A hierarchical classification auxiliary management system for electronic medical record information of critical care specialists

Abstract

The present invention relates to the technical field of electronic medical record management, and specifically to a hierarchical classification auxiliary management system for critical care specialist electronic medical record information. The present invention first obtains the electronic medical records of patients with severe pneumonia through a data acquisition module; further obtains the spine area and the corrected rib area in the target image through a spine and rib positioning module; further obtains the pneumonia lesion index of the target image through a lesion analysis module; finally, in a medical record classification module, the electronic medical records are preliminarily divided into severe or mild categories according to text information; the electronic medical records are classified according to the temporal variation characteristics of the pneumonia lesion index of the patient's chest X-ray film, combined with the distribution characteristics of the flake density increased shadow area.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic medical record management, and in particular to a hierarchical classification auxiliary management system for critical care specialist electronic medical record information. Background Art

[0002] With the continuous advancement of medical technology and the rapid development of medical informatization, the electronic medical record system of critical care has become an indispensable part of modern hospitals. As a key area in the medical system, the management and utilization of medical record information of critical care is of great significance to improving the quality of medical care and helping hospitals analyze the development process of critical diseases. Designing a hierarchical classification auxiliary management system for electronic medical record information of critical care is an inevitable trend in the development of medical informatization.

[0003] However, due to the different conditions of the patients themselves, there are certain differences in the diagnostic results and manifestations of severe pneumonia. The traditional grading and classification of electronic medical records are classified by relevant staff based on the diagnostic results in the medical records, which is prone to classification errors. Moreover, grading and classification based only on the initial diagnostic results cannot show the recovery of severe pneumonia in different patients. Summary of the invention

[0004] In order to solve the technical problem that the traditional electronic medical record classification cannot reflect the patient's recovery status, the purpose of the present invention is to provide a hierarchical classification auxiliary management system for critical care specialist electronic medical record information. The technical solution adopted is as follows:

[0005] A hierarchical classification auxiliary management system for critical care specialist electronic medical record information, the system comprising:

[0006] Data acquisition module: used to acquire electronic medical records of patients with severe pneumonia; the electronic medical records at least contain text information describing the symptoms and at least two chest X-rays taken at different times; the chest X-rays are used as target images in the order of acquisition;

[0007] The vertebra and rib positioning module is used to obtain the vertebra area in the target image according to the continuous characteristics of the pixel values ​​of each column of pixels in the binary image of the target image; obtain the initial rib area in the target image according to the grayscale extension characteristics of the adjacent pixels in the vertebra area; select any of the initial rib areas as the target rib area, and extract the rib main body line according to the distribution of pixels in the target rib area; in the target rib area, screen the pixels according to the distribution characteristics of the pixels on the curve parallel to the rib main body line to obtain the corrected rib area;

[0008] Lesion analysis module: used for performing threshold segmentation on the pixel points outside the spine area and the corrected rib area in the target image to obtain the patchy density increased shadow area; according to the number, area and edge fuzzy characteristics of the patchy density increased shadow area of ​​the target image, obtain the pneumonia lesion index of the target image;

[0009] Medical record classification module: used to preliminarily classify the electronic medical record into severe or mild categories according to the text information; classify the electronic medical record according to the temporal variation characteristics of the pneumonia lesion index of the patient's chest X-ray film, combined with the distribution characteristics of the flake-like increased density shadow area.

[0010] Furthermore, the method for acquiring the spinal region includes:

[0011] In the binary image of the target image, the length of the foreground pixel sequence whose pixel values ​​are continuously 1 and the number of breaks of the foreground pixel sequence are counted in each column of pixels; the sum of the lengths of all the foreground pixel sequences corresponding to each column of pixels and the product of the length of the longest foreground pixel sequence are used as a first numerator, the sum of the number of breaks and a preset positive constant divided by zero are used as a first denominator, and the ratio of the first numerator to the first denominator is normalized to obtain the result as the probability of the spine for each column of pixels;

[0012] Select a column of pixel points with the largest probability of the spine as the starting column and divide it into a spine area, and iteratively extend the pixel points to both sides; when the difference between the spine probability of the extended column of pixel points and the spine probability of the starting column is less than a first preset threshold, divide the extended column of pixel points into a spine area; when the difference between the spine probability of the most recently extended column of pixel points on one side and the spine probability of the starting column is greater than or equal to the first preset threshold, stop the extension process on the corresponding side;

[0013] When the pixel points on both sides of the starting column stop extending, the spine region in the target image is obtained.

[0014] Furthermore, the method for obtaining the initial rib region includes:

[0015] Among the adjacent pixels on the left side of the spine region, any pixel is selected as the pixel to be analyzed; when the difference between the grayscale values ​​of a first preset number of pixel points vertically adjacent to the pixel to be analyzed and the grayscale value of the pixel to be analyzed is less than a second preset threshold, the pixel to be analyzed is marked as a left rib pixel;

[0016] In a first preset neighborhood of each left rib pixel point, a neighborhood pixel point whose grayscale value difference with the left rib pixel point is less than a third preset threshold is marked as a new left rib pixel point; a new left rib pixel point is acquired in the first preset neighborhood of the new left rib pixel point, and when no new left rib pixel point exists in the first preset neighborhood of the latest left rib pixel point, the iteration is stopped, and all the left rib pixel points are used as the initial left rib area of ​​the target image;

[0017] Acquire the right rib pixel points of the target image from the right adjacent pixel points of the spine region; acquire new right rib pixel points in the second preset neighborhood of the right rib pixel points to obtain an initial right rib region; the method for acquiring the initial right rib region is similar to the method for acquiring the initial left rib region;

[0018] The initial right rib region and the initial left rib region are used as initial rib regions in the target image.

