An intelligent processing system and method for piecemeal reimbursement

Abstract

The present invention discloses a sporadic reimbursement intelligent processing system and method, belonging to the technical field of data processing. It includes a bill information acquisition unit for acquiring the text information on the medical bills corresponding to a preset patient. The bill information acquisition unit includes an image scanning module, an identification and editing module, a marking module, an auditing module, and an information storage module. The image scanning module is used to scan the corresponding medical bills and acquire scanned images and / or pictures. The identification and editing module is used to identify the text information of the acquired scanned images and / or pictures, edit the text information according to the identification results, and collect the feature data in the edited text information. The present invention acquires the feature data of the text information on the bills and the text information covered by the seal images on the bills according to the scanned medical bills, and calculates the loss function of the text information overlapping with the seal images through the gradient boosting tree algorithm, realizing the accurate auditing of medical bills and shortening the reimbursement cycle.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a sporadic reimbursement intelligent processing system and method thereof. Background Art

[0002] In the process of medical insurance reimbursement, if the insured fails to directly settle the medical expenses due to special reasons, the individual needs to pay in advance and then apply for reimbursement to the medical insurance agency according to regulations. This is the case of sporadic reimbursement. This situation usually includes expenses for seeking medical treatment at a designated medical institution outside the place of residence but not settled by swiping the card, expenses for seeking medical treatment at a non-designated medical institution due to sudden acute illness, and other medical expenses that meet the regulations.

[0003] Traditional medical insurance sporadic reimbursement adopts the method of offline review and online entry, that is, staff with medical experience need to manually mark and use a computer to calculate the reimbursement amount. This means that all information of the case needs to be manually entered at the stage of entering into the medical insurance system. Not only is the workload large, but also due to the difference between the hospital project name and the medical insurance catalog name, the reimbursement cycle is long and the review is inaccurate, which in turn leads to low satisfaction of the insured. Among them, when staff manually enter the text and data of medical bills, vouchers and / or other reimbursement expense proofs, the seal on the medical bills, vouchers and / or other reimbursement expense proofs often affects the accuracy of text and data recognition and entry. This not only leads to low efficiency of text and data entry, but also leads to incorrect payment of medical reimbursement amounts, thus affecting the supervision of medical reimbursement funds. Therefore, how to improve the efficiency of medical insurance sporadic reimbursement and the accuracy of review has become an urgent problem for medical institutions to solve. Summary of the Invention

[0004] In view of the above problems existing in the current technical field of data processing, the present invention is proposed.

[0005] Therefore, one of the purposes of the present invention is to provide a sporadic reimbursement intelligent processing system and method thereof. By scanning medical bills, it obtains the characteristic data of the text information on the bills and the text information covered by the seal images on the bills, calculates the loss function of the text information overlapping with the seal images through the gradient boosting tree algorithm, can calculate the relevant recognition difficulty according to the clarity of the text information at the overlapping position, and can also adjust the pixel parameters of the contour tomography according to the deep learning of the convolutional neural network to obtain a complete text information contour, and then completely obtain the relevant data, realizing the accurate review of medical bills and shortening the reimbursement cycle.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a sporadic reimbursement intelligent processing system, including:

[0008] A bill information acquisition unit for acquiring text information on medical bills corresponding to a preset patient. The bill information acquisition unit includes an image scanning module, an identification and editing module, a marking module, an auditing module, and an information storage module;

[0009] The image scanning module is used to scan the corresponding medical bill and acquire a scanned image and / or a picture;

[0010] The identification and editing module is used to identify text information from the acquired scanned image and / or picture, edit the text information according to the identification result, and collect feature data in the edited text information;

[0011] The marking module is used to mark the corresponding medical bill according to the acquired scanned image and / or picture to identify the scanned medical bill;

[0012] The auditing module responds to the feature data collected by the identification and editing module and is used to check the collected feature data against the bill data in the hospital inpatient department data management terminal. The bill data includes outpatient medical records, inpatient medical records, examination lists, laboratory test lists, and prescription lists;

[0013] The information storage module is used to store the image and / or picture acquired by the image scanning module and upload the stored image and / or picture to the corresponding medical reimbursement management platform;

[0014] A seal image acquisition module for acquiring the seal image on the corresponding medical bill, obtaining its shape and contour according to the acquired seal image, and simultaneously collecting the text information covered by the seal image;

