CN-122018137-A - Digital pathological section automatic focusing method and device based on five-point heterogeneous constraint

CN122018137ACN 122018137 ACN122018137 ACN 122018137ACN-122018137-A

Abstract

The technical scheme provides a digital pathological section automatic focusing method and device based on five-point heterogeneous constraint, which are characterized in that a low-magnification objective lens lower than a target magnification is used for obtaining a low-magnification preview image of a target section, tissue areas in the low-magnification preview image are segmented, coarse focusing is carried out on the low-magnification preview image to obtain a coarse focal plane model, high-risk areas in the coarse focal plane model are divided based on the coarse focal plane model, fine focusing is carried out on each high-risk area to obtain a plurality of fine focal plane models, focal planes of non-high-risk areas are used as a first focal plane set, the fine focal plane model of the high-risk areas is used as a second focal plane set, and an automatic focusing path of the target section is obtained based on the first focal plane set and the second focal plane set. According to the scheme, five focusing points which are used for achieving overall focal plane reference, scanning direction trend, transverse boundary prediction precision and local focal plane mutation compensation are selected in coarse focusing and fine focusing, the defect of insufficient adaptation of low-order surface fitting to local abnormal areas is overcome, and accurate acquisition of an automatic focusing path is achieved.

Inventors

WANG ZIHAN
HUANG QIANG
LIANG CHAOBIN
CHEN BAICHUAN
KUANG GUOTAO
JIN JIE

Assignees

深圳市生强科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The digital pathological section automatic focusing method based on five-point heterogeneous constraint is characterized by comprising the following steps of: Acquiring a low-power preview image of a target slice by using a low-power objective lens lower than a target power, and dividing a tissue region in the low-power preview image, wherein the target power is a rated imaging power meeting the clinical definition requirement; Selecting five coarse focusing points on the low-power preview image, respectively focusing at the five coarse focusing points by using a low-power objective lens to obtain five coarse focus coordinates, and fitting the five coarse focus coordinates to obtain a coarse focus model; Dividing a low-power preview image into a plurality of focusing subareas to form a focusing subarea set, wherein the range of each focusing subarea corresponds to an objective view field under a target multiplying power, taking the focusing subareas comprising a tissue area in the focusing subarea set as local modeling units, acquiring a local focal plane formed by a plurality of continuous coarse focusing points in each local modeling unit in a coarse focal plane model, taking the local modeling units corresponding to the local focal planes meeting a high-risk area screening rule as a high-risk area, and taking the rest local modeling units as non-high-risk areas; Selecting five fine focus points in each high-risk area, focusing the objective lens under the target multiplying power at the five fine focus points to obtain five fine focus coordinates, and fitting the five fine focus coordinates to obtain a fine focus surface model of the corresponding high-risk area; And acquiring a focal plane of each non-high risk area in the focusing subarea set in the coarse focal plane model as a first focal plane set, and acquiring a focal plane of each high risk area in the focusing subarea set in the corresponding fine focal plane model as a second focal plane set, and acquiring an automatic focusing path of the target slice based on the first focal plane set and the second focal plane set.
2. The method for automatically focusing digital pathological sections based on five-point heterogeneous constraints according to claim 1, wherein the high-risk area screening rule is that the focal confidence of the rough focusing point of the local focal plane is lower than a first threshold, or the focal fitting residual of the rough focusing point of the local focal plane is larger than a second threshold, or the focusing height change gradient of the local focal plane and the adjacent local focal plane is larger than a third threshold.
3. The automatic focusing method for the digital pathological section based on five-point heterogeneous constraint according to claim 2 is characterized by comprising the steps of inputting a local focal plane into a pre-trained focal confidence prediction model to obtain focal confidence of each coarse focal point, respectively calculating differences between each coarse focal point in the local focal plane and five coarse focal points in a low-power preview image on focusing heights, selecting a result with the largest difference as a fitting residual of the corresponding coarse focal point, and taking the largest focusing height difference between the current local focal plane and an adjacent local focal plane as a focusing height change gradient of the current local focal plane.