[0019] Furthermore, a line connecting an upper left pixel point and an upper right pixel point in the target rib region is used as a rib main body line.

[0020] Furthermore, the method for obtaining the modified rib region includes:

[0021] In the target rib area, any pixel point is selected as the target pixel point; a perpendicular line to the rib main line is drawn through the left end point of the rib main line; the rib main line is translated along the perpendicular line through the target pixel point to obtain a reference line of the target pixel point; the number of pixel points from the target pixel point to the left end point of the reference line is taken as a reference number, and the number of pixel points belonging to the target rib area from the target pixel point to the left end point of the reference line is taken as an actual number;

[0022] When the difference between the actual number of pixels and the reference number is greater than a preset difference threshold, the corresponding target pixel is removed from the target rib area, and after analyzing all the pixels in the target rib area, a corrected rib area is obtained.

[0023] Furthermore, the method for obtaining the flake-like high-density shadow area includes:

[0024] The area formed by the pixels in the target image that are outside the spine area and the corrected rib area and whose grayscale values ​​are greater than a preset segmentation threshold is used as a flake-like high-density shadow area.

[0025] Furthermore, the method for obtaining the pneumonia lesion index includes:

[0026] The ratio of the number of pixels in each of the patchy density-increased shadow regions to the gradient average of the edge pixels is used as the pathological factor of each of the patchy density-increased shadow regions;

[0027] The product of the sum of the lesion factors of the patchy density increased shadow areas of the target image and the number of all the patchy density increased shadow areas is used as the pneumonia lesion index of the target image.

[0028] Furthermore, the method of preliminarily classifying the electronic medical records into a serious category or a mild category according to the text information includes:

[0029] The text information is matched with the preset classification keywords of the severe class and the mild class respectively by using the forward maximum matching method. When there is no matching relationship between the text information and the preset classification keywords of the severe class, the electronic medical record is preliminarily classified as the mild class; otherwise, it is classified as the severe class.

[0030] Furthermore, the method for classifying the electronic medical record according to the temporal variation characteristics of the pneumonia lesion index of the chest X-ray of the patient, combined with the distribution characteristics of the flake-like increased density shadow area, includes:

[0031] When the pneumonia lesion index of the chest X-ray of the patient constitutes a lesion index sequence in a time series order; when the elements in the lesion index sequence are monotonically decreasing, the recovery label is a normal recovery class; when the elements in the lesion index sequence are monotonically increasing, the recovery label is an abnormally severe class; when the elements in the lesion index sequence are decreasing, increasing, and then decreasing, the recovery label is a recovery-deterioration-recovery class; when the elements in the lesion index sequence are monotonically decreasing, monotonically increasing, and other than decreasing, increasing, and then decreasing, the recovery label is a to-be-observed class;

[0032] When all the flake-like density-increased shadow areas are on the left side of the corresponding spine area, the position label is left lung pneumonia; when all the flake-like density-increased shadow areas are on the right side of the corresponding spine area, the position label is right lung pneumonia; when the flake-like density-increased shadow areas appear on both sides of the corresponding spine area, the position label is bilateral lung pneumonia;

[0033] The combination of each recovery tag and each location tag is an electronic medical record classification.

[0034] Furthermore, the preset classification keywords of the severe category include at least: dyspnea, impaired consciousness, and renal insufficiency; the preset classification keywords of the mild category include at least: low fever, fatigue, and loss of appetite.

[0035] The present invention has the following beneficial effects:

[0036] The present invention first obtains the electronic medical records of severe pneumonia patients in the data acquisition module to provide a basis for subsequent analysis; further, the spine area in the target image and the corrected rib area are obtained through the spine and rib positioning module, and the pixel area unrelated to the pneumonia symptom is located to exclude the influence of the spine and ribs; further, in the lesion analysis module, the pixel points outside the spine area and the corrected rib area in the target image are threshold segmented to obtain the flake density increased shadow area, accurately locate the lesion area, and facilitate the subsequent acquisition of the pneumonia lesion index; according to the number, area and edge of the flake density increased shadow area of ​​the target image The edge fuzzy features are used to obtain the pneumonia lesion index of the target image, and the severity of pneumonia in the target image is quantified to provide a basis for analyzing the patient's recovery; finally, in the medical record classification module, the electronic medical records are preliminarily divided into severe or mild categories according to the text information, and the electronic medical records are classified from the perspective of text information; according to the temporal change characteristics of the pneumonia lesion index of the patient's chest X-ray, combined with the distribution characteristics of the shadow area with increased flake density, the electronic medical records are classified from the perspective of lesion location and recovery, reflecting the patient's recovery, forming a more detailed medical record management model, which is convenient for relevant personnel to view and perform data analysis. The present invention extracts the flake density increased shadow area in the chest X-ray, analyzes the changes in the pneumonia lesion index, and classifies the electronic medical records in combination with text information, reflecting the patient's symptom degree and recovery, forming a more detailed medical record management model, which is convenient for relevant personnel to view and perform data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A system block diagram of a hierarchical classification auxiliary management system for critical care specialist electronic medical record information provided by an embodiment of the present invention;

[0039] Figure 2 An image of a chest X-ray provided by an embodiment of the present invention;

[0040] Figure 3 A binary image of a chest X-ray provided by an embodiment of the present invention;

[0041] Figure 4 A schematic diagram of a first preset neighborhood provided by an embodiment of the present invention;

[0042] Figure 5 A schematic diagram of a rib main line and a reference line provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a hierarchical classification auxiliary management system for critical care electronic medical record information proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0044] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0045] The specific scheme of the hierarchical classification auxiliary management system of critical care specialist electronic medical record information provided by the present invention is described in detail below with reference to the accompanying drawings.

