Method for identifying and evaluating erythrocyte in lymph node sinus of experimental animalTechnical Field
The invention relates to the field of image processing, in particular to a method for identifying and evaluating erythrocyte in lymph node sinus of an experimental animal.
Background
Currently, in experimental animals, semi-quantitative analysis is mainly carried out on erythrocyte lesions in lymph node sinus by pathological diagnostician. The influence of human factors is large when the pathologist performs semi-quantitative analysis, for example, the diagnosis speed of different diagnosticians is inconsistent with the standard, and even the same diagnostician may have diagnosis drift.
Disclosure of Invention
Based on the problems, the invention aims to provide a method for identifying and evaluating the erythrocyte in the lymph node sinus of an experimental animal, so that the purpose of identifying and evaluating the erythrocyte in the lymph node sinus on an HE (high-speed) staining slice is achieved, and the problem of diagnosis drift of erythrocyte lesions in the lymph node sinus of the experimental animal is solved.
The technical scheme adopted by the invention for achieving the purpose of the invention is that the method for identifying and evaluating the erythrocyte in the lymph node sinus of the experimental animal comprises the following steps:
s1, slice scanning: scanning lymph node HE stained sections of the experimental animals using a digital section scanner;
S2, lymph node tissue identification: identifying a desired lymph node tissue on the stained section and converting it into a first region of interest ROI1;
s3, identifying and evaluating erythrocyte in lymph node sinus: and setting an eosin staining characteristic threshold value smaller than a preset value N1 and a red-blue ratio larger than a preset value N2 on the first region of interest ROI1, identifying the red blood cells in the lymph node sinuses, making corresponding marks, counting the areas of the red blood cells in the lymph node sinuses, and finally converting the areas of the red blood cells in the lymph node sinuses into scores of the red blood cells in the lymph node sinuses.
The method is characterized in that red blood cells in the sinus are distributed in the lymphatic sinus, and the method is characterized in that how to mark the red blood cells in the sinus is a difficult point, if the red blood cells in the sinus, surrounding lymphocytes, macrophages and the like are marked together in a piece, the marked range has the characteristics of the lymphocytes, and the lymphocytes in other areas can be possibly marked as lesion areas, so that if only the red blood cells are required to be marked, the artificial marking difficulty is very high, and the method can furthest save the marking time and improve the marking accuracy by setting the red-blue ratio of the red blood cells and the eosin staining characteristic threshold value for marking; the intracavitary erythrocyte is a lesion where there are erythrocytes in the lymphatic sinus, and erythrocytes in normal blood vessels, but not in the scope of the lesion, there is therefore a need to exclude erythrocytes in blood vessels, and the present invention artificially marks erythrocytes in normal blood vessels as background after marking erythrocytes by means of a red-to-blue ratio and an eosin staining characteristic threshold.
Further, the specific steps of the step S2 lymph node tissue identification are as follows:
S2.1, selecting a certain number of lymph node HE staining sections as a training set, drawing a region of interest including lymph node tissue regions and non-lymph node regions on the staining sections of the training set, marking all the lymph node tissue regions in the region of interest with corresponding marks, for example, blue marks, marking the non-lymph node regions as a background, for example, green marks, training by applying a deep learning U-Net model, namely DEEP LEARNING-U-Net model, and adjusting the training until the lymph node tissue recognition accuracy reaches more than 90%;
S2.2, importing a stained section to be formally subjected to lymph node tissue identification into software for completing model training in the step S2.1, drawing an initial region of interest (ROI 0), carrying out lymph node tissue identification in the initial region of interest (ROI 0), and automatically marking the identified lymph node tissue region as a background and automatically marking the non-lymph node region as a background;
S2.3, carrying out post-treatment, wherein the post-treatment step comprises the following steps:
S2.3.1, filling a blank area in lymph node tissue completely;
s2.3.2, replacing lymph node tissues with areas smaller than a preset value S1 with a background;
s2.3.3, clearing the background, e.g., green background, leaving the stained section with only lymph node tissue;
S2.3.4, transforming the retained lymph node tissue into a first region of interest ROI1.
Still further, the preset value S1 is equal to 10000 μm.