[0015] A text information fusion processing unit, based on the collected text information covered by the seal image, for segmenting the text information and the seal image. Among them, the segmentation of the seal image is the main segmentation, and the text information under the seal image is the secondary segmentation. The number of segmented text information is counted in the secondary segmentation, the overlapping positions of each text information and the seal image are collected, and the text information corresponding to each overlapping position is marked as a loss function through the gradient boosting tree algorithm And through a regularization term Based on the clarity calculation of the text information corresponding to the overlapping position, the recognition difficulty of the text information is calculated as follows:

[0016] Where O represents the color saturation of the seal image overlapping with each text information;

[0017] In the formula, is the loss function used to measure the true value yi and the predicted value u i wherein the predicted value u i is the predicted value given for the color saturation, and the true value y i is the true value obtained for the color saturation of the seal image overlapping with each piece of the text information, and obtain the true value y i and the predicted value u i The squared error loss between them is as follows:

[0018]

[0019] is a regularization term used to control the complexity of the boosting tree and prevent overfitting. The complexity is given according to the color saturation, and the given formula is as follows:

[0020]

[0021] In the formula, r represents the number of leaf nodes of the boosting tree, w is the weight of each leaf node, and λ represents a hyperparameter;

[0022] wherein, the number of leaf nodes corresponds to the number of text information pieces statistically segmented in the sub-segmentation. The more the number of text information pieces, the greater the corresponding weight of the leaf node; the hyperparameter is given according to the adjacent text information interval data. When the text information interval data is small, the system determines that the recognition difficulty of the corresponding text information is large, and vice versa; the text information fusion processing unit includes a calculation module, an exposure module, and an identification warning module.

[0023] As a preferred solution of the present invention, wherein: in the recognition and editing module, the feature data includes total amount of money data, medical insurance co-ordination fund payment data, other payment data, personal account payment data, personal cash payment data, personal out-of-pocket payment data, and personal self-paid data.

[0024] As a preferred solution of the present invention, wherein: the calculation module includes a horizontal calculation module, a vertical calculation module, and an oblique calculation module. Among them, the horizontal calculation module is used to calculate the left and right side lengths of the seal image according to the contour of the seal image; the vertical calculation module is used to calculate the upper and lower side lengths of the seal image according to the contour of the seal image; the oblique calculation module is used to calculate the oblique length of the seal image other than the upper and lower side lengths and the left and right side lengths according to the contour of the seal image; and respectively obtain the number of text information pieces covered in the three length data of the left and right side lengths, the upper and lower side lengths, and the oblique length, and mark an exposure point for any covered text information in the seal image.

[0025] As a preferred embodiment of the present invention, it is provided that: the exposure module responds to the exposure points marked by the calculation module for any covered text information in the seal image, and is used to perform exposure processing on the exposure points. The exposure processing includes performing feedback control on the exposure points according to an automatic exposure algorithm, and presetting a target brightness tolerance threshold for scanning medical bills. The feedback control on the exposure points includes:

[0026] Performing brightness statistics on the currently received seal image;

[0027] Evaluating the current brightness to determine whether the current brightness reaches the target brightness tolerance threshold;

[0028] If the target brightness tolerance threshold is not reached, reset the exposure parameters and scan the medical bill until the current brightness reaches the target brightness tolerance threshold.

[0029] As a preferred embodiment of the present invention, it is provided that: when adjusting the exposure parameters, mark the parameter itself as the input value, mark the brightness of the obtained seal image as the initial threshold. Until the edge contour of the text information is obtained by adjusting the input value, mark the input value as the reference value. At the same time, use the input value and the initial threshold as both ends of an equation to obtain the variation law of the initial threshold with respect to the adjustment of the input value, as follows:

[0030]

[0031] In the formula, F represents the adjusted exposure parameter, t represents the duration of exposure parameter adjustment, L s represents the initial threshold, P represents the initial threshold that changes with the change of the exposure parameter, k represents the factor for adjusting both sides of the equation, and the factor includes the ambient light during the exposure of the seal image.

[0032] As a preferred embodiment of the present invention, it is provided that: the recognition and warning module is based on the reference value, and presets a warning threshold with the reference value. When adjusting the exposure parameters for the exposure points in the future, if the adjusted exposure parameter exceeds the warning threshold and the edge contour of the text information cannot be obtained, the system issues a warning and stops the exposure of the exposure points. Otherwise, no warning is issued.

[0033] As a preferred embodiment of the present invention, the method is as follows: Based on the lengths of the left and right sides and the lengths of the upper and lower sides of the seal image in the horizontal calculation module, text information on the left and right sides and / or the upper and lower sides within the length is collected. The collection method includes providing a piece of text information. When no other text information is collected on the left and right sides and / or the upper and lower sides of the text information, the system determines that the interval data of the text information is large; otherwise, it does not make a determination.

