CN111583275A

CN111583275A - Method, system, device and storage medium for identifying pathological number of section

Info

Publication number: CN111583275A
Application number: CN202010364532.9A
Authority: CN
Inventors: 刘炳宪; 谢菊元; 桂坤; 操家庆; 龙希
Original assignee: Hangzhou Zhituan Information Technology Co ltd
Current assignee: Hangzhou Zhituan Information Technology Co ltd
Priority date: 2020-04-30
Filing date: 2020-04-30
Publication date: 2020-08-25

Abstract

The invention relates to a method, a system, a device and a storage medium for identifying a slice pathology number, wherein the method comprises the steps of acquiring a pathology slice image containing a plurality of pathology slices; sequentially carrying out region segmentation and region filtering on the pathological section image to obtain a pathological label image set in the pathological section image; and identifying the pathology label image set to obtain and output a pathology number corresponding to each pathology label image in the pathology label image set. The invention can automatically identify and input the pathological number on the pathological section, effectively avoids the troubles of naked eye identification and manual input, greatly saves the labor cost, effectively overcomes the error proneness of manual identification and manual input, greatly improves the identification accuracy and the input efficiency of the pathological number, is beneficial to the management of a section management system, is also beneficial to a doctor to look up the pathological section of a specified patient, and improves the working efficiency of the doctor.

Description

Method, system, device and storage medium for identifying pathological number of section

Technical Field

The invention relates to the technical field of slice management, in particular to a method, a system, a device and a storage medium for identifying a slice pathology number.

Background

Judging whether the corresponding tissue and organ of the human body has pathological changes or not by observing the slice tissues of the human body is one of the common and important examination means in modern medicine, so that in order to facilitate the pathological diagnosis of a doctor on a patient, the patient needs to take slice tissues with a certain size and then prepare pathological slices by a histopathology method.

With the development of medical science and technology, each hospital generates a large number of pathological sections every day, the pathological sections are contained in a section box, and each pathological section is marked with a corresponding pathological number. The corresponding pathological number on each pathological section is input into the section system, which has important significance for the management of the pathological section by the hospital and the consultation of the doctor for the pathological section of the appointed patient.

At present, the pathological number of a pathological section is basically identified by naked eyes and depends on a manual mode of inputting a section system, time and labor are wasted, labor cost is greatly wasted, mistakes are easy to make, and the identification accuracy is low.

Disclosure of Invention

The invention aims to solve the technical problems of the prior art and provides a method, a system, a device and a storage medium for identifying a pathological number of a section, which can automatically identify the pathological number on the pathological section and automatically input the pathological number, effectively avoid the troubles of naked eye identification and manual input, save the labor cost and improve the identification accuracy and the input efficiency of the pathological number.

The technical scheme for solving the technical problems is as follows:

a method for identifying a section pathology number comprises the following steps:

step 1: acquiring a pathological section image comprising a plurality of pathological sections;

step 2: sequentially carrying out region segmentation and region filtering on the pathological section image to obtain a pathological label image set in the pathological section image;

and step 3: and identifying the pathology label image set to obtain and output a pathology number corresponding to each pathology label image in the pathology label image set.

According to another aspect of the present invention, there is also provided a system for identifying a pathological section number, which is applied to the method for identifying a pathological section number in the present invention, and comprises an image acquisition module, a region processing module, an identification module, and an output module;

the image acquisition module is used for acquiring a pathological section image containing a plurality of pathological sections;

the region processing module is used for sequentially carrying out region segmentation and region filtering on the pathological section images to obtain a pathological label image set in the pathological section images;

the identification module is used for identifying the pathological label image set to obtain a pathological number corresponding to each pathological label image in the pathological label image set one by one;

and the output module is used for outputting the pathological numbers corresponding to the pathological label images one by one.

According to another aspect of the present invention, there is provided a slice pathology number identification apparatus, comprising a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the computer program realizes the steps of a slice pathology number identification method of the present invention when running.

In accordance with another aspect of the present invention, there is provided a computer storage medium comprising: at least one instruction which, when executed, implements a step in a method of identifying a slice pathology number of the invention.

The method, the system, the device and the storage medium for identifying the slice pathology number have the advantages that: the acquired pathological section images containing a plurality of pathological sections are subjected to region segmentation, so that subsequent filtering is convenient according to the images subjected to region segmentation, the images subjected to region segmentation are subjected to region filtering, images which are completely irrelevant to the images containing pathological number information (namely pathological label images) are convenient to screen out, pathological label images which are in one-to-one correspondence with each pathological section are screened out, a pathological label image set is obtained, the pathological label image set is convenient to be identified subsequently, pathological numbers which are in one-to-one correspondence with each pathological label image are obtained, the pathological number of each pathological section is obtained, and finally the obtained pathological numbers are automatically output one by one; the invention can automatically identify and input the pathological number on the pathological section, effectively avoids the troubles of naked eye identification and manual input, greatly saves the labor cost, effectively overcomes the error proneness of manual identification and manual input, greatly improves the identification accuracy and the input efficiency of the pathological number, is beneficial to the management of a section management system, is also beneficial to a doctor to look up the pathological section of a specified patient, and improves the working efficiency of the doctor.

