Microscopic image intelligent splicing system

Abstract

The invention discloses a microscopic image intelligent splicing system, which comprises a microscopic image acquisition unit and an image splicing unit, and is characterized in that the image splicing unit is used for firstly splicing and repairing microscopic images of a glass slide in a full view collected under a highest-multiple objective lens to obtain a non-joint-mark full-view complete image under the highest-multiple objective lens; and then splicing and repairing the microscopic images under the other multiple objective lenses according to the no-joint-mark full-field-of-view complete image under the highest multiple objective lens to obtain no-joint-mark full-field-of-view complete images under the other multiple objective lenses. According to the method, a non-joint-mark full-view complete image is formed in a grading repairing mode, a real cell image can be accurately reduced, the clinical film reading efficiency is greatly improved, and the missing detection rate and the false detection rate are reduced.

Description

Technical Field

[0001] The invention belongs to the field of cell detection, and in particular relates to a microscopic image intelligent splicing system. Background Art

[0002] When performing a peripheral blood morphology examination, in order to ensure the accuracy of the results and allow doctors to observe features such as cell nuclei or cell particles in detail, a microscope is needed to accurately capture the cells in the blood smear and present high-definition cell images. During morphological microscopy, if you want to observe cells in the entire field of view, using a low-power microscope can only clearly see the distribution of cells, and it is impossible to accurately determine the cell type and then diagnose the disease. Therefore, under normal circumstances, the doctor who reads the film will disassemble a film into multiple magnifications (10×, 40×, 100×, etc.) to present high-definition images of cells in the field of view, so as to clearly observe the cell morphology, read the film, and then determine the cell type and perform pathological analysis and diagnosis. Microscope inspectors need to take pictures and observe in multiple fields of view. In this process, in addition to the possibility of overlapping fields of view, incomplete observation of cells in the edge area, long microscopic examination time, and reliance on the experience of the film reader, there may also be situations where larger cells are difficult to present under a 100× microscope. For example, larger cells with a cell body diameter of 15 to 30 μm are not completely within the same 100× field of view, and often require multiple 100× microscope images to fully present the entire cell. This method not only has low film reading efficiency, but also cannot conveniently achieve high accuracy and fast and automatic cell detection in one field of view.

[0003] At present, stitching multiple local fields of view to form a complete field of view of cell imaging is a common strategy for simplifying high-accuracy and rapid automatic cell detection. The overall cell morphology and distribution of the whole slide can be clearly viewed within a limited field of view. Define a local field of view stitched into a complete field of view under the same magnification as a unit. In order to ensure the precise fit between the stitched unit and the target field of view, a small area of ​​overlap will appear at the interface of the stitching. The overlapping area at the stitching interface is defined as the "joint area". The cell morphology of the microscopic image in the joint area is quite different from the real cell morphology. It is difficult for humans or computers to accurately identify the cells at the joint. There are multiple joints in a complete field of view. On the one hand, some key areas may be missed due to the presence of joints during the switching of the mobile field of view, resulting in errors in the test results and misdiagnosis. On the other hand, although the stitching of different magnification fields of view can easily see the entire slide, the internal structure and composition of the cells cannot be observed because of the overlap of cells at the joints, and the requirements for accurate identification of cell detection cannot be met.

[0004] The processing method for "100× magnification images with stitching marks" is to use existing image repair technology to directly repair and correct the stitching marks to form "100× magnification images without stitching marks". Since the cell image is larger in the 100× magnification, its cell features are more obvious compared to other lower magnification images. When comparing the feature positions and restoring the restoration, the restoration error can be controlled within a small range to restore the true image of the cell.

[0005] Therefore, there is a need for an artificial intelligence stitching image analysis system based on peripheral blood morphological examination, which combines the fusion of acquisition technology and image processing technology, can quickly and accurately find the target cell image, clearly present the entire film without any seams in one picture, and achieve stepless magnification of 10× to 100× magnification images, complete information preservation of peripheral blood smears, high-definition cell smears, and traceability of 100× magnification image seam data. Summary of the invention

[0006] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is to provide a microscopic image intelligent stitching system.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] A microscopic image intelligent stitching system, comprising:

[0009] A microscopic image acquisition unit, used to acquire a microscopic image of the entire field of view of the slide under objective lenses of multiple magnifications; and

[0010] The image stitching unit is used to first stitch and repair the microscopic image of the entire field of view of the slide collected under the highest magnification objective lens, so as to obtain a seamless full-field complete image under the highest magnification objective lens; then stitch and repair the microscopic images under the remaining magnification objective lenses according to the seamless full-field complete image under the highest magnification objective lens, so as to obtain a seamless full-field complete image under the remaining magnification objective lenses.

