented to the doctor, which not only reduces the doctor's burden, but also enhances the accuracy of the diagnosis results. For example, an irregular bulge or ulcerative lesion can be seen in the gastric antrum, with a dirty surface, irregular edges, and may be accompanied by bleeding or necrosis, indicating the possibility of gastric cancer. According to the historical information in the medical record library, it can be obtained that the tumor information of the gastric antrum is related to the tumor, ulcer, and inflammation information of other gastric parts such as the fundus. Therefore, even if the gastric mucosa in other parts such as the fundus, cardia, and pylorus is smooth, normal in color, and no obvious abnormalities are found. However, doctors still need to check to assess whether gastric cancer exists in isolation, or coexists with other lesions such as chronic gastritis and gastric ulcers, and whether there are signs of distant metastasis.

[0108] The steps of obtaining the influence of basic related information on the judgment of the disease condition and selecting and obtaining related images according to the judgment influence are specifically as follows:

[0109] Step S431, extracting the valid image corresponding to the basic association information and using it as the primary image, and re-estimating the patient's condition based on the primary image and the main image.

[0110] Step S432, determining whether the basic patient condition is consistent with the estimated patient condition. If the conditions are consistent, obtaining the degree of auxiliary effect of the primary image on the estimated patient condition.

[0111] Step S433, obtain the attending physician's diagnosis experience of the basic patient's condition and record it as condition experience, and find the corresponding auxiliary effect degree according to the preset condition experience-auxiliary effect degree table and record it as the standard effect degree.

[0112] Step S434: If the auxiliary effect degree of the primary image reaches the standard effect degree, the primary image is recorded as a related image.

[0113] Step S435: If the disease conditions are inconsistent, the primary image is recorded as a related image.

[0114] In actual application, if the image corresponding to the associated information does not change the diagnosis of the disease, it is necessary to determine whether the primary image is useful for the diagnosis of the disease. If the auxiliary effect on the diagnosis of the disease is not great, it is considered redundant information and does not need to be presented to the doctor for review. Doctors have different diagnostic experience and different information needs for gastroscopic images. For less experienced doctors, more gastroscopic images are needed to assist in diagnosis, while for more experienced doctors, some gastroscopic images that do not play a great auxiliary role can be omitted to speed up the diagnosis. The doctor's experience can be evaluated based on the number of years of practice and the misdiagnosis rate. If the primary image changes the result of the disease diagnosis, it means that even if the primary image shows no abnormality, it is still crucial to the diagnosis process and needs to be presented to the doctor for review.

[0115] The steps of obtaining the diagnosis experience of the attending physician, selecting the presentation angles of the key image and the related images according to the diagnosis experience, and obtaining the key image after processing the images according to the presentation angles are as follows:

[0116] Step S51, obtaining the attending physician's medical information, the medical information including the duration of diagnosis and treatment, misdiagnosis rate, and the number of types of diseases diagnosed and treated.

[0117] Step S52, obtaining the key image and the associated image taken at the same location and recording them as location images, judging whether the image needs to be processed according to the medical information, and if the image needs to be processed, forming a key image after the image is processed.

[0118] Step S53: if the image does not need to be processed, the presentation angles of the images at different positions are obtained, and the position images are selected as the key images according to the presentation angles.

[0119] In actual application, some doctors are more experienced and can quickly identify some images with tricky angles and give a diagnosis. However, some doctors with less experience have difficulty identifying some difficult-to-observe gastroscopic images due to the limited number of gastroscopic images they have viewed. In order to reduce the difficulty of viewing gastroscopic images, gastroscopic images can be processed.

[0120] The key image and the associated image taken at the same location are obtained and recorded as the location image. It is determined whether the image needs to be processed according to the medical information. If the image needs to be processed, the steps of forming the key image after the image is processed are specifically as follows:

[0121] Step S521, obtaining a viewing position according to the position image, and determining an actual viewing difficulty of the viewing position.

[0122] Step S522, determine the depth of diagnostic experience of the attending physician based on the medical information, and set the standard viewing difficulty of the attending physician based on the determined experience depth.

[0123] The weight ratios of diagnosis and treatment time, misdiagnosis rate, and number of diagnosed and treated disease types are set respectively to calculate the depth of diagnostic experience. The longer the diagnosis and treatment time, the lower the misdiagnosis rate, and the more diagnosed and treated disease types, the deeper the depth of diagnostic experience.

