CN110738658B - Image quality evaluation method - Google Patents
Image quality evaluation method Download PDFInfo
- Publication number
- CN110738658B CN110738658B CN201911331258.9A CN201911331258A CN110738658B CN 110738658 B CN110738658 B CN 110738658B CN 201911331258 A CN201911331258 A CN 201911331258A CN 110738658 B CN110738658 B CN 110738658B
- Authority
- CN
- China
- Prior art keywords
- quality
- image
- power
- images
- pathological
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention relates to an image quality evaluation method for evaluating the quality of pathological images, which comprises the following steps: judging whether the pathological image is a digital pathological image obtained by scanning or a real-time acquired image obtained by a camera, when the pathological image is the digital pathological image, judging the imaging quality of the image, if the imaging quality is poor, ending the image quality evaluation, and outputting a judgment result; otherwise, judging the content quality of the image and outputting a judgment result; and when the pathological image is a real-time collected image, judging the content quality of the image and outputting a judgment result. The method and the device for judging the quality of the digital pathological image simulation medical expert firstly judge the scanning quality, directly define the pathological image with unqualified obvious scanning quality as an unqualified image, and do not judge the content quality, thereby improving the working efficiency of a quality control link.
Description
Technical Field
The invention relates to the field of image evaluation, in particular to an image quality evaluation method.
Background
The image quality refers to the evaluation of human visual perception of images and also refers to the degree of information provided by images to doctors, and for a medical image processing system, the main body of information is images, and the main index for measuring the system is the image quality. The medical images serve for clinical diagnosis, and a subjective evaluation method is mostly adopted, but the method needs repeated experiments for many times by an organizer, is long in time consumption and high in cost, is easily influenced by the knowledge background, observation purposes, environment and the like of the doctor, is poor in stability and transportability, is difficult to express by a mathematical model, and cannot be widely popularized and applied.
In addition, the combination of artificial intelligence and pathological diagnosis, namely artificial intelligence auxiliary interpretation, is a new trend of improving the efficiency and accuracy of pathological diagnosis in the current medical field, and the quality of pathological images can influence the operation cost of a server and further influence the efficiency of auxiliary interpretation and the reliability of algorithm results, so that how to ensure the quality of input pathological images is also the technical problem which is mainly solved by artificial intelligence auxiliary interpretation.
Disclosure of Invention
The invention aims to apply artificial intelligence to the quality evaluation of pathological images so as to improve the rapidity and the accuracy of the quality evaluation of the pathological images.
The invention realizes the purpose through the following technical scheme: an image quality evaluation method for evaluating the quality of a pathological image, comprising: judging whether the pathological image is a digital pathological image obtained by scanning or a real-time acquired image obtained by a camera, when the pathological image is the digital pathological image, judging the imaging quality of the image, if the imaging quality is poor, ending the image quality evaluation, and outputting a judgment result; otherwise, judging the content quality of the image and outputting a judgment result; and when the pathological image is a real-time collected image, judging the content quality of the image and outputting a judgment result.
Further, the method for judging the imaging quality of the digital pathological image specifically comprises the following steps: the method comprises the steps of carrying out first image sampling on the digital pathological image, collecting a low-power image sample under the magnification of m1, dividing the low-power image sample into a plurality of segmentation images with the same size, inputting the segmentation images with the same size into a low-power quality judgment model for imaging quality judgment, outputting the imaging quality score of each segmentation image by the low-power quality judgment model, judging the imaging quality of the low-power image sample according to the imaging quality score, and outputting a judgment result.
Further, the method for performing quality judgment by the low power quality judgment model comprises the following steps: taking the minimum value from the quality scores of each slice image, and judging the digital pathological image as unqualified scanning quality if the minimum value is less than a threshold value a; otherwise, the scanning quality is qualified.
Further, the method for judging the content quality of the digital pathological image specifically includes: and carrying out second image sampling on the digital pathological image, randomly acquiring a plurality of high-power image samples with set sizes from m2 multiplying power, inputting the high-power image samples into a high-power quality judgment model for content quality judgment, and outputting a judgment result.
Further, the method for performing quality judgment by the high power quality judgment model comprises the following steps: inputting the multiple high-power image samples into a high-power quality judgment model, outputting the quality score of each high-power image sample by the high-power quality judgment model, averaging the obtained quality scores, and judging the digital pathological image to be qualified in content quality if the obtained average value is less than a threshold value b 1; otherwise, the content quality is unqualified.
Further, the method for judging the content quality of the real-time acquired image specifically includes: acquiring images of the real-time acquired images, inputting the acquired real-time acquired images into a high-power quality judgment model for content quality judgment, outputting quality scores of the images by the high-power quality judgment model, and judging the real-time acquired images to be qualified in content quality if the scores of the real-time acquired images are smaller than a threshold value b 3; otherwise, the content quality is unqualified.