Drawings

Fig. 1 is a schematic flowchart of a method for identifying a pathological section number according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a region segmentation process according to an embodiment of the present invention;

FIG. 3 is a schematic flow chart of region filtering according to an embodiment of the present invention;

fig. 4 is a schematic flowchart illustrating a process of identifying a pathology label image according to a first embodiment of the present invention;

FIG. 5 is a schematic view of a complete flow chart of a slice pathology number identification method according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a system for identifying a pathological section number according to a second embodiment of the present invention;

fig. 7 is a schematic structural diagram of a region processing module and an identification module in the second embodiment of the present invention.

Detailed Description

The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.

The present invention will be described with reference to the accompanying drawings.

In a first embodiment, as shown in fig. 1, a method for identifying a pathology number of a slice includes the following steps:

s1: acquiring a pathological section image comprising a plurality of pathological sections;

s2: sequentially carrying out region segmentation and region filtering on the pathological section image to obtain a pathological label image set in the pathological section image;

s3: and identifying the pathology label image set to obtain and output a pathology number corresponding to each pathology label image in the pathology label image set.

The acquired pathological section images containing a plurality of pathological sections are subjected to region segmentation, so that subsequent filtering is convenient according to the images subjected to region segmentation, the images subjected to region segmentation are subjected to region filtering, images which are completely irrelevant to the images containing pathological number information (namely pathological label images) are convenient to screen out, pathological label images which are in one-to-one correspondence with each pathological section are screened out, a pathological label image set is obtained, the pathological label image set is convenient to be identified subsequently, pathological numbers which are in one-to-one correspondence with each pathological label image are obtained, the pathological number of each pathological section is obtained, and finally the obtained pathological numbers are automatically output one by one; this embodiment can automatic identification go out the pathology number on the pathological section and carry out automatic input, effectively avoided the trouble of naked eye discernment and manual input, saved the human cost greatly to effectively overcome the easy mistake nature of manual identification and manual input, improved the discernment rate of accuracy and the input efficiency of pathology number greatly, existing the management that is favorable to section management system, be favorable to the doctor to look up appointed patient's pathological section again, improved doctor's work efficiency.

Specifically, in this embodiment S1, the image of 4618 × 3464 high-resolution pathological section is obtained by using the fertile farmland high-speed scanner with the auto-focus function S1282AF, and the barcode (barcode or two-dimensional code) and the pathological number text corresponding to a plurality of pathological sections can be clearly seen in the image, and meanwhile, in order to reduce the interference of external light, a light source may be installed on the apparatus, so as to improve the quality of the photographed pathological section image.

Preferably, as shown in fig. 2, in S2, the specific step of performing region segmentation on the pathological section image includes:

s21: according to a first preset zooming size, carrying out primary zooming processing on the pathological section image to obtain a primary zoomed section image;

s22: and carrying out image segmentation on the primarily zoomed slice image by adopting an image segmentation method based on a graph to obtain a plurality of slice region segmentation images.

Through carrying out primary zoom to pathology section image, subsequent image segmentation of being convenient for improves image segmentation speed, can improve the recognition efficiency of whole pathology number on the one hand, and subsequent regional segmentation of being convenient for on the other hand to obtain a plurality of pathology label images of better quality. The first preset scaling size can be selected and adjusted according to actual conditions.

Specifically, in the first scaling process in this embodiment S21, scaling is performed according to a method in which the aspect ratio of the image is not changed, and the first preset scaling size specifically includes: length 924 and width 693, i.e., the length W of the first scaled slice image_img924 wide H_img693; the units are pixels.

Specifically, in this embodiment S22, the parameters in the Graph-based image Segmentation method (also called Graph Segmentation algorithm) are: the gaussian kernel σ of the filter is 0.2, the minimum area min _ area is 80, and the minimum class K is 2; the specific operation steps of the Graph Segmentation algorithm are the prior art, and the details are not described herein.

Preferably, as shown in fig. 3, in S2, the specific step of performing region filtering on the pathological section image includes:

s23: respectively carrying out corrosion expansion treatment on each slice region segmentation image to obtain slice region preprocessing images corresponding to each slice region segmentation image one to one;

s24: acquiring the length and width of an external rectangle of each slice area preprocessed image, and primarily filtering the length and width of all external rectangles according to a preset first filtering discriminant to obtain a plurality of slice area primary filtering images;

the first filtering discriminant is specifically:

wherein the content of the first and second substances,

and

length and width of circumscribed rectangle, n, respectively, for preprocessed image of ith slice region_wAnd n_hRespectively, a transverse accommodation quantity of pathological sections accommodated in the transverse direction of the section box and a longitudinal accommodation quantity of pathological sections accommodated in the longitudinal direction, W_imgAnd H_imgRespectively the length and width of the primary scaled slice image;