[0011] Further, the image stitching unit includes a positioning unit and a sorting unit, the positioning unit is used to calculate the pan-tilt coordinates of the target cells of each microscopic image under the highest magnification objective lens according to the coordinate system preset by the automatic microscope acquisition platform, and locate each microscopic image under the highest magnification objective lens according to the pan-tilt coordinates; and respectively calculate the pan-tilt coordinates of the target cells of each microscopic image under each non-highest magnification objective lens, and convert them to the platform coordinates of the target cells of the corresponding microscopic image under the high-magnification objective lens for positioning, and judge the clarity of the image under the high-magnification objective lens, and perform coordinate correction through image similarity comparison;

[0012] The sorting unit is used to splice the microscopic images of the whole field of view of the slide collected under the highest magnification objective lens according to the pan-tilt coordinates of the target cells of each microscopic image under the highest magnification objective lens, so as to obtain a full-field complete image with a joint mark under the highest magnification objective lens, wherein the full-field complete image with a joint mark forms a joint area between two adjacent spliced ​​microscopic images; and to repair the joint area of ​​the full-field complete image with a mark spliced ​​under the highest magnification objective lens through the wavelet transform image fusion technology, so as to obtain a full-field complete image without a joint mark under the highest magnification objective lens; and An algorithm is used to compare the coordinate positions and image similarities of each microscopic image under each non-highest magnification objective lens with the seamless full-view complete image under a higher magnification objective lens, and regions with adjacent coordinates and consistent image comparisons are spliced ​​with traces to form a seamless full-view complete image under the objective lens of that magnification; then, the seamless full-view complete image under each non-highest magnification objective lens is corrected using the seamless full-view complete image under a higher magnification objective lens to obtain the seamless full-view complete images under each non-highest magnification objective lens.

[0013] Furthermore, the positioning unit uses a neural network model to perform image similarity comparison.

[0014] Furthermore, the algorithm adopted by the sorting unit is the Shell sorting algorithm.

[0015] Furthermore, the multiple magnification objective lenses include a "10×" objective lens, a "40×" objective lens and a "100×" objective lens, the highest magnification objective lens is a "100×" objective lens, and the non-highest magnification objective lenses include a "10×" objective lens and a "40×" objective lens.

[0016] Furthermore, the joint area is an overlapping area formed at the joint of two adjacent microscopic images.

[0017] Furthermore, the microscopic image acquisition unit is connected to an automatic microscope acquisition platform, and the automatic microscope acquisition platform is provided with an objective lens converter, and the objective lens converter is used to automatically replace the objective lens on the automatic microscope acquisition platform.

[0018] Furthermore, it also includes a control unit, which is used to control the automatic microscope acquisition platform, the microscopic image acquisition unit and the image stitching unit.

[0019] Furthermore, it also includes a storage unit, which is used to store the collected microscopic images and the seamless full-view complete images obtained after splicing.

[0020] Furthermore, it also includes a display unit, which is used to display a complete image with a seamless full field of view.

[0021] In the present invention, the staff only needs to place the slide in the observation position of the automatic microscope acquisition platform, and the system will automatically perform accurate comparison, filling, correction and splicing of the field of view, and output, display and store the best imaging image of the target cell. A seamless full-field complete image is formed by graded filling, repair, correction and other methods; data can be traced to the source, realizing data information management, avoiding misdiagnosis or missed diagnosis caused by cell omission, and accurately restoring the real image of the cell on the basis of existing technology, greatly improving the efficiency of clinical film reading. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0023] Figure 1 This is a structural block diagram of an embodiment of the microscopic image intelligent stitching system of the present invention.

[0024] Figure 2 Flowchart for acquiring images for the microscopic image acquisition unit.

[0025] Figure 3 Schematic diagram of the image acquisition process.