[0124] Step S523, determining whether the standard viewing difficulty is less than the actual viewing difficulty, if the standard viewing difficulty is less than the actual viewing difficulty, determining that the location image needs to be processed.

[0125] Step S524, based on the diagnostic experience, the attending physician's familiarity with the basic patient's condition is deeply extracted, and a familiarity threshold is set. If the familiarity does not reach the familiarity threshold, image registration is performed based on the position image to obtain a 3D image as the key image.

[0126] Step S525 : if the familiarity reaches the familiarity threshold, then the key information of the location image is selected and locally enhanced to be used as the key image.

[0127] In actual application, the more experienced the doctor is, the more difficult the gastroscopic images he can accept. When the location where the gastroscopic image is taken is difficult for the doctor to view, the gastroscopic image at that location can be taken from multiple angles, and a 3D image can be achieved through image registration, which effectively reduces the difficulty of image viewing. Image registration is the process of matching and superimposing two or more images acquired at different times, different sensors or under different conditions.

[0128] The steps of obtaining the viewing position according to the position image and determining the actual viewing difficulty of the obtained viewing position are specifically:

[0129] Step S5211, obtaining the curvature degree of the cavity and the fold obstruction degree at the viewing position.

[0130] Step S5212, estimate the lesion type corresponding to the viewing position based on the position image, calculate the probability of the lesion type being discovered and record it as the discovery probability.

[0131] Step S5213, extracting the lesion characteristics in the position image, the lesion characteristics include lesion size, lesion morphology and lesion texture, and determining the lesion visibility based on the lesion characteristics.

[0132] Step S5214, respectively set the weight ratios of the curvature of the lumen, the occlusion of the folds, the probability of discovery, and the visibility of the lesion, and calculate the actual viewing difficulty based on the curvature of the lumen, the occlusion of the folds, the probability of discovery, and the visibility of the lesion and the corresponding weight ratios.

[0133] In actual use, the curvature of the digestive tract directly affects the passability and observation field of the gastroscope. For example, the descending segment and bulb of the duodenum are more curved, making observation relatively difficult. The folds of the digestive tract mucosa may obscure the lesion site, making observation difficult. Especially in the gastric antrum, gastric angle and other locations, there are many folds, which require special attention. Some types of lesions are difficult to find, so they are also difficult to observe. At the same time, the intuitiveness of the visualization of lesions displayed in different locations is also different. Intuitive gastroscopic images should be able to fully display the overall picture of the lesion, including the size, shape, color, surface characteristics, etc. of the lesion. This helps doctors to accurately classify and grade lesions.

[0134] If the image does not need to be processed, the steps of obtaining the presentation angles of the images at different positions and selecting the position images as the key images according to the presentation angles are as follows:

[0135] Step S531, extracting the lesion information of the images at different positions, obtaining the presentation angles of the images at different positions, searching for the lesion information corresponding to the presentation angles and recording it as angle information.

[0136] Step S532: If the presentation angles of the position images are consistent, the position image with the largest angle information content is selected as the primary image.

[0137] Step S533, determining whether the angle information of the primary images is consistent, if the angle information is consistent, selecting the position image with higher image quality as the key image.

[0138] Step S534: if the angle information is inconsistent, the position image is not screened.

[0139] In actual application, for complex lesions, images from a single angle may not be able to fully display their characteristics. Therefore, doctors need to observe the lesions from different angles to obtain more comprehensive information. Intuitive gastroscopic images should be able to support this multi-angle observation requirement. Therefore, for gastroscopic images at the same angle, select those with more information. For gastroscopic images at different angles, select gastroscopic images containing different information. In this way, while saving gastroscopic images with different information, gastroscopic images from different angles are selected, and the number of gastroscopic images that doctors need to view is reduced as much as possible. While completing the diagnostic task, the workload of doctors is reduced and the convenience of the diagnostic process is improved.

[0140] An AI-based gastric image analysis and processing system, by applying the above-mentioned AI-based gastric image analysis and processing method, comprises:

[0141] The valid image module obtains all gastroscopic images taken by the gastroscope, determines whether the gastroscopic images are valid, and screens out valid images.

[0142] The main image module obtains the stomach abnormality of the effective image and selects the main image from the effective image according to the stomach abnormality.

[0143] The key image module obtains the attending physician's viewing habits of gastroscopy images and selects key images from the main images based on the viewing habits.

[0144] The associated image module extracts the image information of the key image, obtains the associated information of the image information, searches for the gastroscope image corresponding to the associated information and records it as the associated image.