[0046] See also Figure 1 , which shows a system block diagram of a hierarchical classification auxiliary management system for critical care specialist electronic medical record information provided by an embodiment of the present invention, the system includes: a data acquisition module 101, a spine and rib positioning module 102, a lesion analysis module 103 and a medical record classification module 104.

[0047] Data acquisition module 101: used to obtain electronic medical records of patients with severe pneumonia; the electronic medical records at least contain text information describing the symptoms and at least two chest X-rays taken at different times; the chest X-rays are used as target images in the order of acquisition.

[0048] In an embodiment of the present invention, the electronic medical records of patients with severe pneumonia are mainly graded and classified; considering that electronic medical records usually contain text information describing the symptoms, the patient's condition can be classified through the text information, and the electronic medical records of patients with severe pneumonia will contain chest X-rays of different treatment periods, and the patient's recovery condition can be analyzed through the changes in chest X-rays, so the electronic medical records of patients with severe pneumonia are first obtained; the electronic medical records contain at least text information describing the symptoms and two chest X-rays taken at different times, which provide a basis for subsequent analysis; the chest X-rays are taken as target images in the order of acquisition to facilitate analysis one by one.

[0049] It should be noted that the analysis method for each chest X-ray is the same, and only one chest X-ray is taken as the target image as an example, and no repetitive description is given.

[0050] The spine and rib positioning module 102 is used to obtain the spine region in the target image according to the continuous characteristics of the pixel values ​​of each column of pixels in the binary image of the target image; obtain the initial rib region in the target image according to the grayscale extension characteristics of the adjacent pixels in the spine region; select any initial rib region as the target rib region, and extract the rib main body line according to the distribution of pixels in the target rib region; in the target rib region, screen the pixels according to the distribution characteristics of the pixels on the curve parallel to the rib main body line to obtain the corrected rib region.

[0051] See also Figure 2 , which shows an image of a chest X-ray provided by an embodiment of the present invention. Figure 2 It can be seen that the chest X-ray film contains the spine and ribs, and the image display will show the flake-like increased density of the lungs, so the spine and ribs need to be identified and located in order to eliminate the influence of the spine and ribs.

[0052] Considering that the spine has unique characteristics in the chest X-ray image, the spine is generally displayed in the center, and the human spine is generally vertical, showing a continuous part from top to bottom in the chest X-ray image; considering that the spine will be marked as foreground in the binary image due to its high density, the binary image is more likely to highlight the main structure in the image, which can improve the accuracy and efficiency of spine recognition. Therefore, in the binary image of the target image, the spine area in the target image is obtained according to the continuous characteristics of the pixel values ​​of each column of pixels, and the spine area is located, so as to eliminate the influence of the spine on the recognition of the flake-like increased density shadows of the lungs.

[0053] See also Figure 3 , which shows a binary image of a chest X-ray provided by an embodiment of the present invention. It should be noted that the method of obtaining a binary image is a technical means well known to those skilled in the art. In an embodiment of the present invention, the maximum inter-class variance method is used to obtain the binary image.

[0054] Preferably, in one embodiment of the present invention, the vertebrae are Figure 3 The distribution characteristics of vertical continuous pixel values ​​of 1 are presented in the figure. In each column of pixels, the longer the total length of the foreground pixel sequence with continuous pixel values ​​of 1 and the longer the longest foreground pixel sequence length are, the fewer the number of breaks and the stronger the pixel value continuity characteristics are, indicating that this column of pixels is more likely to be vertebral pixels. Therefore, the total length of the foreground pixel sequence in each column of pixels, the number of breaks and the longest foreground pixel sequence length are used to represent the pixel value continuity characteristics of each column of pixels, and the probability of vertebral bones in each column of pixels can be obtained.

[0055] Considering that the column of pixels with the highest probability of vertebrae must be the pixels of the spine, and due to the compact characteristics of the spine, the pixels of the spine are densely distributed, and each column of pixels in the spine area is closely connected, the probability of the spine is relatively consistent, and other surrounding tissues are more scattered than the spine, so the column of pixels with the highest probability of the spine can be controlled to extend to both sides according to the probability of the spine;

[0056] Based on this, in the binary image of the target image, the length of the foreground pixel sequence whose pixel value is continuously 1 and the number of breaks of the foreground pixel sequence are counted; the sum of the lengths of all foreground pixel sequences corresponding to each column of pixels and the product of the length of the longest foreground pixel sequence are taken as the first numerator, the sum of the number of breaks and a preset positive constant divided by zero are taken as the first denominator, and the ratio of the first numerator to the first denominator is normalized to be used as the spine probability of each column of pixels;

[0057] Select a column of pixels with the largest probability of vertebrae as the starting column and divide it into a vertebrae region, and iteratively extend the pixels to both sides; when the difference between the probability of vertebrae in the extended column of pixels and the probability of vertebrae in the starting column is less than a first preset threshold, divide the extended column of pixels into a vertebrae region; when the difference between the probability of vertebrae in the most recently extended column of pixels on one side and the probability of vertebrae in the starting column is greater than or equal to the first preset threshold, stop the extension process on the corresponding side;

[0058] When the pixels on both sides of the starting column stop extending, the spine region in the target image is obtained.