[0034] As a preferred embodiment of the present invention, the method is as follows: When there are contour breaks in the edge contour of the obtained text information, the contour breaks are marked, and an edge detection algorithm is used for the contour breaks to obtain contour data corresponding to the breaks. The edge detection algorithm includes detection using the Roberts operator and constructing a mixture Gaussian model to obtain the contour shape corresponding to the breaks. The acquisition method includes:

[0035] Initializing matrix parameters in the mixture Gaussian model, where the matrix parameters include pixel parameters;

[0036] Collecting changes in the contour shape according to changes in the matrix parameters;

[0037] When the contour shape is obtained, recording the pixel parameters at the time of acquisition and performing data update;

[0038] Matching the contour shape with the contour break of the corresponding text information. If they can be matched and overlapped, the system determines that the acquisition of the contour data is successful; otherwise, it does not make a determination and issues a warning.

[0039] As a preferred embodiment of the present invention, the method is as follows: When the contour data is successfully obtained, deep learning is performed on the pixel parameters in the mixture Gaussian model through a convolutional neural network to obtain the adjustment changes of the pixel parameters, marking the contour shape corresponding to the adjustment changes, and constructing a fault contour database. When the same fault contour is encountered in a future period, the contour shape is obtained according to the pixel parameters of the deep learning.

[0040] As a preferred embodiment of the present invention, the method is as follows: Obtaining the contour length from the obtained contour data, collecting the number of contour turning angles in the contour length, calculating the turning angle length data for each contour turning angle, dividing the turning angle length data into first data and second data, where the first data is the data from the starting data to the median data in the turning angle length data, and the second data is the data from the median data to the ending data. Analyzing the variation laws of the pixel parameters in the two groups of data and generating a data set.

[0041] As a preferred embodiment of the present invention, among them: in the said dataset, respectively intercept 3 data that are closest to the median data in the two groups of data, and use the 6 data as several evaluation indicators, calculate the weight of each evaluation indicator in the total data in the said dataset, when the weight ≥ 50%, then obtain the said profile data according to the said variation law in the future period, otherwise, make adjustments.

[0042] On the other hand, the present invention also provides a method applied to a sporadic reimbursement intelligent processing system, including the following steps:

[0043] Scan the medical bills of a preset patient, obtain the scanned images and / or pictures, perform text information recognition on the obtained scanned images and / or pictures, edit the text information according to the recognition results, and collect the feature data in the edited text information;

[0044] Mark the obtained scanned images and / or pictures to identify the scanned medical bills;

[0045] Compare the collected feature data with the bill data of the hospital inpatient department data management terminal, and the said bill data includes outpatient medical records, inpatient medical records, inspection lists, test lists, and prescription lists;

[0046] Store the images and / or pictures obtained by the said image scanning module, and upload the stored images and / or pictures to the corresponding medical reimbursement management platform;

[0047] Collect the seal images on the corresponding medical bills, obtain their shapes and contours according to the collected seal images, and at the same time collect the text information covered by the seal images;

[0048] Based on the collected text information covered by the seal image, it is used to segment the said text information and the said seal image, and its segmentation methods include:

[0049] The main segmentation is for the said seal image, and the secondary segmentation is for the text information under the said seal image;

[0050] Count the number of segmented text information in the said secondary segmentation, collect the overlapping positions of each text information and the said seal image, and mark the text information corresponding to each said overlapping position as the loss function L ▽ and through the regularization term Ω ▽ Calculate the recognition difficulty of the text information based on the clarity of the text information corresponding to the said overlapping position.

[0051] The present invention obtains the characteristic data of the text information on the medical bill and the text information covered by the seal image on the bill by scanning the medical bill. By performing primary and secondary segmentation on the seal image and the covered text information, and calculating the loss function of the text information overlapping with the seal image through the gradient boosting tree algorithm, the recognition difficulty can be calculated according to the clarity of the text information at the overlapping position, and the text information with high recognition difficulty is marked. Then, the marked points are exposed to obtain the edge contour of the text information. At the same time, when there is a contour break in the edge contour, the pixel parameters of the contour break can be adjusted according to the deep learning of the convolutional neural network to obtain a complete text information contour, and then the relevant data can be obtained completely to ensure the correct and error-free transmission of the reimbursement amount. This not only reduces the workload of the staff, but also can accurately review medical bills and shorten the reimbursement cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0053] Figure 1 It is a modular structure diagram of the piecemeal reimbursement intelligent processing system according to an embodiment of the present invention;

[0054] Figure 2 It is a modular structure diagram of the bill information acquisition unit according to an embodiment of the present invention;

[0055] Figure 3 It is a schematic flowchart of the method according to an embodiment of the present invention;

[0056] Reference numerals in the figures: 110 - bill information acquisition unit; 1101 - image scanning module; 1102 - recognition and editing module; 1103 - marking module; 1104 - auditing module; 1105 - information storage module; 120 - seal image acquisition module; 130 - text information fusion processing unit; 1301 - calculation module; 1302 - exposure module; 1303 - recognition and warning module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the embodiments of the present invention with reference to the drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present invention.