s25: carrying out region merging on the primary filtered images of all the slice regions to obtain a plurality of slice region merged images;

s26: acquiring the length and width of a circumscribed rectangle of each slice region merged image, and filtering all slice region merged images again according to a preset second filtering discrimination formula to obtain a plurality of slice region re-filtered images;

the second filtering discriminant is specifically:

wherein the content of the first and second substances,

and

the length and width of a circumscribed rectangle of the merged image for the jth slice region respectively;

s27: according to a second preset scaling size, re-scaling each re-filtered image of each slice area respectively to obtain a slice area scaled image corresponding to each re-filtered image of each slice area one by one;

s28: extracting gradient direction histogram characteristics corresponding to the zoomed images of each slice region one by one, and clustering the zoomed images of each slice region according to the gradient direction histogram characteristics by adopting a DBSCAN density clustering method to obtain a plurality of clustering categories and slice region zoomed image sets and image quantities under each clustering category; and determining a slice region scaling image set under the cluster category corresponding to the maximum value in all the image numbers as the pathology label image set.

By respectively carrying out corrosion expansion processing on the segmentation images of each slice region, the interference of noise can be reduced, the image quality is improved, the length and the width of a circumscribed rectangle of the preprocessed image of each slice region after the corrosion expansion processing are conveniently and accurately obtained, and further the preprocessed images of the slice regions are conveniently subjected to primary filtering; because the maximum region and the minimum region in the slice region preprocessed images generally refer to regions, such as edges of the slice box, of the slice box which are irrelevant to the pathological section labels, the maximum region and the minimum region in the slice region preprocessed images can be filtered according to the first filtering discriminant, so that images containing label information of pathological sections can be primarily screened, and the primarily filtered images of a plurality of slice regions can be obtained; because the pathological section label is provided with the pathological number text information, the bar code and other information, the label is easily divided into two different areas, and therefore the pathological number text information, the bar code and other information on the same pathological section label can be combined together through area combination to obtain a plurality of combined images of the section areas; because the smaller area in the combined images of the slice areas is not necessarily the pathological section label, the smaller area in the combined images of the slice areas after combination can be filtered according to the second filtering discrimination, and the identification effect of the pathological section label is further improved; because the merged images of the plurality of slice regions have different specifications, if the Histogram feature of the gradient direction (i.e. Histogram of OrientedGradient feature, HOG feature for short) is directly extracted, the feature dimensions of the merged images of each slice region are different, which is not beneficial to subsequent clustering and identification, and through the rescaling treatment again, the feature dimensions of the merged images of each slice region can be unified into one specification, which is convenient for subsequent clustering and identification; by extracting HOG characteristics, the Clustering effect can be improved, and finally, a DBSCAN Density Clustering method (namely, a sensitivity-Based Spatial Clustering of Applications with Noise algorithm) is adopted, so that more accurate pathological label images which correspond to each pathological section one by one can be identified, and the pathological label images form a pathological label image set.

Wherein, the Histogram of Oriented Gradient (HOG) feature is a feature descriptor used for object detection in computer vision and image processing, and the HOG feature forms a feature by calculating and counting the Histogram of the Gradient direction of a local area of an image; the feature description method can keep good invariance to the deformation of image geometry and optics; the DBSCAN Density Clustering method (namely, Density-Based Spatial Clustering of applications with Noise algorithm) is a Density-Based Clustering algorithm, and generally, the class can be determined by the compactness degree of sample distribution, and closely connected samples can be classified into one class; the clustering method can be used for clustering dense data sets in any shapes, abnormal points can be found while clustering, the abnormal points in the data sets are insensitive, and clustering results are not biased. It should be noted that, the specific operation steps of the HOG feature extraction and DBSCAN density clustering method are the prior art, and the specific details are not described herein again.

Specifically, in the present embodiment S24, the lateral accommodation number n of pathological sections accommodated in the lateral direction by the section box_w10, the number n of pathological sections accommodated in the longitudinal direction in the section cassette_h2; the second preset zoom size may be selected and adjusted according to actual situations, in this embodiment S27, the second preset zoom size is 64 × 64, that is, the size of the zoomed image of each slice region is 64 × 64; in the HOG feature extraction process in this embodiment S28, the size of the sliding window and the block of the HOG feature extractor are 64 × 64, the size of each cell and the sliding size in the block are 16 × 16, and each cell outputs 18 feature histograms, that is, the HOG feature dimensions are: (64/64 × 64/64) (64/16 × 64/16) × 18 ═ 288, and each slice region scaled image is a 288-dimensional dot.