Further, the training process of the low power quality judgment model comprises: s10, training data are collected: acquiring a plurality of digital pathology full-field pictures, performing low-power sampling on the digital pathology full-field pictures to acquire digital pathology images with the magnification of m1, adjusting the acquired digital pathology images to be the same size, cutting each digital pathology image into n1 sub-images with the same size, and dividing all the sub-images into low-power positive samples and low-power negative samples according to the imaging quality; s11, training a deep learning model: and establishing a convolutional neural network as a first slice quality judger, and training by adopting a low-power positive sample and a low-power negative sample to obtain a low-power quality judgment model.
Further, the training process of the high power quality judgment model comprises: s20, training data are collected: acquiring a plurality of digital pathology whole-field images, performing high-power sampling on the digital pathology whole-field images to obtain digital pathology images with the magnification of m2, dividing each digital pathology image into n2 divided images with the same size, and dividing all the divided images into high-power positive samples and high-power negative samples according to the content quality; s21, training a deep learning model: and establishing a convolutional neural network as a second slice quality judger, and training by adopting a high-power positive sample and a high-power negative sample to obtain a high-power quality judgment model.
Compared with the prior art, the invention has the following substantial effects: according to the method, the quality control link of the pathological images is placed before image processing, subsequent artificial intelligence auxiliary interpretation is not performed on the pathological images which are unqualified in evaluation, the operation cost of a server is reduced, and the efficiency of intelligent analysis and the reliability of algorithm results are improved; (2) for the judgment process of the digital pathological image simulation medical expert, the scanning quality is judged firstly, the pathological image with unqualified obvious scanning quality is directly defined as the unqualified image, and the content quality is not judged, so that the working efficiency of a quality control link is improved.
Drawings
Fig. 1 is a flowchart of an image quality evaluation method according to the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings in which:
the pathological image in this embodiment is a high-resolution digital pathological image acquired by scanning a pathological slide with a scanner, and is a real-time acquired image acquired by acquiring a pathological section placed under a microscope in real time with a microscope camera mounted on the microscope.
The image quality evaluation method is used for evaluating the image quality of pathological images, for digital pathological images, the image quality comprises the imaging quality and the content quality, the imaging quality comprises the defects of whether pathological slide scanning is distorted, whether a large amount of stains, dust, bubbles, water drops and the like exist on a section, whether an effective tissue area is contained, and the content quality comprises the defects of whether cell cavities exist in the section, few cells, much blood and the like. Under the low power visual field, the range of the observed digital pathological image is large, which is beneficial to judging the image quality on the whole, but the details of the cell level can not be observed, so that the digital pathological image with qualified imaging quality needs to be converted into the high power visual field so as to judge the content quality of the pathological image more pertinently.
As shown in fig. 1, after receiving a pathological image, it is first determined whether the pathological image is a digital pathological image or a real-time captured image. If the image is a digital pathological image, firstly judging the imaging quality of the image: performing first image sampling on a digital pathological image, acquiring a low-power image sample under m1 magnification (for example, under 1.25 magnification), dividing the low-power image sample into a plurality of divided images with the same size, inputting the plurality of divided images with the same size into a low-power quality judgment model for imaging quality judgment, outputting an imaging quality score of each slice image by the low-power quality judgment model, taking a minimum value from the quality scores of each slice image, judging the digital pathological image to be unqualified in scanning quality if the minimum value is less than a threshold value a, ending the imaging quality evaluation, and outputting a judgment result; otherwise, the content quality of the image is further judged for the qualified scanning quality. Performing second image sampling on the digital pathological image, randomly acquiring a plurality of high-power image samples with set sizes from m2 multiplying power, inputting the plurality of high-power image samples into a high-power quality judgment model for content quality judgment, outputting the quality score of each high-power image sample by the high-power quality judgment model, averaging the plurality of obtained quality scores, and judging the digital pathological image to be qualified in content quality if the average value is less than a threshold b 1; otherwise, the content quality is unqualified. Wherein the threshold a and the threshold b1 are selected according to experiments and different scenes of quality level requirements.