[0059] As an example, the formula for calculating the probability of a vertebra includes:

[0060] ;

[0061] Where t represents the serial number of the target image; k represents the serial number of each column of pixels; represents the probability of the spine at the pixel point in the kth column in the tth target image; represents the linear normalization function; Represents the number of foreground pixel sequences of the k-th column of pixels in the t-th target image; represents the number of breaks of the k-th column of pixels in the t-th target image; j represents the sequence number of the foreground pixel sequence; The length of the j-th foreground pixel sequence of the k-th column of pixels in the t-th target image is also the number of inner pixels of the j-th foreground pixel sequence; Represents the set of lengths of all foreground pixel sequences at the k-th column of pixels in the t-th target image; represents the length of the longest foreground pixel sequence of the k-th column of pixels in the t-th target image; c represents a preset positive constant that divides by zero, and in this example, c=0.001.

[0062] It should be noted that the difference between the vertebral probability of an extended column of pixel points and the vertebral probability of the starting column is represented by the absolute value of the difference between the two vertebral probabilities. In one embodiment of the present invention, the first preset threshold is 0.1, and the column of pixel points with the largest vertebral probability is taken as the starting column and divided into the vertebral area. A column of pixel points adjacent to the vertebral area and with a vertebral probability difference less than 0.1 is divided into the vertebral area, and the extension and division judgment is performed again.

[0063] It should be noted that, in other embodiments of the present invention, the implementer may also set other preset positive constants that divide by zero, and set other first preset thresholds in [0.05-0.15].

[0064] Considering that the ribs and spine are directly connected, and the grayscale of the pixels on the ribs is relatively consistent, with obvious grayscale extension features, the initial rib area in the target image is obtained based on the grayscale extension features of the adjacent pixels in the spine area, and the rib area is preliminarily located to prepare for further correction of the initial rib area.

[0065] Preferably, in one embodiment of the present invention, considering that the ribs are distributed on both sides of the spine, adjacent pixels in the spine region can extend in different directions when analyzing grayscale extension features, so both sides of the ribs are analyzed separately;

[0066] Among the adjacent pixels on the left side of the spine area, any pixel is selected as the pixel to be analyzed; considering that the rib has a certain width, the rib pixel has a certain grayscale similarity in the vertical direction; when the difference between the grayscale value of the first preset number of pixel points vertically adjacent to the pixel to be analyzed and the grayscale value of the pixel to be analyzed is less than a second preset threshold, the pixel to be analyzed is marked as a left rib pixel;

[0067] In a first preset neighborhood of each left rib pixel point, a neighborhood pixel point whose grayscale value difference with the left rib pixel point is less than a third preset threshold is marked as a new left rib pixel point; a new left rib pixel point is obtained in the first preset neighborhood of the new left rib pixel point, and when there is no new left rib pixel point in the first preset neighborhood of the latest left rib pixel point, the iteration is stopped, and all left rib pixels are used as the initial left rib area of ​​the target image;

[0068] As an example, the first preset number is 10, the second preset threshold is 30, the third preset threshold is 5, and the first preset neighborhood is the area consisting of one pixel below the center pixel and three pixels in a row to the left of the center pixel in the eight neighborhoods with the left rib pixel as the center pixel. Figure 4 , which shows a schematic diagram of a first preset neighborhood provided by an embodiment of the present invention; Figure 4 Each square in the figure is a pixel point, the one in the center is the central pixel point, and the four pixel points framed are the first preset neighborhood of the central pixel point.

[0069] The absolute value of the grayscale value difference between the neighborhood pixel point and the left rib pixel point at the center represents the grayscale value difference between the two, and the neighborhood pixel points corresponding to the grayscale value difference less than 5 are marked as new left rib pixel points; then, a new left rib pixel point is obtained within the first preset neighborhood of the new left rib pixel point, and the grayscale extension characteristics of the adjacent pixel points in the spine area are represented by continuously iteratively extending the pixel points, and the iteration is repeated until the iteration stop condition is met to obtain the initial left rib area.

[0070] Similarly, the right rib pixel of the target image is obtained from the right adjacent pixel of the spine region; within the second preset neighborhood of the right rib pixel, a new right rib pixel is obtained to obtain an initial right rib region; the method for obtaining the initial right rib region is similar to the method for obtaining the initial left rib region, which will not be described in detail;

[0071] As an example, the second preset neighborhood is an area consisting of one pixel below the center pixel and three pixels in a row to the right of the center pixel in the eight neighborhoods with the right rib pixel as the center pixel, which can be represented by a matrix as follows: , where 5 is the central pixel, and 3, 6, 8 and 9 are the second preset neighborhood of the central pixel.