Preferably, S25 specifically includes:

s251: optionally selecting two adjacent slice region primary filtering images in all slice region primary filtering images, and respectively calculating the image areas of the two selected slice region primary filtering images and the overlapping area and the similarity between the two selected slice region primary filtering images according to the length and the width of the circumscribed rectangle of the two selected slice region primary filtering images;

the specific formula for calculating the similarity between the k slice region primary filtered image and the k +1 slice region primary filtered image is as follows:

similarity_kk+1＝min(|w_k+1-w_k|,|h_k+1-h_k|)；

wherein, similarity_kk+1For the similarity between the first filtered image of the k-th slice region and the first filtered image of the k + 1-th slice region, w_kAnd h_kThe length and width, w, of the circumscribed rectangle of the primary filtered image for the kth slice region, respectively_k+1And h_k+1The length and the width of a circumscribed rectangle of the primary filtered image of the (k + 1) th slice region are respectively;

s252: respectively calculating the overlapping degree and the coverage degree between the selected two slice region primary filtering images according to the image areas of the selected two slice region primary filtering images and the overlapping area between the selected two slice region primary filtering images;

the specific formulas for calculating the overlapping degree and the coverage degree between the k slice region primary filtered image and the k +1 slice region primary filtered image are respectively as follows:

wherein the IOA_kk+1And IOU_kk+1Overlap and coverage, area, between the k-th slice region primary filtered image and the k + 1-th slice region primary filtered image, respectively_kAnd area_k+1Respectively setting the image area of the primary filtered image of the k slice region and the image area of the primary filtered image of the k +1 slice region, overlap_kk+1An overlapping area between the k slice region primary filtered image and the k +1 slice region primary filtered image;

s253: judging whether the overlapping degree, the coverage degree and the similarity between the two selected slice region primary filtering images meet a preset merging judgment formula, if so, merging the two selected slice region primary filtering images to obtain a corresponding slice region merging image, otherwise, returning to S251 until every two adjacent slice region primary filtering images in all the slice region primary filtering images are traversed;

the merging discriminant is specifically:

s254: in the method of S251 to S253, a plurality of slice region merged images are obtained.

Because two adjacent slice regions primarily filter images which are possibly two different region images divided by information such as pathological number text information and bar codes on pathological section labels, the method for calculating the overlapping degree, the coverage degree and the similarity between any two adjacent slice regions primarily filter images substitutes the overlapping degree, the coverage degree and the similarity into a merging discriminant, so that the pathological number text information and the bar codes on the same pathological section label can be accurately merged together for subsequent clustering and identification, and the pathological label image with high identification accuracy is obtained.

Preferably, as shown in fig. 4, S3 specifically includes:

s31: optionally selecting one pathological label image in the pathological label image set, performing bar code identification on the selected pathological label image, judging whether the bar code exists in the selected pathological label image, if so, sequentially executing S32 and S34, and if not, sequentially executing S33 and S34;

s32: judging whether the bar code is a two-dimensional code or not, if so, analyzing the bar code by the two-dimensional code to obtain and output a pathology number corresponding to one selected pathology label image, and if not, searching a preset pathology number database according to the bar code to obtain and output a pathology number corresponding to one selected pathology label image;

s33: performing character recognition on the selected pathological label image by using an OCR character recognition method to obtain and output a pathological number corresponding to the selected pathological label image;

s34: and traversing each pathological label image in the pathological label image set to obtain and output a pathological number corresponding to each pathological label image one by one.

Because the pathological number usually appears in the form of a one-dimensional bar code or a two-dimensional code and/or a text in each pathological section label, wherein the one-dimensional bar code and the two-dimensional code are commonly referred to as bar codes, firstly, the bar codes of any pathological label image are identified, whether the bar codes exist in the pathological label image is judged, if yes, the pathological label image is identified according to the identification method corresponding to the bar codes, and if not, the pathological label image is identified according to the identification method corresponding to the text identification; in the process of identifying the bar code, if the bar code is a two-dimensional bar code, the two-dimensional bar code is directly analyzed to obtain a corresponding pathological number and output the pathological number, and if the bar code is a one-dimensional bar code, a preset pathological number database is inquired, and the pathological number corresponding to the one-dimensional bar code is inquired and output; the method can automatically select a proper identification method according to the actual situation, automatically identify the corresponding pathological number and automatically output the pathological number, has high identification efficiency and identification accuracy, and effectively avoids the troubles of traditional naked eye identification and manual input.

Specifically, in this embodiment S33, an OCR character recognition method (i.e., an Optical character recognition method) is a method for converting a print character into an image of a black-and-white dot matrix by an Optical method and then recognizing the image of the black-and-white dot matrix; the specific operation steps are as follows:

(1) adopting a ctpn network to identify a text region in the pathological label image, and carrying out noise reduction processing and binarization processing on the text region;

(2) acquiring a region with the largest aspect ratio in the processed text region, namely a pathological number region;

(3) identifying the content of the regional pathological number of the pathological number by using crnn;

(4) and (4) after the identified pathological number is manually rechecked, obtaining the final pathological number and outputting the final pathological number.