For the pathological section directly observed under the microscope, the picture of the pathological section under the microscope lens can be collected through the microscope camera arranged under the microscope, the collected pathological image is called as a real-time collected image in the application, and in consideration of the practical work, when a doctor observes the pathological section under the microscope, the microscope is usually adjusted to be suitable for the observed high-power visual field, and the collected real-time collected image is adapted to the size of the training sample of the high-power quality judgment model. Therefore, for the real-time collected image, the acquired picture of the pathological section under the current microscope visual field under the high-power visual field is acquired by default, namely the acquired pathological image under the high-power visual field is acquired, the acquired real-time collected image is input into the high-power quality judgment model for content quality judgment, the high-power quality judgment model outputs the quality score of the real-time collected image, and if the quality score of the real-time collected image is smaller than a threshold value b3, the real-time collected image is judged to be qualified in content quality; otherwise, the content quality is unqualified. The threshold b3 is selected according to experiment and the required scene of quality level.
When carrying out the low power quality judgment model and the high power quality judgment model training, the method comprises the following steps:
1. collecting training data: the training data set contains about one thousand full-field slices of good or poor quality digital pathology.
a) And performing low-power sampling on all the digital full-field slices to obtain a full-field image with the magnification of 1.25 (the side length of the general image is between 2000 and 3000 pixels), then scaling the obtained full-field image to 2048 × 2048 pixels, then cutting the full-field image into a plurality of cut images of 1024 × 1024 pixels, finally dividing all the cut images into a low-power positive sample and a low-power negative sample according to the imaging quality of the cut images, using 1 to represent the positive sample, and using 0 to represent the negative sample.
b) High-power sampling is carried out on all digital full-field slices to obtain full-field images with the magnification of 20 times (the side length of a general image is between 30000 and 50000 pixels), each full-field image is segmented into a plurality of segmented images with the size of 1024 x 1024 pixels, finally the segmented images are divided into high-power positive samples and high-power negative samples according to the content quality, 1 represents the positive samples, and 0 represents the negative samples.
2. Training a deep learning model: establishing two Convolutional Neural Networks (CNN) with the same structure as a slice quality judger, and respectively training by using high-power sampled training data and low-power sampled training data, wherein the obtained models are a high-power quality judgment model and a low-power quality judgment model. The model consists of convolutional layer, pooling layer, activation function, BN (BatchNormal) layer, full-link layer and jump link. The convolution layer is used for coding the characteristics of the upper layer; the activation function produces a non-linear transformation; the pooling layer reduces the dimension of the characteristic diagram, so that the parameter quantity of the network is reduced, and the training and reasoning of the model are accelerated; the full connection layer is used for integrating and classifying the obtained features; the hopping connection is used to transfer the preceding feature map directly via a shortcut into a deeper network structure than via the main path. The layers and functions are organized by a certain design structure to form the neural network structure used in the invention. The input of the neural network is an image of 1024 x 1024 pixels, and the output is a fraction between 0 and 1. A score closer to 1 represents better image quality, whereas a score closer to 0 represents worse image quality.
The slice quality judging method simulates the process of human expert judgment, and for digital pathological images, whether slice scanning is distorted, whether a large amount of stains, dust, bubbles and water drops exist on the slices, whether the slices contain effective tissue areas and the like are roughly judged from a low-power visual field. If the slice does not contain the above-mentioned problems in the low-magnification visual field, the slice is further sampled and judged from the high-magnification visual field because details at the cellular level, such as cellular cavities, few cells, much blood, etc., cannot be observed in the low-magnification visual field. The efficiency and the accuracy of pathological image quality judgment are improved through step-by-step judgment of scanning quality and content quality, and preparation is made for artificial intelligent auxiliary interpretation.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed.
Claims (7)
1. An image quality evaluation method for evaluating quality of a pathological image, comprising:
judging whether the pathological image is a digital pathological image obtained by scanning or a real-time acquired image obtained by acquiring the picture of the pathological section under the microscope lens under a high-power visual field through a microscope camera arranged on a microscope;
when the pathological image is a digital pathological image, judging the imaging quality of the image, if the imaging quality is poor, finishing the image quality evaluation, and outputting a judgment result; otherwise, judging the content quality of the image, performing second image sampling on the digital pathological image, inputting a high-power quality judgment model for content quality judgment, and outputting a judgment result by the high-power quality judgment model;
when the pathological image is a real-time collected image, inputting the collected real-time collected image into a high-power quality judgment model for content quality judgment, and outputting a judgment result by the high-power quality judgment model; the collected real-time collected image is adaptive to the size of a training sample of the high-power quality judgment model;
the method for judging the imaging quality of the digital pathological image specifically comprises the following steps: the method comprises the steps of carrying out first image sampling on the digital pathological image, collecting a low-power image sample under the magnification of m1, dividing the low-power image sample into a plurality of segmentation images with the same size, inputting the segmentation images with the same size into a low-power quality judgment model for imaging quality judgment, outputting the imaging quality score of each segmentation image by the low-power quality judgment model, judging the imaging quality of the low-power image sample according to the imaging quality score, and outputting a judgment result.