Further, the specific steps of identifying and evaluating the erythrocyte in the sinus of the lymph node S3 are as follows:
s3.1, selecting a certain number of lymph node HE staining sections as a training set, drawing a region of interest on the staining sections of the training set, setting an eosin staining characteristic threshold value smaller than a preset value N1, setting a red-blue ratio larger than a preset value N2, clearing out sinus red blood cells with an area smaller than a preset value S2, marking all sinus red blood cell regions in the region of interest as blue, marking non-sinus red blood cell regions as a background such as green, marking the non-sinus red blood cell regions including blood vessels and blood vessel red blood cells, training by using a deep learning semantic segmentation model, namely DEEP LEARNING-DeepLabv3+ model, and adjusting the training until the sinus red blood cell recognition accuracy reaches more than 90%;
S3.2, importing the first region of interest ROI1 which is required to formally identify the red blood cells in the sinus of the lymph node into software for completing model training in the step S3.1, setting the eosin staining characteristic threshold value smaller than a preset value N1, setting the red-blue ratio larger than a preset value N2, identifying the red blood cells in the sinus of the lymph node, and automatically marking the identified red blood cell region in the sinus as a corresponding mark and automatically marking the non-sinus red blood cell region as a background;
S3.3, then carrying out post-treatment, wherein the post-treatment step comprises the following steps:
S3.3.1, removing the red blood cells in the sinus with the area smaller than a preset value S2;
S3.3.2, removing the background, such as a green background, so that the stained section only retains the red blood cells in the sinus;
s3.4, counting to obtain the area of red blood cells in the sinuses of each lymph node;
S3.5, converting the area of the red blood cells in the sinuses of each lymph node into scores of the red blood cells in the sinuses of each lymph node.
Further, the preset value N1 ranges from 160 to 180, and the preset value N2 ranges from 3 to 6.
Still further, the preset value N1 is equal to 170, and the preset value N2 is equal to 5.
Still further, the preset value S2 is equal to 2 μm.
Further, the specific method for score conversion in step S3.5 is as follows: the in-sinus red blood cell score of the area less than or equal to S10 is equal to 0, the in-sinus red blood cell score of the area greater than S10 and less than or equal to S11 is equal to 1, the in-sinus red blood cell score of the area greater than S11 and less than or equal to S12 is equal to 2, the in-sinus red blood cell score of the area greater than S12 and less than or equal to S13 is equal to 3, the in-sinus red blood cell score of the area greater than S13 and less than or equal to S14 is equal to 4, the in-sinus red blood cell score of the area greater than S14 is equal to 5, and the S10, S11, S12, S13 and S14 are all preset values.
Still further, the preset value S10 is 8000-10000 mu m, the preset value S11 is 40000-60000 mu m, the preset value S12 is 90000-110000 mu m, the preset value S13 is 230000 ～ 270000 mu m, and the preset value S14 is 330000 ～ 370000 mu m.
The beneficial effects of the invention are as follows:
the invention can realize artificial intelligence of identification and evaluation of the erythrocyte in the lymph node sinus of the experimental animal, realize rapid and accurate batch operation and solve the problem of diagnosis drift of the erythrocyte lesion in the lymph node sinus of the experimental animal.
Drawings
FIG. 1 is a schematic diagram of a general flow chart of an embodiment of the present invention;
FIG. 2 is a diagram showing the effect of the step S2;
FIG. 3 is an original view of the unlabeled color corresponding to FIG. 2;
FIG. 4 is a diagram showing the effect of step S3 according to the embodiment of the present invention;
Fig. 5 is an original view of the unlabeled color corresponding to fig. 4.
Detailed Description
In order that the above-recited objects, features and advantages of the present invention will be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
FIG. 1 shows a specific embodiment of the method for identifying and evaluating erythrocyte in lymph node sinus of experimental animal according to the present invention, which comprises the steps of:
s1, slice scanning: scanning the lymph node HE stained sections of the experimental animal using a digital section scanner, and importing the digital sections into Visiospharm software;
S2, lymph node tissue identification: identifying a desired lymph node tissue on the stained section and converting it into a first region of interest ROI1;
S3, identifying and evaluating erythrocyte in lymph node sinus: and setting an eosin staining characteristic threshold value smaller than 170 and a red-blue ratio larger than 5 on the first region of interest (ROI 1), identifying the red blood cells in the lymph node sinuses, making corresponding marks, counting the areas of the red blood cells in the lymph node sinuses, and finally converting the areas of the red blood cells in the lymph node sinuses into scores of the red blood cells in the lymph node sinuses.
In this embodiment, the specific steps of step S2 lymph node tissue identification are as follows:
S2.1, selecting a certain number of lymph node HE staining sections as a training set, drawing a region of interest including lymph node tissue regions and non-lymph node regions on the staining sections of the training set, marking all the lymph node tissue regions in the region of interest as blue (class 2), marking the non-lymph node regions as green (class 1), then training by applying a deep learning U-Net model, namely DEEP LEARNING-U-Net model, and adjusting the training until the lymph node tissue recognition accuracy reaches more than 90%;
S2.2, importing a stained section to be formally subjected to lymph node tissue identification into software for completing model training in the step S2.1, drawing an initial region of interest (ROI 0), performing lymph node tissue identification in the initial region of interest (ROI 0), automatically marking the identified lymph node tissue region as blue (class 2), and automatically marking the non-lymph node region as green (class 1);
S2.3, re-running a post processing program, wherein the post processing steps comprise:
s2.3.1, setting a parameter Fill Holes Lbl, namely class2, namely filling a blank area in lymph node tissue completely;
S2.3.2, setting parameters change by shape Lbl, namely replacing class 2 with a class 1 with an area smaller than 10000 mu m, namely replacing blue lymph node tissues with an area smaller than 10000 mu m with a green background;
S2.3.3, setting parameters change Lbl, namely replacing class1 with clear, namely clearing the green background, and enabling the stained section to only retain blue lymph node tissues;
S2.3.4 setting parameters outline as ROI Lbl, outputting class2 as ROI1, i.e. converting the remaining blue lymph node tissue into a first region of interest ROI1.