The pathological number on the pathological section label without the bar code can be accurately identified by an OCR character identification method, wherein the identification accuracy can be further improved by manually rechecking the identified pathological number.

Specifically, the method further includes, before the step S32 of this embodiment, the following steps:

and establishing the pathology number database in advance.

By establishing the pathological number database in advance, the corresponding pathological number can be conveniently and directly inquired according to the one-dimensional bar code, the method is convenient and quick, and the management of a slicing system is facilitated. The pathological number database stores the one-dimensional bar code and pathological number one-to-one correspondence.

Preferably, in S32, the pathology number database stores a one-to-one correspondence table between the barcode and the pathology number;

then in S32, before searching the preset pathology number database according to the barcode, the method further includes:

and judging whether the relation table is acquired or not, if so, inquiring the relation table according to the bar code to acquire and output a pathology number corresponding to the selected pathology label image, otherwise, removing the bar code in the selected pathology label image, and executing S33.

When the pathological number database (namely the relation table) is inquired for the one-dimensional bar code, the condition that the pathological number database is not acquired or is lost may occur, so that by judging whether the relation table is acquired or not, if the relation table is acquired, the relation table is directly inquired to obtain the corresponding pathological number for outputting, and if the relation table is not acquired, the one-dimensional bar code in the pathological label image is removed, and then the pathological number is identified according to a character identification method; the method can ensure the success rate of pathological number identification and further effectively ensure the identification effect.

Preferably, the specific step of removing the barcode in the selected pathological label image includes:

sequentially carrying out graying processing and reverse binarization processing on the selected pathological label image to obtain a foreground image and a background image corresponding to the selected pathological label image;

performing corrosion expansion processing on the foreground image corresponding to the selected pathological label image to obtain a foreground processing image corresponding to the selected pathological label image;

extracting a bar code region image corresponding to the selected pathological label image from the foreground processing image corresponding to the selected pathological label image according to a preset extraction length-width ratio threshold;

and assigning the bar code region image corresponding to the selected pathological label image according to the pixel value of the background image corresponding to the selected pathological label image, and finishing bar code removal.

Through graying and reverse binarization processing, a foreground image (including a bar code region image and a text region image) in the pathological label image can be distinguished from a background region, so that the foreground image can be conveniently processed subsequently; because the length-width ratio of the one-dimensional bar code is usually in a certain specification, the length-width ratio of the foreground region is calculated, the bar code region image in the foreground image can be extracted according to the preset extraction length-width ratio, and then the extracted bar code region image is assigned according to the pixel value of the background image, so that the bar code region image can be treated as the background region, and the purpose of removing the bar code region image is achieved; the step of removing the bar code can effectively improve the accuracy of subsequent character recognition.

Specifically, in this embodiment, after performing graying processing on a selected pathological label image, a grayscale image is obtained, then in the reverse binarization processing process, a pixel threshold is set, an area lower than the pixel threshold in the grayscale is determined as a foreground area, an area higher than the pixel threshold is determined as a background area, and then the grayscale value of the background area is assigned to 0, so as to obtain a black background image; wherein the pixel threshold may be set to 0.6 of the pixel mean of the grayscale image.

Specifically, in the process of performing erosion dilation processing on the foreground image corresponding to the selected pathological label image, the size of the erosion dilation kernel is set to be 3 × 3.

Specifically, in the process of extracting the barcode region image corresponding to the selected pathological label image, the embodiment first calculates the aspect ratio of each sub-region in the foreground image, and then sets the preset extraction aspect ratio threshold to x<0.2 or x>5(x is the length-width ratio of the subareas), two types of subareas are extracted and are respectively marked as Area₉₀And Area₀(ii) a Because the one-dimensional bar codes on the pathological sections are all pasted manually when the pathological sections are manufactured, most of the one-dimensional bar codes are pasted horizontally or vertically, x is<A sub-region of 0.2 may correspond to a vertical barcode, and the image corresponding to the sub-region may be a barcode region image, x>5, the sub-region may correspond to a transverse bar code, and the image corresponding to the sub-region may also be a bar code region image; then, the center coordinates of the foreground images are taken as feature descriptors, and the DBSCAN density clustering method is adopted again to perform clustering on the areas₉₀And Area₀These two types of subregions are carried outClustering, counting the number of the two types of sub-regions, determining the one type of sub-region with a larger number as a bar code region, wherein the corresponding image is the bar code region image; and finally, assigning the gray value of the image of the barcode region as 0, namely finishing the removal of the barcode, and then performing character recognition on the image after the removal of the barcode by adopting an OCR character recognition method to obtain a corresponding pathological number and outputting the pathological number.

Specifically, in a preferred embodiment of the present embodiment, the complete flow of slice pathology number identification is shown in fig. 5.