[0072] The initial right rib region and the initial left rib region are used as initial rib regions in the target image.

[0073] It should be noted that when rib pixels are obtained by screening adjacent pixels in the spine region, since there are gaps in the upper and lower distribution of the ribs, there are also gaps in the regions formed by the rib pixels, and there are multiple initial rib regions.

[0074] Considering that the density-enhanced shadow may be connected to the rib area and will be classified into the rib area, it is necessary to select any initial rib area as the target rib area for further analysis; considering that the rib body is relatively regular, the pixel points mistakenly classified into the rib area have a large distribution difference from the rib body, so the rib body line is extracted according to the pixel point distribution in the target rib area, providing a rough reference for the rib direction; in the target rib area, according to the distribution characteristics of the pixel points on the curve parallel to the rib body line, the pixel points are screened to obtain the corrected rib area, remove the pixel interference in the non-rib area, improve the accuracy of rib recognition, and facilitate more accurate analysis of the flake density-enhanced shadow area in the future.

[0075] Preferably, in one embodiment of the present invention, considering that the ribs have obvious boundaries, the upper left and upper right pixel points usually correspond to the edge points of the rib area. By connecting these two points, the line formed can summarize the direction and overall trend of the ribs in the image, so the line between the upper left pixel point and the upper right pixel point in the target rib area is used as the rib main line.

[0076] Preferably, in one embodiment of the present invention, in the target rib area, any pixel point is selected as the target pixel point; a perpendicular line to the rib main line is drawn through the left end point of the rib main line to form a reference line for translating the rib main line, so as to obtain a curve parallel to the rib main line; the rib main line is translated along the perpendicular line through the target pixel point to obtain a reference line of the target pixel point, which is equivalent to projecting the geometric relationship of the rib main line around the target pixel point, thereby constructing a local reference system of the target pixel point;

[0077] The number of pixels from the target pixel to the left end point of the reference line is taken as the reference number, which provides the expected distribution characteristics of the target pixel under the ideal regular distribution condition; the number of pixels belonging to the target rib area among the pixels from the target pixel to the left end point of the reference line is taken as the actual number, which reflects the actual distribution characteristics of the target pixel;

[0078] When the difference between the actual number of pixels and the reference number is greater than a preset difference threshold, the corresponding target pixel is removed from the target rib area, and after analyzing all the pixels in the target rib area, a corrected rib area is obtained.

[0079] As an example, the difference between the actual number and the reference number is expressed by the absolute value of the difference between the actual number and the reference number; the preset difference threshold is 3. When the difference between the actual number and the reference number of target pixels is greater than 3, it means that the target pixel may not meet the regular features of the rib area and is removed. Figure 5 , which shows a schematic diagram of a rib main body line and a reference line provided by an embodiment of the present invention, Figure 5In the figure, the squares represent pixels, and the rib main line, vertical line, reference line, left endpoint, right endpoint, and target pixel are marked. n represents the number of references.

[0080] In another embodiment of the present invention, considering that skeleton extraction is a commonly used morphological image processing method, extracting the skeleton of the target rib area as the rib main line can provide topological structure information and indicate the direction of the ribs, so as to more accurately determine the rib position and shape. Therefore, the skeleton of the target rib area is extracted as the rib main line through the existing Skeletonize function, and the pixel points are screened according to the distribution characteristics of the pixel points on the curve parallel to the rib main line to obtain the corrected rib area.

[0081] Specifically, in the target rib area, any pixel point is selected as the target pixel point; the rib main body line is translated up and down in the vertical direction, and when it passes through the target pixel point, it is used as the reference line of the target pixel point; the number of pixel points belonging to the target rib area on the reference line of the target pixel point is obtained as the actual number, which represents the distribution characteristics of the pixel points on the curve parallel to the rib main body line; the number of pixel points on the rib main body line is obtained as the expected number;

[0082] The absolute value of the difference between the actual number of target pixels and the expected number and the ratio of the actual number to the expected number are taken as the distribution difference of the target pixels. When the distribution difference is greater than 0.1, the target pixels are eliminated. All pixels in the target rib area are traversed to obtain the corrected rib area.

[0083] In another embodiment of the present invention, the initial rib region is directly used as the corrected rib region, that is, the correction process of the rib region is skipped. Among the three methods, this method has the smallest amount of calculation and the lowest accuracy. The implementer can choose a suitable method from the three methods.

[0084] The lesion analysis module 103 is used to perform threshold segmentation on the pixel points outside the spine area and the corrected rib area in the target image to obtain the patchy density increased shadow area; according to the number, area and edge fuzzy characteristics of the patchy density increased shadow area of ​​the target image, the pneumonia lesion index of the target image is obtained.

[0085] After locating the areas where the spine and ribs are located, the interference of the spine and ribs can be eliminated and analysis can be performed in other areas of the image; considering the pathological conditions of patients with severe pneumonia in the target image to be extracted in order to analyze the changing characteristics of the pathological conditions, and thus classify the electronic medical records according to the patient's recovery, it is first necessary to extract the flake-like increased density shadow area; and considering that the flake-like increased density shadow area has a large grayscale difference with the background, the pixel points outside the spine area and the corrected rib area in the target image are threshold segmented to obtain the flake-like increased density shadow area, accurately locate the lesion area, and facilitate the subsequent acquisition of the pneumonia lesion index.