4. The automatic focusing method of digital pathological section based on five-point heterogeneous constraint according to claim 1, wherein the tissue type of each local modeling unit is classified based on a tissue type recognition model, the local modeling units comprising the tissue type as a dead zone, a mucus zone and a fat zone are used as high risk areas based on the classification result, and the local modeling units comprising the junction of the tissue area and a background part are used as high risk areas.
5. The five-point heterogeneous constraint-based automatic focusing method for the digital pathological section is characterized in that five coarse focusing points are a center representative point, a scanning forward prediction point, a first transverse boundary point, a second transverse boundary point and a risk compensation point on a low-power preview image, five fine focusing points are a center representative point, a scanning forward prediction point, a first transverse boundary point, a second transverse boundary point and a risk compensation point on a high-risk area, wherein the center representative point is a point which is in the geometric center range of a corresponding image and is positioned on a tissue area, the scanning forward prediction point is a point which is in the tissue area with highest tissue continuity in the scanning direction of the corresponding image, the first transverse boundary point is a point which is in the tissue area and has the smallest X-axis coordinate of the corresponding image, the second transverse boundary point is a point which is in the tissue area and has the largest X-axis coordinate of the corresponding image, and the risk compensation point is a point in the tissue area with highest abnormal probability of a focal plane.
6. The method for automatically focusing a digital pathological section based on five-point heterogeneous constraint according to claim 1, wherein the objective lens with target magnification moves along the Z axis at each fine focusing point with a preset step distance and collects a plurality of intermediate images, the definition of each intermediate image is calculated, and the coordinate position corresponding to the intermediate image with highest definition is the fine focusing point coordinate.
7. The five-point heterogeneous constraint-based digital pathological section automatic focusing method according to claim 1, wherein a fine focal plane model of a current high-risk area is used as a fitting priori of a next adjacent high-risk area, wherein the adjacent relation between the high-risk areas is judged based on a scanning direction, and the fitting priori is used for constraining a fitting process of the fine focal plane model of the next adjacent high-risk area.
8. Digital pathological section automatic focusing device based on five heterogeneous constraints, characterized by comprising: The acquisition module is used for acquiring a low-power preview image of a target slice by using a low-power objective lens lower than a target power, and dividing a tissue region in the low-power preview image, wherein the target power is a rated imaging power meeting clinical definition requirements; the coarse focusing module is used for selecting five coarse focusing points on the low-power preview image, focusing the five coarse focusing points by the low-power objective lens to obtain five coarse focus coordinates, and fitting the five coarse focus coordinates to obtain a coarse focus model; The risk judging module is used for dividing the low-power preview image into a plurality of focusing subareas to form a focusing subarea set, wherein the range of each focusing subarea corresponds to the field of view of the objective lens under the target multiplying power, the focusing subareas comprising the tissue area in the focusing subarea set are used as local modeling units, the local focal plane formed by a plurality of continuous coarse focusing points in each local modeling unit is obtained in a coarse focal plane model, the local modeling units corresponding to the local focal plane meeting the high-risk area screening rule are taken as high-risk areas, and the rest local modeling units are non-high-risk areas; The fine focusing module is used for selecting five fine focusing points in each high-risk area, focusing the objective lens under the target multiplying power at the five fine focusing points to obtain five fine focus coordinates, and fitting the five fine focus coordinates to obtain a fine focus model corresponding to the high-risk area; The automatic focusing path acquisition module is used for acquiring a focal plane of each non-high risk area in the focusing subarea set in the coarse focal plane model as a first focal plane set, acquiring a focal plane of each high risk area in the focusing subarea set in the corresponding fine focal plane model as a second focal plane set, and acquiring an automatic focusing path of the target slice based on the first focal plane set and the second focal plane set.
9. An electronic device comprising a memory and a processor, wherein the memory has stored therein a computer program, the processor being arranged to run the computer program to perform a five-point heterogeneous constraint based digital pathological slice autofocus method according to any of claims 1-7.
10. A readable storage medium, wherein a computer program is stored in the readable storage medium, and when the computer program is executed by a processor, the computer program realizes a five-point heterogeneous constraint-based digital pathological section automatic focusing method according to any one of claims 1 to 7.