In a second embodiment, as shown in fig. 6, a system for identifying a pathological section number is applied to the method for identifying a pathological section number in the first embodiment, and includes an image acquisition module, a region processing module, an identification module, and an output module;

The system for identifying a slice pathology number according to the embodiment performs region segmentation on a pathology slice image containing a plurality of pathology slices acquired by an image acquisition module through a region processing module, facilitates subsequent filtering according to the region segmented image through the region processing module, and performs region filtering on the region segmented image, thereby facilitating screening of an image completely unrelated to an image (namely a pathology label image) containing pathology number information, screening out pathology label images corresponding to each pathology slice one by one to obtain a pathology label image set, facilitating subsequent identification of the pathology label image set by the identification module to obtain a pathology number corresponding to each pathology label image one by one, namely obtaining the pathology number of each pathology slice, and finally automatically outputting the obtained pathology numbers one by one through an output module; this embodiment can automatic identification go out the pathology number on the pathological section and carry out automatic input, effectively avoided the trouble of naked eye discernment and manual input, saved the human cost greatly to effectively overcome the easy mistake nature of manual identification and manual input, improved the discernment rate of accuracy and the input efficiency of pathology number greatly, existing the management that is favorable to section management system, be favorable to the doctor to look up appointed patient's pathological section again, improved doctor's work efficiency.

Preferably, as shown in fig. 7, the region processing module includes a first scaling unit and an image segmentation unit;

the first zooming unit is used for carrying out primary zooming processing on the pathological section image according to a first preset zooming size to obtain a primary zoomed section image;

and the image segmentation unit is used for carrying out image segmentation on the primary scaled slice image by adopting an image segmentation method based on a graph to obtain a plurality of slice region segmentation images.

Carry out the primary zoom through first zooming unit to pathological section image and handle, can be convenient for subsequent image segmentation, improve the image and cut apart the speed, can improve the recognition efficiency of whole pathology number on the one hand, on the other hand can be convenient for subsequent image segmentation unit's region segmentation to obtain a plurality of pathology label images of better quality.

Preferably, as shown in fig. 7, the region processing module further includes a corrosion expansion unit, a primary filtering unit, a region merging unit, a secondary filtering unit, a second scaling unit, a feature extraction unit, and a clustering unit;

the corrosion expansion unit is used for respectively carrying out corrosion expansion processing on each slice region segmentation image to obtain slice region preprocessing images corresponding to each slice region segmentation image one to one;

the primary filtering unit is used for acquiring the length and the width of an external rectangle of each slice area preprocessed image, and performing primary filtering on the length and the width of all the external rectangles according to a preset first filtering discriminant to obtain a plurality of slice area primary filtered images;

the region merging unit is used for performing region merging on the primary filtered images of all the slice regions to obtain a plurality of slice region merged images;

the secondary filtering unit is used for acquiring the length and the width of a circumscribed rectangle of the combined image of each slice region, and filtering the combined image of all the slice regions again according to a preset second filtering discriminant to obtain a plurality of secondary filtered images of the slice regions;

the second zooming unit is used for respectively carrying out zooming processing on the re-filtered image of each slice region again according to a second preset zooming size to obtain a zoomed image of the slice region corresponding to the re-filtered image of each slice region one by one;

the feature extraction unit is used for extracting gradient direction histogram features corresponding to the zoomed images of each slice region one by one;

the clustering unit is used for clustering the scaled images of each slice region according to the histogram characteristics of each gradient direction by adopting a DBSCAN density clustering method to obtain a plurality of clustering categories and slice region scaled image sets and image quantities under each clustering category; and determining a slice region scaling image set under the cluster category corresponding to the maximum value in all the image numbers as the pathology label image set.

The corrosion expansion unit can reduce the noise interference, improve the image quality, and facilitate the accurate acquisition of the length and width of the external rectangle of the preprocessed image of each slice region after corrosion expansion treatment, thereby facilitating the primary filtration of the preprocessed image of the slice regions; the initial filtering unit can filter the maximum area and the minimum area in the preprocessed images of the slice areas according to the first filtering discriminant so as to initially screen out images containing label information of pathological slices, and the initial filtered images of the slice areas are obtained; through the region merging unit, pathological number text information, bar codes and other information on the same pathological section label can be merged together to obtain a plurality of section region merged images; the smaller area in the combined image of the combined slice area can be filtered by the secondary filtering unit according to the second filtering discrimination, so that the identification effect on the pathological slice label is further improved; the feature dimensions of the merged image of each slice region can be unified into a specification through the second zooming unit for zooming again, so that subsequent clustering and identification are facilitated; the HOG features are extracted by the feature extraction unit, the clustering effect can be improved, and finally, more accurate pathological label images which correspond to each pathological section one by one can be identified by the clustering unit by adopting a DBSCAN density clustering method, and the pathological label images form a pathological label image set.