[0058] In the past, when staff manually entered the text and data of medical bills, vouchers, and / or other reimbursement expense proofs, the seals on the medical bills, vouchers, and / or other reimbursement expense proofs often affected the accuracy of text and data recognition and entry. This not only led to low efficiency in text and data entry but also resulted in incorrect disbursement of medical reimbursement amounts, thereby affecting the supervision of medical reimbursement funds. Therefore, how to improve the efficiency of sporadic medical insurance reimbursement and the accuracy of review has become an urgent problem for medical institutions to solve.

[0059] Based on this, the present invention proposes a sporadic reimbursement intelligent processing system and method. In this application, by scanning medical bills, the characteristic data of the text information on the bills and the text information covered by the seal images on the bills are obtained. The loss function of the text information overlapping with the seal images is calculated through the gradient boosting tree algorithm. The recognition difficulty can be calculated according to the clarity of the text information at the overlapping position, and the pixel parameters of the contour tomography can be adjusted through the deep learning of the convolutional neural network to obtain the complete text information contour, and then the relevant data can be completely obtained, realizing the accurate review of medical bills and shortening the reimbursement cycle.

[0060] The following further specifically describes this solution through embodiments in combination with the drawings.

[0061] Referring to Figures 1 to 3 , as an embodiment of the present invention, this embodiment provides a sporadic reimbursement intelligent processing system, including:

[0062] A bill information acquisition unit 110, configured to acquire the text information on the medical bills corresponding to a preset patient. The bill information acquisition unit 110 includes an image scanning module 1101, a recognition and editing module 1102, a marking module 1103, a review module 1104, and an information storage module 1105;

[0063] The image scanning module 1101 is configured to scan the corresponding medical bills and acquire the scanned images and / or pictures;

[0064] The image scanning module 1101 includes a scanning exposure machine;

[0065] The recognition and editing module 1102 is configured to recognize the text information of the acquired scanned images and / or pictures, edit the text information according to the recognition results, and collect the characteristic data in the edited text information;

[0066] It should be emphasized in this embodiment that the characteristic data includes total amount data, medical insurance co-ordination fund payment data, other payment data, personal account payment data, personal cash payment data, personal out-of-pocket data, and personal self-paid data;

[0067] The marking module 1103 is used to mark the corresponding medical bills according to the acquired scanned images and / or pictures to identify the scanned medical bills;

[0068] The auditing module 1104 responds to the feature data collected by the recognition and editing module 1102 and is used to check the collected feature data with the bill data of the hospital inpatient department data management terminal. The bill data includes outpatient medical records, inpatient medical records, inspection lists, test lists, and prescription lists;

[0069] The information storage module 1105 is used to store the images and / or pictures acquired by the image scanning module 1101 and upload the stored images and / or pictures to the corresponding medical reimbursement management platform;

[0070] The seal image acquisition module 120 is used to acquire the seal image on the corresponding medical bill, obtain its shape and contour according to the acquired seal image, and simultaneously acquire the text information covered by the seal image;

[0071] The text information fusion processing unit 130, based on the collected text information covered by the seal image, is used to segment the text information and the seal image. Among them, the segmentation of the seal image is the main segmentation, and the text information under the seal image is the secondary segmentation. The number of segmented text information is counted in the secondary segmentation, the overlapping positions of each text information and the seal image are acquired, and the text information corresponding to each overlapping position is marked as a loss function through the gradient boosting tree algorithm And through the regularization term Based on the clarity calculation of the text information corresponding to the overlapping position, the recognition difficulty of the text information is calculated as follows:

[0072] Among them, O represents the color saturation of the seal image overlapping with each text information;

[0073] In the formula, is the loss function, used to measure the true value y i and the predicted value u i The gap between them, where the predicted value u i is the predicted value given to the color saturation, and the true value y i is the true value obtained for the color saturation of the seal image overlapping with each text information, and the true value y i and the predicted value u i The squared error loss between them is as follows:

[0074]

[0075] is a regularization term used to control the complexity of the boosting tree and prevent overfitting. The complexity is given according to the color saturation, and the given formula is as follows:

[0076]

[0077] In the formula, r represents the number of leaf nodes of the boosting tree, w is the weight of each leaf node, and λ represents the hyperparameter;

[0078] Among them, the number of leaf nodes corresponds to the number of text information counted in the segmentation. The more the number of text information, the greater the weight of the corresponding leaf node; the hyperparameter is given according to the interval data of adjacent text information. When the interval data of text information is small, the system determines that the recognition difficulty of the corresponding text information is large, otherwise, it is not determined; the text information fusion processing unit 130 includes a calculation module 1301, an exposure module 1302, and an identification and warning module 1303;

[0079] Specifically in this embodiment, the calculation module 1301 includes a horizontal calculation module, a vertical calculation module, and an oblique calculation module. Among them, the horizontal calculation module is used to calculate the lengths of the left and right sides of the seal image according to the contour of the seal image; the vertical calculation module is used to calculate the lengths of the upper and lower sides of the seal image according to the contour of the seal image; the oblique calculation module is used to calculate the oblique length of the seal image other than the lengths of the upper and lower sides and the left and right sides according to the contour of the seal image; and the number of text information covered is obtained from the three length data of the left and right side lengths, the upper and lower side lengths, and the oblique length, and in the seal image, an exposure point is marked for any covered text information;

[0080] The exposure module 1302 responds to the exposure point marked by the calculation module 1301 for any covered text information in the seal image, and is used to perform exposure processing on the exposure point. The exposure processing includes performing feedback control on the exposure point according to the automatic exposure algorithm, and presetting the target brightness tolerance threshold for scanning medical bills, and the feedback control on the exposure point includes:

[0081] Performing brightness statistics on the currently received seal image;

[0082] Evaluating the current brightness to determine whether the current brightness reaches the target brightness tolerance threshold;

[0083] If the target brightness tolerance threshold is not reached, reset the exposure parameters and scan the medical bill until the current brightness reaches the target brightness tolerance threshold;

[0084] Among them, when adjusting the exposure parameters, the parameter itself is marked as the input value, and the brightness of the obtained seal image is marked as the initial threshold. Until the edge contour of the text information is obtained by adjusting the input value, the input value is marked as the reference value. At the same time, the input value and the initial threshold are used as both ends of the equation to obtain the variation law of the initial threshold with respect to the adjustment of the input value as follows:

[0085]

[0086] In the formula, F represents the adjusted exposure parameter, t represents the duration of the exposure parameter adjustment, L s represents the initial threshold, P represents the initial threshold that changes with the exposure parameter, and k represents the factor for adjusting both sides of the equation. The factor includes the ambient light during the exposure of the seal image;

[0087] Past experience in scanning including paper objects shows that for the obtained pictures, locally exposing them can ensure that every detail in the scanned pictures is clearly visible, make the colors of the pictures more vivid and real, and thus is conducive to obtaining the edge contour of the text information covered by the seal image;

[0088] The recognition and warning module 1303 is based on the reference value and preset the warning threshold with the reference value. When adjusting the exposure parameter for the exposure point in the future period, if the adjusted exposure parameter exceeds the warning threshold and the edge contour of the text information cannot be obtained, the system issues a warning and stops exposing the exposure point. Otherwise, there is no warning;

[0089] On this basis, in this embodiment, further, based on the lengths of the left and right sides of the seal image in the horizontal calculation module and the lengths of the upper and lower sides of the seal image in the vertical calculation module, the text information on the left and right sides and / or the upper and lower sides of the text information within the collected length is collected. The collection method includes giving a piece of text information. When no other text information is collected on the left and right sides and / or the upper and lower sides of the text information, the system determines that the interval data of the text information is large. Otherwise, it does not determine;

[0090] It should be noted in this embodiment that when there are contour breaks in the edge contour of the obtained text information, the contour breaks are marked, and the edge detection algorithm is used for the contour breaks to obtain the contour data corresponding to the breaks. The edge detection algorithm includes detecting and obtaining by using the Roberts operator, and constructing a mixture Gaussian model to obtain the contour shape corresponding to the breaks. The obtaining method includes:

[0091] Initializing the matrix parameters in the mixture Gaussian model, and the matrix parameters include pixel parameters;

[0092] Collecting the change of the contour shape according to the change of the matrix parameters;

[0093] When the contour shape is obtained, record the pixel parameters at the time of acquisition and update the data;