[0026] Figure 4 Schematic diagram of the principle of hill climbing method.

[0027] Figure 5 Schematic diagram of the stitching principle of microscopic images. DETAILED DESCRIPTION

[0028] The following describes the implementation methods of the present invention through specific examples. The illustrations provided in the following embodiments are only used to schematically illustrate the basic concept of the present invention. The following embodiments and features in the embodiments may be combined with each other without conflict.

[0029] See also Figure 1 , Figure 1The block diagram of the structure of an embodiment of the microscopic image intelligent stitching system of the present invention. The microscopic image intelligent stitching system of this embodiment includes a microscopic image acquisition unit and an image stitching unit, and the microscopic image acquisition unit is used to collect microscopic images of the entire field of view of a slide (a carrier for placing specimen cells) under multiple magnification objective lenses. In this embodiment, the slide is a blood smear, and the multiple magnification objective lenses include a "10×" objective lens, a "40×" objective lens, and a "100×" objective lens. The highest magnification objective lens is a "100×" objective lens, and the non-highest magnification objective lens includes a "10×" objective lens and a "40×" objective lens. In order to facilitate automatic completion of image acquisition and objective lens switching, the microscopic image acquisition unit is connected to an automatic microscope acquisition platform, and the automatic microscope acquisition platform is provided with an objective lens converter, and the objective lens converter is used to automatically replace the "10×" objective lens, the "40×" objective lens, and the "100×" objective lens on the automatic microscope acquisition platform.

[0030] See also Figure 2 and Figure 3 , the method for collecting images by the microscopic image collection unit may include the following steps:

[0031] S1. A spatial rectangular coordinate system is established with the plane where the blood smear is located as the XOY plane and the elevation direction as the Z axis direction. Generally, a corner point of the blood smear is used as the origin of the spatial rectangular coordinate system, and the length direction and width direction of the blood smear are used as the X axis direction and Y axis direction of the XOY plane, respectively. In this embodiment, a spatial rectangular coordinate system is established with the corner point of the lower left corner of the blood smear as the origin, the length direction of the blood smear as the X direction, the width direction of the blood smear as the Y direction, and the elevation direction as the Z axis direction.

[0032] S2. Taking a single field of view of the microscope (i.e., the area corresponding to one field of view of the microscope) as a unit scanning area, a "battlement-like" moving method is used to scan the entire blood smear along an "S"-shaped scanning path.

[0033] S3. The unit scanning area corresponding to the starting position of the scanning path is taken as the first unit scanning area. The first unit scanning area of ​​the scanning path is photographed using a microscope, and the focus area is divided into n layers along the Z-axis direction. The spacing between adjacent layers is generally 50nm to 500nm, which can be determined according to the actual situation of the specimen in the blood smear. The layer image with the highest clarity is identified along the Z-axis direction by the hill climbing method as the microscopic image of the first unit scanning area, that is, the final image of the first unit scanning area; the layer position with the highest clarity found is the clearest focus position of the microscope.

[0034] See also Figure 4The hill climbing method is a commonly used method for identifying the layer image with the highest definition. Its principle is: scan the microscopic image with a continuous single objective lens field of view as a unit, determine the focus elevation (Z direction) range h, estimate the coordinate position of the clear layer, and then perform layered scanning in the positive Z direction within h with a unit distance of m to determine the clarity of the scanned layer and compare it with the clarity of the estimated clear layer. When the clarity ratio exceeds p, determine the coordinates of the layer position (X1, Y1, Z1); immediately perform layered scanning in the negative Z direction of the estimated clear layer with a unit distance of m to determine the clarity of the scanned layer and compare it with the clarity of the estimated clear layer. When the clarity ratio exceeds p, determine the coordinates of the layer position (X2, Y2, Z2); at this time, take (|Z1-Z2|) as the elevation, defined as h', within h', take m' as the unit distance, repeat the above method steps, continuously determine and select the clear layer, and finally segment n layers and select the clearest layer in the area as the final image of the position.

[0035] According to the actual situation of the specimen in the blood smear, since the imaging distance of the specimen varies only within a very small range, the change in the sharpest focus position of two adjacent unit scanning areas is very small. Therefore, the sharpest focus position of the previous unit scanning area (that is, the Z coordinate of the layer image with the highest clarity) can be directly used as the center to float up and down in a small range to find the layer image with the highest clarity in the next unit scanning area. Its Z coordinate is the sharpest focus position of the unit scanning area, which can greatly reduce the number of segmented layers.