Preferably, as shown in fig. 7, the identification module includes a barcode identification unit, a first determination unit, a character identification unit, a second determination unit, a two-dimensional code analysis unit, and a query unit;

the bar code identification unit is used for selecting one pathological label image in the pathological label image set optionally and carrying out bar code identification on the selected pathological label image;

the first judging unit is used for judging whether a bar code exists in a selected pathological label image or not;

the character recognition unit is used for recognizing characters of the selected pathological label image by using an OCR character recognition method when the first judgment module judges that the bar code does not exist in the selected pathological label image, so as to obtain and output a pathological number corresponding to the selected pathological label image;

the second judging unit is used for judging whether the bar code is a two-dimensional code or not when the first judging module judges that the bar code exists in the selected pathological label image;

the two-dimensional code analyzing unit is used for analyzing the bar code to obtain and output a pathology number corresponding to the selected pathology label image when the second judging unit judges that the bar code is the two-dimensional code;

and the query unit is used for searching a preset pathological number database according to the bar code when the second judgment unit judges that the bar code is not the two-dimensional code, obtaining and outputting a pathological number corresponding to the selected pathological label image.

Through the recognition module with the structure, a proper recognition method can be automatically selected according to actual conditions, the corresponding pathological number can be automatically recognized and automatically output, the recognition efficiency and the recognition accuracy rate are high, and the troubles of traditional naked eye recognition and manual input are effectively avoided.

Preferably, the identification module further comprises a database establishing unit;

the database establishing unit is used for establishing the pathology number database in advance.

Preferably, as shown in fig. 7, a relationship table in which barcodes correspond to pathology numbers one to one is stored in the pathology number database; the identification module also comprises a third judgment unit and a bar code removal unit;

the third judging unit is used for judging whether to acquire the relation table or not when the second judging unit judges that the bar code is not the two-dimensional code;

the bar code removing unit is used for removing a bar code in a selected pathological label image when the third judging unit judges that the relation table is not acquired, and sending the pathological label image with the bar code removed to the character recognition unit for character recognition to obtain and output a corresponding pathological number;

the query unit is specifically configured to query the relationship table according to the barcode when the third determination unit determines that the relationship table is obtained, obtain and output a pathology number corresponding to the selected pathology label image.

Through the identification module that still includes third judgement unit and bar code removal unit above-mentioned, can guarantee the success rate of pathology number discernment, further effectively guarantee the recognition effect.

Preferably, the barcode removing unit is specifically configured to:

The bar code removing unit can distinguish a foreground image (including a bar code region image and a text region image) from a background region in the pathological label image through graying and reverse binarization processing, so that the foreground image can be conveniently processed subsequently; because the length-width ratio of the one-dimensional bar code is usually in a certain specification, the length-width ratio of the foreground region is calculated, the bar code region image in the foreground image can be extracted according to the preset extraction length-width ratio, and then the extracted bar code region image is assigned according to the pixel value of the background image, so that the bar code region image can be treated as the background region, and the purpose of removing the bar code region image is achieved; the step of removing the bar code can effectively improve the accuracy of subsequent character recognition.

In a third embodiment, based on the first embodiment and the second embodiment, the present embodiment further discloses a slice pathology number identification apparatus, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor, and when the computer program is executed, the specific steps of S1 to S3 shown in fig. 1 are implemented.

The identification of the slice pathological number is realized by the computer program stored in the memory and running on the processor, the pathological number on the pathological section can be automatically identified and automatically input, the troubles of naked eye identification and manual input are effectively avoided, the labor cost is greatly saved, the error easiness of manual identification and manual input is effectively overcome, the identification accuracy and the input efficiency of the pathological number are greatly improved, the management of a slice management system is facilitated, a doctor can look up the pathological section of a specified patient, and the working efficiency of the doctor is improved.

The present embodiment also provides a computer storage medium having at least one instruction stored thereon, where the instruction when executed implements the specific steps of S1-S3.

The identification of the slice pathological number is realized by executing the computer storage medium containing at least one instruction, the pathological number on the pathological slice can be automatically identified and automatically input, the troubles of naked eye identification and manual input are effectively avoided, the labor cost is greatly saved, the error easiness of manual identification and manual input is effectively overcome, the identification accuracy and the input efficiency of the pathological number are greatly improved, the management of a slice management system is facilitated, a doctor can look up the pathological slice of a specified patient, and the working efficiency of the doctor is improved.

Details of S1 to S3 in this embodiment are not described in detail in the first embodiment and the detailed description of fig. 1 to fig. 5, which are not repeated herein.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims

1. A method for identifying a section pathology number is characterized by comprising the following steps:

2. The method for identifying a slice pathology number according to claim 1, wherein in the step 2, the specific step of performing region segmentation on the pathology slice image includes:

step 21: according to a first preset zooming size, carrying out primary zooming processing on the pathological section image to obtain a primary zoomed section image;

step 22: and carrying out image segmentation on the primarily zoomed slice image by adopting an image segmentation method based on a graph to obtain a plurality of slice region segmentation images.