Description

Digital pathological section automatic focusing method and device based on five-point heterogeneous constraint Technical Field The application relates to the field of digital pathological section scanning, in particular to a digital pathological section automatic focusing method and device based on five-point heterogeneous constraint. Background The core target of digital pathological section scanning is to convert pathological tissue samples on a glass slide into panoramic digital images with high definition and high consistency through a high-magnification microscopic optical system. Under the scanning scene of 20 times, 40 times and higher multiplying power commonly used in clinic, the depth of field of the microscope objective is extremely shallow, and the optimal focal point height of different space positions of the same slice can be obviously changed due to a plurality of factors such as thickness difference of pathological tissues, inclination of a cover glass, uneven distribution of sealing medium, clamping error of the glass slide, flatness deviation of a moving platform, thermal drift of a mechanism in the running process of equipment and the like. At present, the main flow technical scheme in the field mainly comprises four major categories of single-point focusing, multi-point plane fitting, multi-point curved surface fitting and continuous focus following in the scanning process, but the scheme has the following defects in practical application that firstly, the focusing time can be obviously increased although the fitting precision can be improved by simply increasing the number of focusing points, the whole scanning flux is influenced, secondly, the common multi-point fitting is usually defaulting to have equal reliability of all focusing points and is easily interfered by blank areas, fold areas, bubble areas, uneven dyeing areas and tissue edge warp areas, thirdly, the texture characteristics and focusing reliability of different pathological tissue categories have obvious differences, if uniform weights are adopted, the focal surface modeling result is easily influenced excessively, fourthly, the partial scheme only carries out once global fitting, the condition that the whole slow change and the partial mutation of the tissue surface coexist is difficult to be considered, and fifth, the historical focal surface information in the scanning direction is not fully utilized, so that each partial area needs repeated complete modeling, and the efficiency is lower. Therefore, there is a need for an auto-focus method that achieves high robustness and high efficiency with a reduced number of focus points and that can accommodate differences in different pathological tissue categories. Disclosure of Invention The embodiment of the application provides a five-point heterogeneous constraint-based digital pathological section automatic focusing method and device, which are used for selecting five focusing points which are compatible with overall focal plane reference, scanning direction trend, transverse boundary prediction precision and local focal plane mutation compensation in coarse focusing and fine focusing, so that the defect of insufficient adaptation of low-order curved surface fitting to a local abnormal region is overcome, and the accurate acquisition of an automatic focusing path is realized. In a first aspect, an embodiment of the present application provides a digital pathological section auto-focusing method based on five-point heterogeneous constraint, where the method includes: Acquiring a low-power preview image of a target slice by using a low-power objective lens lower than a target power, and dividing a tissue region in the low-power preview image, wherein the target power is a rated imaging power meeting the clinical definition requirement; Acquiring a low-power preview image of a target slice by using a low-power objective lens lower than a target power, and dividing a tissue region in the low-power preview image, wherein the target power is a rated imaging power meeting the clinical definition requirement; Selecting five coarse focusing points on the low-power preview image, respectively focusing at the five coarse focusing points by using a low-power objective lens to obtain five coarse focus coordinates, and fitting the five coarse focus coordinates to obtain a coarse focus model; Dividing a low-power preview image into a plurality of focusing subareas to form a focusing subarea set, wherein the range of each focusing subarea corresponds to an objective view field under a target multiplying power, taking the focusing subareas comprising a tissue area in the focusing subarea set as local modeling units, acquiring a local focal plane formed by a plurality of continuous coarse focusing points in each local modeling unit in a coarse focal plane model, taking the local modeling units corresponding to the local focal planes meeting a high-risk area screening rule as a high-risk ar