[0094] Match the contour shape with the contour section of the corresponding text information. If they can be matched and overlapped, the system determines that the acquisition of the contour data is successful; otherwise, it does not determine and issues a warning;

[0095] Further, based on the above, when the contour data is successfully acquired, in the Gaussian mixture model, perform deep learning on the pixel parameters through a convolutional neural network, obtain the adjustment changes of the pixel parameters, mark the contour shape corresponding to the adjustment changes, and construct a tomographic contour database. When encountering the same tomographic contour in the future period, obtain the contour shape according to the pixel parameters of the deep learning;

[0096] It should be determined in this embodiment that the contour length is obtained from the acquired contour data, the number of contour turning angles is collected from the contour length, and the turning angle length data is calculated for each contour turning angle. The turning angle length data is divided into first data and second data, where the first data is the data from the starting data to the median data in the turning angle length data, and the second data is the data from the median data to the end data. Analyze the change rules of the pixel parameters in the two groups of data and generate a data set;

[0097] Further, in the data set, respectively intercept 3 data in the two groups of data that are closest to the median data, and use the 6 data as several evaluation indicators. Calculate the weight of each evaluation indicator in the total data of the data set. When the weight ≥ 50%, the contour data is obtained according to the change rule in the future period; otherwise, adjustments are made.

[0098] Based on the above, this application obtains the characteristic data of the text information on the medical bill and the text information covered by the seal image on the bill by scanning the medical bill, calculates the loss function of the text information overlapping with the seal image through the gradient boosting tree algorithm, can calculate the relevant recognition difficulty according to the clarity of the text information at the overlapping position, and can also adjust the pixel parameters of the contour section through the deep learning of the convolutional neural network to obtain the complete text information contour, and then completely obtain the relevant data, realizing the accurate review of medical bills and shortening the reimbursement cycle.

[0099] Combined with the above sporadic reimbursement intelligent processing system, this embodiment also proposes the working method of the system as follows:

[0100] Scan the medical bills of the preset patients, obtain the scanned images and / or pictures, perform text information recognition on the obtained scanned images and / or pictures, edit the text information according to the recognition results, and collect the characteristic data in the edited text information;

[0101] Mark the acquired scanned images and / or pictures to identify the scanned medical bills;

[0102] Compare the collected feature data with the bill data in the data management terminal of the hospital inpatient department. The bill data includes outpatient medical records, inpatient medical records, examination lists, test lists, and prescription lists;

[0103] Store the images and / or pictures obtained by the image scanning module, and upload the stored images and / or pictures to the corresponding medical reimbursement management platform;

[0104] Collect the seal image on the corresponding medical bill, obtain its shape and contour based on the collected seal image, and at the same time collect the text information covered by the seal image;

[0105] Based on the collected text information covered by the seal image, it is used to segment the text information and the seal image. The segmentation methods include:

[0106] The main segmentation is for the seal image, and the secondary segmentation is for the text information under the seal image;

[0107] In the secondary segmentation, count the number of segmented text information, collect the overlapping positions of each text information and the seal image, and mark the text information corresponding to each overlapping position as the loss function L through the gradient boosting tree algorithm ▽ , and through the regularization term Ω ▽ Calculate the recognition difficulty of the text information based on the clarity of the text information corresponding to the overlapping position.

[0108] In summary, the present invention obtains the feature data of the text information on the bill and the text information covered by the seal image on the bill by scanning the medical bill. By performing primary and secondary segmentation on the seal image and the covered text information, and calculating the loss function of the text information overlapping with the seal image through the gradient boosting tree algorithm, the relevant recognition difficulty can be calculated according to the clarity of the text information at the overlapping position, and the text information with high recognition difficulty is marked. Then, the edge contour of the text information is obtained by exposing the marked points. At the same time, when there is a contour break in the edge contour, the pixel parameters of the contour break can be adjusted according to the deep learning of the convolutional neural network to obtain a complete text information contour, and then the relevant data can be obtained completely to ensure the correct and error-free sending of the reimbursement amount. This not only reduces the workload of the staff, but also can accurately review medical bills and shorten the reimbursement cycle.