The effect chart after the step is performed is shown in fig. 2, fig. 3 is an original chart of unlabeled colors corresponding to fig. 2, and the effect after the blue lymph node tissue is screened out in the step can be seen by comparing fig. 3 and fig. 2.
In this embodiment, the specific steps of identifying and evaluating the erythrocyte in the sinus of the lymph node S3 are as follows:
S3.1, selecting a certain number of lymph node HE staining sections as a training set, drawing a region of interest on the staining sections of the training set, selecting a threshold value threshold as an algorithm in tissue Classification, selecting class 2, setting an eosin staining characteristic threshold value to be less than 170, setting a red-blue ratio to be greater than 5, and setting parameters change by shape Lbl: replacing class 2 with the area smaller than 2 [ mu ] m and the area smaller than 2 [ mu ] m with clear, running a threshold algorithm to obtain an image of marked sinus red blood cells, marking all the sinus red blood cell areas in the region of interest as blue, marking the non-sinus red blood cell areas as green, marking the non-sinus red blood cell areas including blood vessels and red blood cells in the blood vessels, training by using a deep learning semantic segmentation model, namely DEEP LEARNING-DeepLabv & lt+ & gt, and adjusting the training until the accuracy of identifying the sinus red blood cells reaches more than 90%; by the arrangement, the intracavitary erythrocytes can be marked to the greatest extent, if only the eosin staining characteristic threshold value is set to be smaller than 170, many nuclei can be marked, so that one condition is added, namely the red-blue ratio parameter is larger than 5, and the intracavitary erythrocytes are removed again by setting the area size;
s3.2, importing the first region of interest (ROI 1) which is required to formally identify the red blood cells in the sinus of the lymph node into software for completing model training in the step S3.1, selecting a threshold value threshold in tissue Classification, selecting class 2, setting an eosin staining characteristic threshold value smaller than 170, setting a red-blue ratio larger than 5, and carrying out the red blood cell identification in the sinus of the lymph node, wherein the identified red blood cell area is automatically marked as blue, and the non-sinus red blood cell area is automatically marked as green;
S3.3, re-running a post processing program, wherein the post processing steps comprise:
S3.3.1, setting parameters change by shape Lbl: replacing class 2 with the area smaller than 2 mu m with clear, namely removing red blood cells in sinuses with the area smaller than 2 mu m;
S3.3.2, setting a parameter change Lbl, namely replacing class 1 with clear, namely clearing a green background, and enabling the dyed slice to only retain blue sinus red blood cells;
s3.4, selecting class2 from output variables output variables, and counting after operation to obtain the area of red blood cells in the sinuses of each lymph node;
S3.5, converting the area of the red blood cells in the sinuses of each lymph node into scores of the red blood cells in the sinuses of each lymph node.
In this embodiment, the preset value S10 is 10000 μm, S11 is 50000 μm, S12 is 100000 μm, S13 is 250000 μm, S14 is 350000 μm, and therefore, the specific method for score conversion in step S3.5 is as follows:
The area is less than or equal to 10000 mu m, the score of red blood cells in the sinus is equal to 0, the area is greater than 10000 mu m and less than or equal to 50000 mu m, the score of red blood cells in the sinus is equal to 1, the area is greater than 50000 mu m and less than or equal to 100000 mu m, the score of red blood cells in the sinus is equal to 2, the score of red blood cells in the sinus is greater than 100000 mu m and less than or equal to 250000 mu m is equal to 3, the score of red blood cells in the sinus is greater than 250000 mu m and less than 350000 mu m is equal to 4, and the score of red blood cells in the sinus is greater than 350000 mu m is equal to 5.
The effect chart after the step is performed is shown in fig. 4, fig. 5 is an original chart of unlabeled color corresponding to fig. 4, and comparing fig. 5 and fig. 4 can see the effect after the step of screening out blue in-sinus red blood cells.
Pathologists in toxicity pathology practice need to score according to the severity and extent of the lesions, and the area conversion obtained in this example can be better accessed into the pathologist's diagnostic scoring system for better application in toxicity pathology diagnosis.
The embodiment of the invention is realized by adopting an AI-based image analysis software Visiogenome platform in the prior art, and can also be realized by adopting other software platforms with similar functions.
The above examples of the present invention are merely illustrative of the present invention and are not intended to limit the embodiments of the present invention. Other variations and modifications of the present invention will be apparent to those of ordinary skill in the art in light of the foregoing description. Not all embodiments are exhaustive. Obvious changes and modifications which are extended by the technical proposal of the invention are still within the protection scope of the invention.