[0145] The key image module obtains the diagnosis experience of the attending physician, selects the presentation angles of the key image and related images based on the diagnosis experience, and obtains the key image after processing the image according to the presentation angle.

[0146] The image report module sorts the key images according to the viewing habits of the attending physician, obtains the image sequence, and packages the key images according to the image sequence to form a gastroscopy image report.

[0147] In actual use, the images are initially screened based on whether they are useful for diagnosis, and then a second screening is performed based on whether the stomach condition shown in the gastroscopic image is abnormal. The key images are selected based on the doctor's viewing habits, and related images are found based on the diagnosis of the related images. The images are processed according to the presentation angles of the key images and related images to obtain the key images. Finally, the key images are sorted to form a gastroscopic image report and sent to the doctor. Without affecting the diagnosis, gastroscopic images that fit the doctor's viewing habits and diagnostic experience are selected as much as possible, and the gastroscopic image report is formed according to the doctor's viewing habits, which can effectively improve the diagnostic efficiency and provide convenience for the doctor.

[0148] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. An AI-based gastric image analysis and processing method, characterized in that: The following steps are involved: Obtain all gastroscopic images taken by a gastroscope, determine whether the gastroscopic images are valid, and screen out valid images; Obtaining stomach abnormalities of valid images, and screening the valid images to obtain a main image according to the stomach abnormalities; Obtain the attending physician's viewing habits of gastroscopy images, and select key images from the main images based on the viewing habits; Extracting image information of the key image, obtaining associated information of the image information, searching for gastroscopic images corresponding to the associated information and recording them as associated images; Obtain the diagnosis experience of the attending physician, select the presentation angles of the key images and related images based on the diagnosis experience, and obtain the key images after processing the images according to the presentation angles; According to the viewing habits of the attending physician, the key images are sorted to obtain the image sequence, and the key images are packaged according to the image sequence to form a gastroscopy image report; The steps of obtaining the diagnosis experience of the attending physician, selecting the presentation angles of the key image and the associated image according to the diagnosis experience, and obtaining the key image after processing the image according to the presentation angle are specifically as follows: Obtaining the attending physician's medical information, including the duration of treatment, misdiagnosis rate, and number of types of conditions treated; Obtain key images and related images taken at the same location and record them as location images, determine whether the images need to be processed based on the medical information, and if so, process the images to form a key image; If the image does not need to be processed, the presentation angles of the images at different positions are obtained, and the position images are selected as the key images according to the presentation angles; The steps of obtaining a key image and a related image taken at the same position and recording them as position images, determining whether the image needs to be processed according to the medical information, and if the image needs to be processed, processing the image to form a key image, are specifically as follows: Obtaining a viewing position according to the position image, and determining an actual viewing difficulty of the viewing position; Determine the depth of the attending physician's diagnostic experience based on the medical information, and set the standard viewing difficulty of the attending physician based on the determined experience depth; Determine whether the standard viewing difficulty is less than the actual viewing difficulty. If the standard viewing difficulty is less than the actual viewing difficulty, it is determined that the location image needs to be processed; The attending physician's familiarity with the basic patient's condition is deeply extracted based on diagnostic experience, and a familiarity threshold is set. If the familiarity does not reach the familiarity threshold, image registration is performed based on the position image to obtain a 3D image as the key image; If the familiarity reaches the familiarity threshold, the key information of the location image is selected and locally enhanced as the key image; The step of obtaining the viewing position according to the position image and determining the actual viewing difficulty of the viewing position is specifically: Obtain the curvature of the cavity and the degree of fold obstruction at the viewing position; The lesion type corresponding to the location is estimated based on the location image, and the probability of the lesion type being discovered is calculated and recorded as the discovery probability; Extracting lesion characteristics in the position image, the lesion characteristics including lesion size, lesion morphology and lesion texture, and determining the lesion visibility based on the lesion characteristics; The weight ratios of the curvature of the lumen, the obstruction of the folds, the probability of discovery and the visibility of the lesion are set respectively, and the actual viewing difficulty is calculated according to the curvature of the lumen, the obstruction of the folds, the probability of discovery and the visibility of the lesion and the corresponding weight ratios.

2. The AI-based gastric image analysis and processing method according to claim 1, characterized in that: The step of obtaining the attending physician's viewing habits of gastroscopic images and screening key images from main images according to the viewing habits is specifically as follows: Obtain the historical records of the attending physician viewing gastroscopic images, and count the browsing time of the attending physician viewing different gastroscopic images based on the historical records; Setting a browsing time threshold, filtering gastroscope images that reach the browsing time threshold and recording them as the first images; Counting the magnified part images of the gastroscopic images in the historical records and recording them as the second images; The first image and the second image are combined to form a historical image set, and the same image features of the historical images are counted and used as image focuses; The importance of the main image is obtained, and the key image is screened from the main image in combination with the image focus.