[0086] Preferably, in one embodiment of the present invention, considering that the lesion may be manifested as abnormal density of lung tissue, such as a local increased density area caused by inflammation, exudate or fibrosis. These abnormal areas are usually manifested in the image as flake shadows with grayscale values ​​significantly higher than those of normal lung tissue; therefore, the area formed by the pixels in the target image that are outside the spine area and the corrected rib area and whose grayscale values ​​are greater than the preset segmentation threshold is taken as the flake density increased shadow area.

[0087] As an example, the preset segmentation threshold is an empirical value of 200.

[0088] Considering that the number and area of ​​the flake-like density-increased shadow areas reflect the severity of the patient's symptoms, and the edge blur characteristics reflect the diffusion characteristics of inflammation, the pneumonia lesion index of the target image is obtained according to the number, area and edge blur characteristics of the flake-like density-increased shadow areas of the target image, and the severity of pneumonia in the target image is quantified, providing a basis for analyzing the patient's recovery.

[0089] Preferably, in one embodiment of the present invention, considering that the more pixel points in the flake-like density-increased shadow area in the image, the larger the area, the larger the range of the inflammation area, the greater the severity of pneumonia, and the larger the pneumonia lesion index; the larger the number of flake-like density-increased shadow areas, the more inflammation areas, the greater the severity of pneumonia, and the larger the pneumonia lesion index; when inflammation occurs, fluid exudation and fibrosis in the alveoli will cause uneven density of lung tissue, resulting in blurred edge features in imaging, and the smaller the gradient average value of the edge pixel points of each flake-like density-increased shadow area, the blurred edge, reflecting that the diffusion characteristics of the inflammation are more obvious, the severity of pneumonia is greater, and the pneumonia lesion index is larger;

[0090] Based on this, the ratio of the number of pixels in each patchy density-increasing shadow area to the gradient average of the edge pixels is used as the lesion factor of each patchy density-increasing shadow area.

[0091] The product of the sum of the lesion factors of the patchy density increased shadow areas of the target image and the number of all the patchy density increased shadow areas is used as the pneumonia lesion index of the target image.

[0092] The calculation formula of pneumonia lesion index includes:

[0093] ;

[0094] Where, t represents the serial number of the target image; n represents the serial number of the flake-like high-density shadow area; represents the pneumonia lesion index of the t-th target image; Indicates the number of flake-like high-density shadow areas in the t-th target image; Indicates the number of pixels in the nth flake-like density-increased shadow area in the tth target image; Represents the average gradient of the edge pixels of the nth flake-like density-increased shadow region in the tth target image; Represents the lesion factor of the nth flake-like high-density shadow area in the tth target image.

[0095] Among them, the area of ​​the patchy density increased shadow area is represented by the number of pixels in the patchy density increased shadow area, the pixel value of the patchy density increased shadow area is set to 1, and the pixel values ​​of other areas are set to 0. The connected component analysis method is used to count the patchy density increased shadow areas to obtain the number, and the gradient average of the edge pixels of the patchy density increased shadow area is used as the denominator of the ratio, that is, the reciprocal of the gradient average of the edge pixels of the patchy density increased shadow area, to represent the edge blur feature, reflecting the diffusion feature of inflammation. The blurrier the edge, the more obvious the diffusion feature of inflammation is verified, so as to obtain the pneumonia lesion index of the target image and characterize the severity of pneumonia.

[0096] It should be noted that the use of connected component analysis methods to count, locate edge pixels, and calculate gradient values ​​are all existing technologies and will not be described in detail.

[0097] Medical record classification module 104: used to preliminarily classify electronic medical records into severe or mild categories based on text information; classify electronic medical records based on the temporal variation characteristics of the pneumonia lesion index of the patient's chest X-ray film, combined with the distribution characteristics of the flake-like increased density shadow area.

[0098] Considering that text information and chest X-ray image information are two classification bases for electronic medical records, we first analyze the severity of the symptoms based on the text information and preliminarily divide the electronic medical records into severe or mild categories.

[0099] Preferably, in one embodiment of the present invention, the forward maximum matching method is used to perform keyword matching between the text information and the preset classification keywords of the severe class and the mild class respectively. When there is no matching relationship between the text information and the preset classification keywords of the severe class, the electronic medical record is preliminarily classified into the mild class; otherwise, it is classified into the severe class.

[0100] In one embodiment of the present invention, the preset classification keywords of the severe category include at least: dyspnea, impaired consciousness, and renal insufficiency; the preset classification keywords of the mild category include at least: low fever, fatigue, and loss of appetite.

[0101] It should be noted that the forward maximum matching method is a technical means well known to those skilled in the art. In other embodiments of the present invention, the implementer may also perform data cleaning on the text information, extract the symptom keywords using the bag-of-words model, match the symptom keywords with the preset classification keywords through the Levenshtein distance, and determine that the corresponding two words match when the similarity of the two words exceeds 80%;

[0102] Taking into account that different relevant personnel have different descriptions of the symptoms, the preset classification keywords for the severe category may also include respiratory distress and impaired consciousness; implementers can adjust the preset classification keywords for the severe and mild categories according to actual needs.