[0036] S4. Use a microscope to photograph the next unit scanning area of ​​the scanning path, and divide 2m+1 layers along the Z axis direction based on the standard layer Z coordinate of the unit scanning area, 2m+1<n. The method of dividing the focus area of ​​the recognition image into 2m+1 layers along the Z axis direction based on the standard layer Z coordinate is:

[0037] Take the Z coordinate of the standard layer as the standard layer, start from the standard layer and step upward with a predetermined step value and divide it into m layers in sequence, start from the standard layer and step downward with a predetermined step value and divide it into m layers in sequence, and get 2m+1 layers including the standard layer. The step value generally ranges from 50nm to 500nm, which can be determined according to the actual situation of the specimen in the blood smear. Combined with the implementation situation, in order to simplify the calculation, the value of m is generally 2 or 3 to find a clear image.

[0038] In this embodiment, m=2. For example, after obtaining the microscopic image of the first unit scanning area, the second unit scanning area is photographed. At this time, according to the coordinate Y a1Determine the standard layer of the second unit scanning area, first obtain the layer image of the standard layer as the first layer image, then based on the standard layer, move up by one step value from the standard layer to obtain the layer image as the second layer image, and continue to move up by one step value to obtain the layer image as the third layer image, and move down by one step value from the standard layer to obtain the layer image as the fourth layer image, and continue to move down by one step value to obtain the layer image as the fifth layer image, thereby obtaining 5 layer images. Of course, depending on the implementation, sometimes m=3 can also be taken, so that 3 layers are divided above and below the standard layer, thereby obtaining 7 layer images.

[0039] S5. Identify the layer image with the highest definition from the 2m+1 layer images as the microscopic image of the unit scanning area, and calculate the Z-axis coordinate of the layer image with the highest definition as the standard layer Z coordinate of the next unit scanning area. In this embodiment, a convolutional neural network model is used to identify the layer image with the highest definition, and the specific method is as follows:

[0040] A convolutional neural network model for predicting the clarity of an image is pre-set, and the convolutional neural network model is trained to obtain a clarity recognition model. During recognition, each layer image is input into the clarity recognition model to obtain a clarity prediction value of each layer image, and the layer image with the highest clarity prediction value is output as the layer image with the highest clarity.

[0041] In order to test the accuracy of the clarity recognition model, in this step, the clarity factor of each layer image is calculated according to the grayscale of the image and it is determined whether the layer image output by the clarity recognition model is consistent with the layer image with the largest clarity factor value. If they are consistent, it can be considered that the layer image output by the clarity recognition model is the image with the highest clarity, that is, the clarity recognition model output is accurate. If the layer image output by the clarity recognition model is inconsistent with the layer image with the largest clarity factor value, it is considered that the layer image output by the clarity recognition model may not be the image with the highest clarity. The layer image output by the clarity recognition model is marked and can be manually verified later. If it is confirmed that the layer image output by the clarity recognition model is not the image with the highest clarity, the clarity recognition model can be improved. The formula for calculating the clarity factor of each layer image is as follows:

[0042]

[0043] Where σ represents the clarity factor; N represents the number of all pixels in the area; x i represents the gray level of the i-th pixel; μ represents the average gray level of the image.

[0044] Since the prediction method using the convolutional neural network model is more accurate, the layer image output by the convolutional neural network model is always used as the microscopic image of the unit scanning area. The calculation formula of the clarity factor is only used to assist in verifying the output of the convolutional neural network model and will not be used as the main basis for judgment.

[0045] S6. Repeat steps S4 and S5 to obtain a microscopic image of the entire blood smear.

[0046] Using the above image acquisition method, only the first unit scanning area needs to use the hill climbing method to identify the clearest image. Starting from the second unit scanning area, the processing of subsequent unit scanning areas no longer divides the layers of each area into n layers and then identifies and screens the clearest image, but only divides it into 5 layers, and identifies and screens the clearest image in the 5 layers. Compared with the existing method, when this embodiment processes clear images of the same number of fields of view, the number of required segmentations and the number of identifications after segmentation are reduced, and the processing volume is reduced by about n / 5 times that of the existing method, which greatly reduces the requirements for computing power and improves the imaging speed. It can also truly restore various cell features, perfectly present the delicate texture, color and internal details of the cells, including the clear reproduction of cell lines, cytoplasm, particles, etc.