3. The method for identifying a slice pathology number according to claim 2, wherein in the step 2, the specific step of performing region filtering on the pathology slice image includes:

step 23: respectively carrying out corrosion expansion treatment on each slice region segmentation image to obtain slice region preprocessing images corresponding to each slice region segmentation image one to one;

step 24: acquiring the length and width of an external rectangle of each slice area preprocessed image, and primarily filtering the length and width of all external rectangles according to a preset first filtering discriminant to obtain a plurality of slice area primary filtering images;

the first filtering discriminant is specifically:

wherein the content of the first and second substances,

and

respectively as the i-th slice regionLength and width of circumscribed rectangle of image, n_wAnd n_hRespectively, a transverse accommodation quantity of pathological sections accommodated in the transverse direction of the section box and a longitudinal accommodation quantity of pathological sections accommodated in the longitudinal direction, W_imgAnd H_imgRespectively the length and width of the primary scaled slice image;

step 25: carrying out region merging on the primary filtered images of all the slice regions to obtain a plurality of slice region merged images;

step 26: acquiring the length and width of a circumscribed rectangle of each slice region merged image, and filtering all slice region merged images again according to a preset second filtering discrimination formula to obtain a plurality of slice region re-filtered images;

the second filtering discriminant is specifically:

or

Wherein the content of the first and second substances,

and

step 27: according to a second preset scaling size, carrying out secondary scaling processing on the filtered image of each slice area respectively to obtain a slice area scaled image corresponding to the filtered image of each slice area one by one;

step 28: extracting gradient direction histogram characteristics corresponding to the zoomed images of each slice region one by one, and clustering the zoomed images of each slice region according to the gradient direction histogram characteristics by adopting a DBSCAN density clustering method to obtain a plurality of clustering categories and slice region zoomed image sets and image quantities under each clustering category; and determining a slice region scaling image set under the cluster category corresponding to the maximum value in all the image numbers as the pathology label image set.

4. The method for identifying a slice pathology number according to claim 3, wherein the step 25 specifically comprises:

step 251: optionally selecting two adjacent slice region primary filtering images in all slice region primary filtering images, and respectively calculating the image areas of the two selected slice region primary filtering images and the overlapping area and the similarity between the two selected slice region primary filtering images according to the length and the width of the circumscribed rectangle of the two selected slice region primary filtering images;

similarity_kk+1＝min(|w_k+1-w_k|,|h_k+1-h_k|)；

step 252: respectively calculating the overlapping degree and the coverage degree between the selected two slice region primary filtering images according to the image areas of the selected two slice region primary filtering images and the overlapping area between the selected two slice region primary filtering images;

step 253: judging whether the overlapping degree, the coverage degree and the similarity between the two selected slice region primary filtering images meet a preset merging judgment formula, if so, merging the two selected slice region primary filtering images to obtain corresponding slice region merging images, and if not, returning to the step 251 until every two adjacent slice region primary filtering images in all the slice region primary filtering images are traversed;

the merging discriminant is specifically:

step 254: according to the method of the step 251 to the step 253, a plurality of slice region merged images are obtained.

5. The method for identifying a slice pathology number according to claim 1, wherein the step 3 specifically comprises:

step 31: optionally selecting one pathological label image in the pathological label image set, performing bar code identification on the selected pathological label image, judging whether the bar code exists in the selected pathological label image, if so, sequentially executing the step 32 and the step 34, and if not, sequentially executing the step 33 and the step 34;

step 32: judging whether the bar code is a two-dimensional code or not, if so, analyzing the bar code by the two-dimensional code to obtain and output a pathology number corresponding to one selected pathology label image, and if not, searching a preset pathology number database according to the bar code to obtain and output a pathology number corresponding to one selected pathology label image;

step 33: performing character recognition on the selected pathological label image by using an OCR character recognition method to obtain and output a pathological number corresponding to the selected pathological label image;

step 34: and traversing each pathological label image in the pathological label image set to obtain and output a pathological number corresponding to each pathological label image one by one.

6. The method for identifying a section pathology number according to claim 5, wherein in the step 32, a relationship table in which bar codes and pathology numbers are in one-to-one correspondence is stored in the pathology number database;

then, in step 32, before searching the preset pathology number database according to the barcode, the method further includes:

and judging whether the relation table is acquired or not, if so, inquiring the relation table according to the bar code to obtain and output a pathology number corresponding to the selected pathology label image, otherwise, removing the bar code in the selected pathology label image, and executing the step 33.

7. The method for identifying the pathological section number according to claim 6, wherein the step of removing the barcode in the selected pathological label image comprises:

8. A system for identifying a slice pathology number, which is applied to the method for identifying a slice pathology number according to any one of claims 1 to 7, and comprises an image acquisition module, a region processing module, an identification module, and an output module;

9. An apparatus for identifying a pathology number of a slice, comprising a processor, a memory and a computer program stored in the memory and executable on the processor, the computer program when executed implementing the method steps according to any one of claims 1 to 7.

10. A computer storage medium, the computer storage medium comprising: at least one instruction which, when executed, implements the method steps of any one of claims 1 to 7.