[0103] In order to realize the grading and classification of electronic medical records and reflect the patient's recovery status at the same time, taking into account that the distribution of the flake-density increased shadow area reflects the location of the disease, the electronic medical records are finally classified according to the temporal change characteristics of the pneumonia lesion index of the patient's chest X-ray film, combined with the distribution characteristics of the flake-density increased shadow area, to reflect the patient's recovery status.

[0104] Preferably, in one embodiment of the present invention, when the pneumonia lesion index of the chest X-ray of the patient constitutes a lesion index sequence in a time series order; when the elements in the lesion index sequence are monotonically decreasing, the recovery label is a normal recovery class; when the elements in the lesion index sequence are monotonically increasing, the recovery label is an abnormally severe class; when the elements in the lesion index sequence are decreasing, increasing, and then decreasing, the recovery label is a recovery-deterioration-recovery class; when the elements in the lesion index sequence are monotonically decreasing, monotonically increasing, and other than decreasing, increasing, and then decreasing, the recovery label is a to-be-observed class;

[0105] When all the flake-like density-increasing shadow areas are on the left side of the corresponding vertebral region, the position label is left lung pneumonia; when all the flake-like density-increasing shadow areas are on the right side of the corresponding vertebral region, the position label is right lung pneumonia; when the flake-like density-increasing shadow areas appear on both sides of the corresponding vertebral region, the position label is bilateral lung pneumonia;

[0106] The combination of each recovery label and each location label is an electronic medical record classification.

[0107] As an example, the first-level classification of electronic medical records in the system is the symptom severity classification, which is divided into mild or severe categories; the second-level classification is the symptom location classification, which is divided into left-lung pneumonia, right-lung pneumonia, or bilateral lung pneumonia; the tertiary classification is the recovery status classification, which is divided into normal recovery, abnormal severity, recovery-worsening-recovery, or to be observed.

[0108] As an example, in one embodiment of the present invention, several lesion index sequences obtained after data extraction are [0.77, 0.15, 0.86, 0.70, 0.83, 0.18], and the corresponding recovery label is the to-be-observed class; [0.85, 0.78, 0.58, 0.94, 0.30], and the corresponding recovery label is the recovery-deterioration-recovery class; [0.12, 0, 20, 0.75], and the corresponding recovery label is the abnormally severe class; [0.20, 0.19, 0.11, 0.08], and the corresponding recovery label is the normal recovery class; among which the pneumonia lesion index only takes the first two decimal places.

[0109] In summary, in order to solve the technical problem that the traditional electronic medical record classification cannot reflect the patient's recovery status, the present invention proposes a hierarchical classification auxiliary management system for critical care specialist electronic medical record information. The present invention first obtains the electronic medical records of patients with severe pneumonia through the data acquisition module; further through the spine and rib positioning module: obtains the spine area and the corrected rib area in the target image; further through the lesion analysis module, obtains the pneumonia lesion index of the target image; finally, in the medical record classification module, the electronic medical records are preliminarily divided into severe or mild categories according to the text information; the electronic medical records are classified according to the time series change characteristics of the pneumonia lesion index of the patient's chest X-ray film, combined with the distribution characteristics of the flake density increased shadow area.

[0110] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0111] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A hierarchical classification auxiliary management system for critical care specialist electronic medical record information, characterized in that: The system comprises: Data acquisition module: used to acquire electronic medical records of patients with severe pneumonia; the electronic medical records at least contain text information describing the symptoms and at least two chest X-rays taken at different times; the chest X-rays are used as target images in the order of acquisition; The vertebra and rib positioning module is used to obtain the vertebra area in the target image according to the continuous characteristics of the pixel values ​​of each column of pixels in the binary image of the target image; obtain the initial rib area in the target image according to the grayscale extension characteristics of the adjacent pixels in the vertebra area; select any of the initial rib areas as the target rib area, and extract the rib main body line according to the distribution of pixels in the target rib area; in the target rib area, screen the pixels according to the distribution characteristics of the pixels on the curve parallel to the rib main body line to obtain the corrected rib area; Lesion analysis module: used for performing threshold segmentation on the pixel points outside the spine area and the corrected rib area in the target image to obtain the patchy density increased shadow area; according to the number, area and edge fuzzy characteristics of the patchy density increased shadow area of ​​the target image, obtain the pneumonia lesion index of the target image; Medical record classification module: used to preliminarily classify the electronic medical record into a severe class or a mild class according to the text information; classify the electronic medical record according to the temporal variation characteristics of the pneumonia lesion index of the chest X-ray film of the patient, combined with the distribution characteristics of the flake-like increased density shadow area; The method for obtaining the spinal region comprises: In the binary image of the target image, the length of the foreground pixel sequence whose pixel values ​​are continuously 1 and the number of breaks of the foreground pixel sequence are counted in each column of pixels; the sum of the lengths of all the foreground pixel sequences corresponding to each column of pixels and the product of the length of the longest foreground pixel sequence are used as a first numerator, the sum of the number of breaks and a preset positive constant divided by zero are used as a first denominator, and the ratio of the first numerator to the first denominator is normalized to obtain the result as the probability of the spine for each column of pixels; Select a column of pixel points with the largest probability of the spine as the starting column and divide it into a spine area, and iteratively extend the pixel points to both sides; when the difference between the spine probability of the extended column of pixel points and the spine probability of the starting column is less than a first preset threshold, divide the extended column of pixel points into a spine area; when the difference between the spine probability of the most recently extended column of pixel points on one side and the spine probability of the starting column is greater than or equal to the first preset threshold, stop the extension process on the corresponding side; When the pixel points on both sides of the starting column stop extending, the spine region in the target image is obtained.