[0047] The image stitching unit is used to first stitch and repair the microscopic image of the entire field of view of the blood smear collected under the "100×" objective lens (i.e., the highest magnification objective lens) to obtain a seamless full-field complete image under the "100×" objective lens; then, according to the seamless full-field complete image under the "100×" objective lens, stitch and repair the microscopic images under the remaining magnification objective lenses (i.e., the "10×" objective lens and the "40×" objective lens) to obtain seamless full-field complete images under the "10×" objective lens and the "40×" objective lens, respectively.

[0048] The image stitching unit may include a positioning unit and a sorting unit. The positioning unit is used to calculate the PTZ coordinates of the target cells (i.e., the cells to be observed and counted) of each microscopic image under the "100×" objective lens according to the preset coordinate system of the automatic microscope acquisition platform, and to locate each microscopic image under the "100×" objective lens according to the PTZ coordinates; and to calculate the PTZ coordinates of the target cells of each microscopic image under each non-highest magnification objective lens (i.e., the "10×" objective lens and the "40×" objective lens), and convert them to the platform coordinates of the target cells of the corresponding microscopic image under the high-magnification objective lens (i.e., the objective lens with a magnification higher than the non-highest magnification objective lens) for positioning. For example, when positioning each microscopic image under the "40×" objective lens, convert it to the "100×" objective lens, and when positioning each microscopic image under the "10×" objective lens, convert it to the "40×" objective lens or the "100×" objective lens. And the clarity of the image under the high-magnification objective lens is judged, and the coordinates are corrected by image similarity comparison to achieve precise positioning. In this embodiment, the positioning unit uses a neural network model to perform image similarity comparison.

[0049] The sorting unit is used to splice the microscopic images of the whole field of view of the blood smear collected under the "100×" objective lens according to the pan-tilt coordinates of the target cells of each microscopic image under the "100×" objective lens, so as to obtain a full-field complete image with a joint under the "100×" objective lens, wherein the full-field complete image with a joint forms a joint area between two adjacent spliced ​​microscopic images. Since there is no objective lens with a larger magnification to correct the 100× objective lens image, the joint area of ​​the full-field complete image with a joint under the "100×" objective lens is repaired by wavelet transform image fusion technology, so as to correct the full-field complete image with a joint into a full-field complete image without a joint, and obtain a full-field complete image without a joint under the "100×" objective lens.

[0050] The seam area is the overlapping area formed at the joint of two adjacent microscopic images. In this embodiment, the width of the seam area is 15 microns. In peripheral blood, the diameter of red blood cells is 6 to 9 microns, the diameter of white blood cells is 10 to 15 microns, and the diameter of platelets is 1.5 to 3 microns. A 15-micron overlapping area is reserved at the seam position, which is exactly the width of a cell, meeting the requirement of basically restoring all the morphologies of various cells in peripheral blood.

[0051] The sorting unit is also used to use an algorithm to compare the coordinate positions and image similarities of each microscopic image under the "10×" objective lens and the "40×" objective lens with the seamless full-view complete image under the higher magnification objective lens, and to perform traced splicing on the areas with adjacent coordinates and consistent image comparisons to form a seamless full-view complete image under the objective lens of that magnification. Afterwards, the seamless full-view complete image under the higher magnification objective lens is used to correct the joint areas of the seamless full-view complete images under the "10×" objective lens and the "40×" objective lens, respectively, to obtain the seamless full-view complete images under the "10×" objective lens and the "40×" objective lens, respectively. In this embodiment, the algorithm used by the sorting unit is the Hill sorting algorithm.

[0052] In order to automatically complete the entire image acquisition and stitching process, the microscopic image intelligent stitching system also includes a control unit, which is used to control the automatic microscope acquisition platform, the microscopic image acquisition unit and the image stitching unit. The control unit can ensure that each time the lens of the objective lens is switched, it will move within the coordinate range according to the same movement mode, and the field of view image acquisition of the "10×" objective lens, the "40×" objective lens and the "100×" objective lens is performed according to the same movement mode, thereby ensuring the consistency of the observation process.