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An intelligent processing system for sporadic reimbursement, characterized in that: include: A bill information acquisition unit, used to acquire text information on a medical bill corresponding to a preset patient, the bill information acquisition unit comprising an image scanning module, a recognition and editing module, a marking module, a review module and an information storage module; The image scanning module is used to scan the corresponding medical bill and obtain the scanned image and / or picture; The recognition and editing module is used to recognize text information on the acquired scanned image and / or picture, edit the text information according to the recognition result, and collect feature data in the edited text information; The marking module is used to mark the corresponding medical bill according to the acquired scanned image and / or picture, so as to identify the scanned medical bill; The audit module responds to the feature data collected by the identification and editing module, and is used to check the collected feature data with the bill data of the hospital inpatient department data management terminal, and the bill data includes outpatient medical records, inpatient medical records, examination sheets, test sheets and prescription sheets; The information storage module is used to store the images and / or pictures acquired by the image scanning module, and upload the stored images and / or pictures to the corresponding medical reimbursement management platform; A seal image acquisition module, used to acquire the seal image on the corresponding medical bill, and obtain its shape and outline according to the acquired seal image, and at the same time acquire text information covered by the seal image; The text information fusion processing unit is used to segment the text information and the seal image based on the collected text information covered by the seal image, wherein the segmentation of the seal image is the main segmentation, and the text information under the seal image is the secondary segmentation, and the number of segmented text information is counted in the secondary segmentation, and the overlapping position of each text information and the seal image is collected, and the text information corresponding to each overlapping position is marked as a loss function through a gradient boosting tree algorithm. And through the regularization term Ω ▽ The difficulty of recognizing the text information is calculated based on the clarity of the text information corresponding to the overlapping position, as shown below: Wherein, O represents the color saturation of the seal image overlapping with each of the text information; In the formula, is the loss function used to measure the true value y i and the predicted value u i The gap between the predicted value u i is the predicted value given for the color saturation, the true value y i The real value y is obtained for the color saturation of the seal image overlapping with each of the text information, and the real value y is obtained. i and the predicted value u i The squared error loss between , is as follows: is a regularization term used to control the complexity of the boosted tree to prevent overfitting. The complexity is given according to the color saturation. The given formula is as follows: In the formula, r represents the number of leaf nodes of the boosted tree, w is the weight of each leaf node, and λ represents the hyperparameter; Among them, the number of leaf nodes corresponds to the number of text information statistically segmented in the sub-segmentation. The more text information there is, the greater the corresponding leaf node weight is; the hyperparameter is given according to the adjacent text information interval data. When the text information interval data is small, the system determines that the recognition difficulty of the corresponding text information is large, otherwise, no determination is made; the text information fusion processing unit includes a calculation module, an exposure module and an identification and warning module.

2. The intelligent processing system for sporadic reimbursement as claimed in claim 1, characterized in that: In the identification and editing module, the characteristic data includes total amount data, medical insurance fund payment data, other payment data, personal account payment data, personal cash payment data, personal out-of-pocket payment data and personal self-funded data.

3. The intelligent processing system for sporadic reimbursement as claimed in claim 1, characterized in that: The calculation module includes a horizontal calculation module, a vertical calculation module and an oblique calculation module, wherein the horizontal calculation module is used to calculate the length of the left and right sides of the seal image according to the outline of the seal image; the vertical calculation module is used to calculate the length of the upper and lower sides of the seal image according to the outline of the seal image; the oblique calculation module is used to calculate the oblique length of the seal image other than the upper and lower sides and the left and right sides according to the outline of the seal image; and the number of covered text information is obtained from the three length data of the left and right sides length, the upper and lower sides length and the oblique length, and an exposure point mark is performed on any covered text information in the seal image.

4. The intelligent processing system for sporadic reimbursement as claimed in claim 3, characterized in that: The exposure module responds to the exposure point marked by the calculation module for any covered text information in the seal image, and is used to perform exposure processing on the exposure point. The exposure processing includes feedback control of the exposure point according to the automatic exposure algorithm, and presets the target brightness tolerance threshold for the medical bill scanning, and the feedback control of the exposure point includes: Perform brightness statistics on the currently received seal image; Evaluate the current brightness to determine whether the current brightness reaches the target brightness allowable threshold; If the target brightness tolerance threshold is not reached, the exposure parameters are reset and the medical document is scanned until the current brightness reaches the target brightness tolerance threshold.

5. The intelligent processing system for sporadic reimbursement as claimed in claim 4, characterized in that: When adjusting the exposure parameter, the parameter itself is marked as the input value, and the obtained seal image brightness is marked as the initial threshold value, until the adjusted input value obtains the edge contour of the text information, the input value is marked as the reference value, and the input value and the initial threshold value are used as the two ends of the equation to obtain the change rule of the initial threshold value to the input value adjustment, as follows: Where F represents the adjusted exposure parameter, t represents the duration of exposure parameter adjustment, and L s represents the initial threshold, P represents the initial threshold that changes with the exposure parameter, and k represents the factor for adjusting both sides of the equation, wherein the factor includes the ambient light when the seal image is exposed.