2. According to claim 1, a hierarchical classification auxiliary management system for critical care specialist electronic medical record information is characterized in that: The method for obtaining the initial rib region comprises: Among the adjacent pixels on the left side of the spine region, any pixel is selected as the pixel to be analyzed; when the difference between the grayscale values ​​of a first preset number of pixel points vertically adjacent to the pixel to be analyzed and the grayscale value of the pixel to be analyzed is less than a second preset threshold, the pixel to be analyzed is marked as a left rib pixel; In a first preset neighborhood of each left rib pixel point, a neighborhood pixel point whose grayscale value difference with the left rib pixel point is less than a third preset threshold is marked as a new left rib pixel point; a new left rib pixel point is acquired in the first preset neighborhood of the new left rib pixel point, and when no new left rib pixel point exists in the first preset neighborhood of the latest left rib pixel point, the iteration is stopped, and all the left rib pixel points are used as the initial left rib area of ​​the target image; Acquire the right rib pixel points of the target image from the right adjacent pixel points of the spine region; acquire new right rib pixel points in the second preset neighborhood of the right rib pixel points to obtain an initial right rib region; the method for acquiring the initial right rib region is similar to the method for acquiring the initial left rib region; The initial right rib region and the initial left rib region are used as initial rib regions in the target image.

3. According to claim 1, a hierarchical classification auxiliary management system for critical care specialist electronic medical record information is characterized in that: The line connecting the upper left pixel point and the upper right pixel point in the target rib region is taken as the rib main body line.

4. A hierarchical classification auxiliary management system for critical care specialist electronic medical record information according to claim 3, characterized in that: The method for obtaining the modified rib region comprises: In the target rib area, any pixel point is selected as the target pixel point; a perpendicular line to the rib main line is drawn through the left end point of the rib main line; the rib main line is translated along the perpendicular line through the target pixel point to obtain a reference line of the target pixel point; the number of pixel points from the target pixel point to the left end point of the reference line is taken as a reference number, and the number of pixel points belonging to the target rib area from the target pixel point to the left end point of the reference line is taken as an actual number; When the difference between the actual number of pixels and the reference number is greater than a preset difference threshold, the corresponding target pixel is removed from the target rib area, and after analyzing all the pixels in the target rib area, a corrected rib area is obtained.

5. According to claim 1, a hierarchical classification auxiliary management system for critical care electronic medical record information is characterized in that: The method for obtaining the flake-like high-density shadow area comprises: The area formed by the pixels in the target image that are outside the spine area and the corrected rib area and whose grayscale values ​​are greater than a preset segmentation threshold is used as a flake-like high-density shadow area.

6. A hierarchical classification auxiliary management system for critical care specialist electronic medical record information according to claim 1, characterized in that: The method for obtaining the pneumonia lesion index comprises: The ratio of the number of pixels in each of the patchy density-increased shadow regions to the gradient average of the edge pixels is used as the pathological factor of each of the patchy density-increased shadow regions; The product of the sum of the lesion factors of the patchy density increased shadow areas of the target image and the number of all the patchy density increased shadow areas is used as the pneumonia lesion index of the target image.

7. A hierarchical classification auxiliary management system for critical care specialist electronic medical record information according to claim 1, characterized in that: The method for preliminarily classifying the electronic medical record into a serious category or a mild category according to the text information comprises: The text information is matched with the preset classification keywords of the severe class and the mild class respectively by using the forward maximum matching method. When there is no matching relationship between the text information and the preset classification keywords of the severe class, the electronic medical record is preliminarily classified as the mild class; otherwise, it is classified as the severe class.

8. The hierarchical classification auxiliary management system for critical care electronic medical record information according to claim 1, characterized in that: The method for classifying the electronic medical record according to the temporal variation characteristics of the pneumonia lesion index of the chest X-ray of the patient and the distribution characteristics of the flake-like increased density shadow area includes: When the pneumonia lesion index of the chest X-ray of the patient constitutes a lesion index sequence in a time series order; when the elements in the lesion index sequence are monotonically decreasing, the recovery label is a normal recovery class; when the elements in the lesion index sequence are monotonically increasing, the recovery label is an abnormally severe class; when the elements in the lesion index sequence are decreasing, increasing, and then decreasing, the recovery label is a recovery-deterioration-recovery class; when the elements in the lesion index sequence are monotonically decreasing, monotonically increasing, and other than decreasing, increasing, and then decreasing, the recovery label is a to-be-observed class; When all the flake-like density-increased shadow areas are on the left side of the corresponding spine area, the position label is left lung pneumonia; when all the flake-like density-increased shadow areas are on the right side of the corresponding spine area, the position label is right lung pneumonia; when the flake-like density-increased shadow areas appear on both sides of the corresponding spine area, the position label is bilateral lung pneumonia; The combination of each recovery tag and each location tag is an electronic medical record classification.

9. A hierarchical classification auxiliary management system for critical care specialist electronic medical record information according to claim 7, characterized in that: The preset classification keywords of the severe category include at least: dyspnea, impaired consciousness, and renal insufficiency; the preset classification keywords of the mild category include at least: low fever, fatigue, and loss of appetite.

Images