[0053] In order to facilitate the storage of images, the microscopic image intelligent stitching system further includes a storage unit, which is used to store the collected microscopic images and the seamless full-view complete images obtained after stitching.

[0054] In order to display the spliced ​​image for easy viewing, the microscopic image intelligent splicing system further includes a display unit, which is used to display a complete image with no seams and full field of view.

[0055] The working principle of this embodiment is as follows:

[0056] See also Figures 1 to 5 , place the blood smear on the automatic microscope acquisition platform, the automatic microscope acquisition platform first uses a "100×" objective lens to scan the entire blood smear in an "S-shaped" manner, the positioning unit calculates the node coordinates of each microscopic image under the "100×" objective lens according to the coordinate system preset by the automatic microscope acquisition platform, the sorting unit splices the images with adjacent coordinates with traces, and reserves a 15-micron overlapping area (i.e., the seam area) to obtain a full-field complete image with seams under the "100×" objective lens, and then repairs the seam area of ​​the full-field complete image with traces spliced ​​under the "100×" objective lens through wavelet transform image fusion technology to obtain a full-field complete image without seams under the "100×" objective lens.

[0057] Afterwards, the automatic microscope acquisition platform switches to the "40×" objective lens through the objective lens converter, and scans the entire blood smear in the same "S-shaped" manner. The positioning unit calculates the node coordinates of each microscopic image under the "40×" objective lens according to the coordinate system preset by the automatic microscope acquisition platform. Then, the lens converter is moved to the "100×" objective lens, and the sorting unit compares the visual field similarity according to the estimated position and performs the "40×" objective lens visual field with trace splicing in the same way, and reserves a 15-micron overlapping area (i.e., the seam area) to obtain a complete image of the full field of view with seams under the "40×" objective lens. Afterwards, the positioning unit obtains the target coordinates and coordinate range under the "100×" objective lens according to the coordinate conversion and coordinate correction of the "40×" objective lens and the "100×" objective lens, and compares the cells at the joint. The positioning unit analyzes the captured image within the coordinate range under the "100×" objective lens. When the clearest image is analyzed, the exact coordinate position of the clearest image is recorded, and the clearest image is transmitted to the sorting unit. The sorting unit accurately compares, fills, and splices the images through a preset program, thereby repairing the joint area of ​​the full-field complete image with joints under the "40×" objective lens, and obtains a full-field complete image without joints under the "40×" objective lens.

[0058] Afterwards, the automatic microscope acquisition platform is switched to the "10×" objective lens through the objective lens converter, and the traced stitching is performed in the same way as the "40×" objective lens field of view, obtaining a complete image of the full field of view with a trace under the "10×" objective lens. Afterwards, the same repair method as the "40×" objective lens is used to repair the traced full field of view complete image under the "10×" objective lens based on the complete image of the full field of view without a trace under the "100×" objective lens or the "40×" objective lens, and obtain a complete image of the full field of view without a trace under the "10×" objective lens.

[0059] In this process, the storage unit stores the seamless full-view complete image, and the display unit can display the cell image in the seamless full-view complete image.

[0060] When the stitching work is completed, the image stitching unit can send an initialization signal through a preset program. After receiving the initialization signal, the control unit initializes all units to restore the state to be analyzed, which is convenient for the next cell analysis work.

[0061] The current film reading method has the following problems and impacts: the large number of images taken and stored leads to low reading efficiency and large storage space, information fragmentation leads to inconvenient management, incomplete edge cells in the image lead to the inability to identify edge cells, resulting in missed detection, false detection and other events, and cells with larger diameters cannot be fully presented under a high-power microscope, which increases the difficulty for doctors to read and diagnose. Based on these pain points, in this embodiment, the staff only needs to place the slide with the target cells on the observation position of the automatic microscope acquisition platform. The system will automatically perform accurate comparison, filling, correction and splicing of the field of view, and output the best imaging image of the target cell for display and storage. A complete image with a full field of view without seams is formed through graded filling, repair, correction and other methods; data can be traced, and data information management is realized to avoid misdiagnosis or missed diagnosis caused by missing cells. On the basis of existing technologies, the true image of the cell is accurately restored, greatly improving the efficiency of clinical film reading. It can reduce the missed detection and false detection rates,