3. The AI-based gastric image analysis and processing method according to claim 2, characterized in that: The step of obtaining the importance of the main image and selecting the key image from the main image in combination with the image focus is specifically as follows: Obtain historical gastroscopic images, and estimate the basic patient condition based on the comparison between the main image and the historical gastroscopic images; Determine whether the image focus is useful for basic patient condition judgment. If not, obtain the importance of the main image to the basic patient condition judgment and record it as the first importance, and select the key image according to the first importance; If useful, the importance of the image focus to the basic patient condition judgment is obtained and recorded as the second most important level; Extract the image focus content ratio in the main image, and multiply the second importance level and the image focus content to obtain the third importance level; The first importance level and the third importance level are superimposed to obtain the image importance level, and the gastroscopic images whose image importance levels reach a preset image importance level threshold are selected as key images.

4. The AI-based gastric image analysis and processing method according to claim 3, characterized in that: The steps of extracting image information of the key image, obtaining associated information of the image information, searching for a gastroscopic image corresponding to the associated information and recording it as an associated image are specifically as follows: Extract the image information of the key image, search for the associated information of the image information from the hospital medical record database and record it as the basic associated information; Obtain patient information associated with the patient's basic patient condition and determine whether the basic associated information is associated with the patient information; If the basic related information is not related to the patient information, the influence of the basic related information on the judgment of the condition is obtained, and the related images are obtained by screening according to the judgment influence; If the basic related information is related to the patient information, determine whether the patient information is a sufficient condition for the basic related information; The basic associated information corresponding to the sufficient condition of the patient information being the basic associated information is selected and recorded as the primary associated information; Find the valid image corresponding to the primary associated information and record it as the associated image.

5. The AI-based gastric image analysis and processing method according to claim 4, characterized in that: The step of obtaining the influence of the basic related information on the judgment of the disease condition and selecting and obtaining the related images according to the judgment of the influence condition is specifically as follows: Extract the effective image corresponding to the basic correlation information and use it as the primary image, and estimate the patient's condition again based on the primary image and the main image; Determine whether the basic patient condition is consistent with the estimated patient condition, and if so, obtain the degree to which the primary image assists the estimated patient condition; Obtain the attending physician's diagnosis experience of the basic patient's condition and record it as condition experience, and find the corresponding auxiliary effect degree according to the preset condition experience-auxiliary effect degree table and record it as the standard effect degree; If the auxiliary effect of the elementary image reaches the standard effect, the elementary image is recorded as a related image; If the disease conditions are inconsistent, the elementary image is recorded as the associated image.

6. The AI-based gastric image analysis and processing method according to claim 5, characterized in that: If the image does not need to be processed, the step of obtaining the presentation angles of the images at different positions and selecting the position images as the key images according to the presentation angles is specifically as follows: Extracting lesion information from images at different positions, obtaining presentation angles of images at different positions, searching for lesion information corresponding to the presentation angles and recording them as angle information; If the presentation angles of the position images are consistent, the position image with the largest angle information content is selected as the primary image; Determine whether the angle information of the first-level images is consistent. If the angle information is consistent, select the position image with higher image quality as the key image; If the angle information is inconsistent, the position image will not be screened.

7. An AI-based gastric image analysis and processing system, characterized in that: By applying the AI-based gastric image analysis and processing method as described in any one of claims 1 to 6, comprising: The valid image module obtains all gastroscopic images taken by the gastroscope, determines whether the gastroscopic images are valid, and screens out valid images; The main image module obtains the stomach abnormality of the effective image, and selects the main image from the effective image according to the stomach abnormality; The key image module obtains the attending physician's viewing habits of gastroscopy images and selects key images from the main images based on the viewing habits; The associated image module extracts the image information of the key image, obtains the associated information of the image information, searches for the gastroscope image corresponding to the associated information and records it as the associated image; The key image module obtains the diagnosis experience of the attending physician, selects the presentation angles of the key image and the related images according to the diagnosis experience, and obtains the key image after processing the image according to the presentation angle; The image report module sorts the key images according to the viewing habits of the attending physician, obtains the image sequence, and packages the key images according to the image sequence to form a gastroscopy image report.

Images