6. The intelligent processing system for sporadic reimbursement as claimed in claim 5, characterized in that: The identification and warning module is based on the reference value and presets a warning threshold with the reference value. When the exposure parameter of the exposure point is adjusted in the future period, if the adjusted exposure parameter exceeds the warning threshold and the edge contour of the text information cannot be obtained, the system issues a warning and stops exposing the exposure point. Otherwise, no warning is issued.

7. The intelligent processing system for sporadic reimbursement as claimed in claim 3, characterized in that: Based on the length of the left and right sides of the seal image in the horizontal calculation module and the length of the upper and lower sides of the seal image in the vertical calculation module, the text information on the left and right sides and / or the upper and lower sides of the text information within the length is collected. The collection method includes giving a text information. When no other text information is collected on the left and right sides and / or the upper and lower sides of the text information, the system determines that the text information interval data is large, otherwise, no determination is made.

8. The intelligent processing system for sporadic reimbursement as claimed in claim 5, characterized in that: When there is a contour fault in the edge contour of the acquired text information, the contour fault is marked, and an edge detection algorithm is used to obtain contour data of the corresponding fault. The edge detection algorithm includes using a Roberts operator for detection and acquisition, and constructing a mixed Gaussian model to obtain the contour shape of the corresponding fault. The acquisition method includes: Initializing matrix parameters in the Gaussian mixture model, the matrix parameters including pixel parameters; Collecting the change of the contour shape according to the change of the matrix parameters; When the contour shape is obtained, the pixel parameters at the time of obtaining are recorded, and data is updated; The contour shape is matched with the contour fault of the corresponding text information. If they match and overlap, the system determines that the contour data is successfully acquired. Otherwise, no determination is made and an early warning is issued.

9. The intelligent processing system for sporadic reimbursement as claimed in claim 8, characterized in that: When the contour data is successfully acquired, the pixel parameters are deeply learned through a convolutional neural network in the mixed Gaussian model to obtain the adjusted changes of the pixel parameters, and the contour shapes corresponding to the adjusted changes are marked, and a fault contour database is constructed. When the same fault contour is encountered in a future period, the contour shape is acquired according to the deeply learned pixel parameters.

10. The intelligent processing system for sporadic reimbursement as claimed in claim 8, characterized in that: The contour length is obtained from the acquired contour data, the number of contour corners is collected in the contour length, and the corner length data is calculated at each contour corner, and the corner length data is divided into first data and second data, wherein the first data is the starting data to the median data in the corner length data, and the second data is the median data to the ending data, the changing rules of the pixel parameters in the two sets of data are analyzed, and a data set is generated.

11. The intelligent processing system for sporadic reimbursement as claimed in claim 10, characterized in that: In the data set, three data closest to the median data in the two groups of data are respectively intercepted, and the six data are divided into several evaluation indicators, and the weight of each evaluation indicator in the total data in the data set is calculated. When the weight is ≥50%, the profile data is obtained in the future time period according to the change law, otherwise, it is adjusted.

12. A method applied to a sporadic reimbursement intelligent processing system as claimed in claim 1, characterized in that: The following steps are involved: Scanning the medical bills of a preset patient and obtaining scanned images and / or pictures, performing text information recognition on the obtained scanned images and / or pictures, editing the text information according to the recognition results, and collecting feature data in the edited text information; Marking the acquired scanned images and / or pictures to identify the medical bills that have been scanned; The collected characteristic data is compared with the bill data of the hospital inpatient department data management terminal, wherein the bill data includes outpatient medical records, inpatient medical records, examination sheets, test sheets and prescription sheets; Storing the images and / or pictures acquired by the image scanning module, and uploading the stored images and / or pictures to a corresponding medical reimbursement management platform; Collecting the seal image on the corresponding medical bill, and acquiring its shape and outline according to the collected seal image, and collecting text information covered by the seal image; Based on the collected text information covered by the seal image, the text information and the seal image are segmented, and the segmentation method includes: The segmentation of the seal image is primary segmentation, and the segmentation of the text information under the seal image is secondary segmentation; In the sub-segmentation, the number of segmented text information is counted, the overlapping position of each text information and the seal image is collected, and the text information corresponding to each overlapping position is marked as a loss function L by a gradient boosting tree algorithm. ▽ , and through the regularization term Ω ▽ The recognition difficulty of the text information is calculated based on the clarity of the text information corresponding to the overlapping position.

Images