[0062] The above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A microscopic image intelligent stitching system, characterized in that: include A microscopic image acquisition unit, used for acquiring microscopic images of the entire field of view of the slide under objective lenses of multiple magnifications; as well as The image stitching unit is used to first stitch and repair the microscopic image of the entire field of view of the slide collected under the highest magnification objective lens, so as to obtain a seamless full-field complete image under the highest magnification objective lens; then stitch and repair the microscopic images under the remaining magnification objective lenses according to the seamless full-field complete image under the highest magnification objective lens, so as to obtain a seamless full-field complete image under the remaining magnification objective lenses.

2. The microscopic image intelligent stitching system according to claim 1, characterized in that: The image stitching unit includes a positioning unit and a sorting unit, wherein the positioning unit is used to calculate the pan-tilt coordinates of the target cells of each microscopic image under the highest magnification objective lens according to the coordinate system preset by the automatic microscope acquisition platform, and to position each microscopic image under the highest magnification objective lens according to the pan-tilt coordinates; and respectively calculating the pan-tilt coordinates of the target cells in each microscopic image under each non-highest magnification objective lens, and converting them to the platform coordinates of the target cells in the corresponding microscopic image under the high-magnification objective lens for positioning, and judging the clarity of the image under the high-magnification objective lens, and correcting the coordinates by image similarity comparison; The sorting unit is used to splice the microscopic images of the entire field of view of the slide collected under the highest magnification objective lens according to the pan-tilt coordinates of the target cells of each microscopic image under the highest magnification objective lens, so as to obtain a complete image of the entire field of view with a joint under the highest magnification objective lens, wherein the complete image of the entire field of view with a joint forms a joint area between two adjacent spliced ​​microscopic images; and to repair the joint area of ​​the complete image of the entire field of view with a joint under the highest magnification objective lens by using the wavelet transform image fusion technology, so as to obtain a complete image of the entire field of view without a joint under the highest magnification objective lens; And use an algorithm to compare the coordinate position and image similarity of each microscopic image under each non-highest magnification objective lens with the seamless full-view complete image under a higher magnification objective lens, and splice the areas with adjacent coordinates and consistent image comparisons with traces to form a seamless full-view complete image under the objective lens of that magnification; then, the joint area of ​​the seamless full-view complete image under each non-highest magnification objective lens is corrected using the seamless full-view complete image under a higher magnification objective lens, so as to obtain the seamless full-view complete images under each non-highest magnification objective lens.

3. The microscopic image intelligent stitching system according to claim 2, characterized in that: The positioning unit uses a neural network model to perform image similarity comparison.

4. The microscopic image intelligent stitching system according to claim 2, characterized in that: The algorithm adopted by the sorting unit is the Shell sorting algorithm.

5. The microscopic image intelligent stitching system according to claim 1, characterized in that: The multiple magnification objective lenses include a "10×" objective lens, a "40×" objective lens and a "100×" objective lens, the highest magnification objective lens is a "100×" objective lens, and the non-highest magnification objective lenses include a "10×" objective lens and a "40×" objective lens.

6. The microscopic image intelligent stitching system according to claim 1, characterized in that: The joint area is an overlapping area formed at the joint of two adjacent microscopic images.

7. The microscopic image intelligent stitching system according to any one of claims 1 to 6, characterized in that: The microscopic image acquisition unit is connected to an automatic microscope acquisition platform, and the automatic microscope acquisition platform is provided with an objective lens converter, and the objective lens converter is used to automatically replace the objective lens on the automatic microscope acquisition platform.

8. The microscopic image intelligent stitching system according to claim 7, characterized in that: It also includes a control unit, which is used to control the automatic microscope acquisition platform, the microscopic image acquisition unit and the image stitching unit.

9. The microscopic image intelligent stitching system according to any one of claims 1 to 6, characterized in that: It also includes a storage unit, which is used to store the collected microscopic images and the seamless full-view complete images obtained after splicing.

10. The microscopic image intelligent stitching system according to claim 9, characterized in that: It also includes a display unit, which is used to display a seamless full-view complete image.

Images