2. The image quality evaluation method according to claim 1, wherein the method of performing quality judgment by the low power quality judgment model includes:
taking the minimum value from the quality scores of each obtained segmentation image, and judging the digital pathological image as unqualified scanning quality if the minimum value is less than a threshold value a; otherwise, the scanning quality is qualified.
3. The image quality evaluation method according to claim 1, wherein the method for determining the content quality of the digital pathology image specifically includes: and carrying out second image sampling on the digital pathological image, randomly acquiring a plurality of high-power image samples with set sizes from m2 multiplying power, inputting the high-power image samples into a high-power quality judgment model for content quality judgment, and outputting a judgment result.
4. The image quality evaluation method according to claim 3, wherein the method of performing quality judgment by the high power quality judgment model includes:
inputting the multiple high-power image samples into a high-power quality judgment model, outputting the quality score of each high-power image sample by the high-power quality judgment model, averaging the obtained quality scores, and judging the digital pathological image to be qualified in content quality if the obtained average value is less than a threshold value b 1; otherwise, the content quality is unqualified.
5. The image quality evaluation method according to claim 1, wherein the method for determining the content quality of the real-time captured image specifically comprises: acquiring images of the real-time acquired images, inputting the acquired real-time acquired images into a high-power quality judgment model for content quality judgment, outputting quality scores of the images by the high-power quality judgment model, and judging the real-time acquired images to be qualified in content quality if the scores of the real-time acquired images are smaller than a threshold value b 3; otherwise, the content quality is unqualified.
6. The image quality evaluation method according to claim 2, wherein the training process of the low power quality determination model includes:
s10, training data are collected:
acquiring a plurality of digital pathology full-field pictures, performing low-power sampling on the digital pathology full-field pictures to acquire digital pathology images with the magnification of m1, adjusting the acquired digital pathology images to be the same size, cutting each digital pathology image into n1 sub-images with the same size, and dividing all the sub-images into low-power positive samples and low-power negative samples according to the imaging quality;
s11, training a deep learning model:
and establishing a convolutional neural network as a first slice quality judger, and training by adopting a low-power positive sample and a low-power negative sample to obtain a low-power quality judgment model.
7. The image quality evaluation method according to claim 3, 4 or 5, wherein the training process of the high power quality determination model includes:
s20, training data are collected:
acquiring a plurality of digital pathology whole-field images, performing high-power sampling on the digital pathology whole-field images to obtain digital pathology images with the magnification of m2, dividing each digital pathology image into n2 divided images with the same size, and dividing all the divided images into high-power positive samples and high-power negative samples according to the content quality;
s21, training a deep learning model:
and establishing a convolutional neural network as a second slice quality judger, and training by adopting a high-power positive sample and a high-power negative sample to obtain a high-power quality judgment model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911331258.9A CN110738658B (en) | 2019-12-21 | 2019-12-21 | Image quality evaluation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911331258.9A CN110738658B (en) | 2019-12-21 | 2019-12-21 | Image quality evaluation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110738658A CN110738658A (en) | 2020-01-31 |
| CN110738658B true CN110738658B (en) | 2020-09-15 |
Family
ID=69274559
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911331258.9A Active CN110738658B (en) | 2019-12-21 | 2019-12-21 | Image quality evaluation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110738658B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111986157B (en) * | 2020-07-21 | 2024-02-09 | 万达信息股份有限公司 | Digital pathological image quality evaluation system |
| CN113962976B (en) * | 2021-01-20 | 2022-09-16 | 赛维森(广州)医疗科技服务有限公司 | Quality evaluation method for pathological slide digital image |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070031043A1 (en) * | 2005-08-02 | 2007-02-08 | Perz Cynthia B | System for and method of intelligently directed segmentation analysis for automated microscope systems |
| CN109271969B (en) * | 2018-10-29 | 2020-03-24 | 北京青燕祥云科技有限公司 | Brain glioma grading evaluation method and device |
| CN109859232A (en) * | 2019-01-11 | 2019-06-07 | 安徽师范大学 | A kind of animal brain's slice microscopic image segmentation |
| CN110376198B (en) * | 2019-08-29 | 2021-08-10 | 广州锟元方青医疗科技有限公司 | Cervical liquid-based cell slice quality detection system |
-
2019
- 2019-12-21 CN CN201911331258.9A patent/CN110738658B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110738658A (en) | 2020-01-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220343623A1 (en) | Blood smear full-view intelligent analysis method, and blood cell segmentation model and recognition model construction method | |
| WO2021139258A1 (en) | Image recognition based cell recognition and counting method and apparatus, and computer device | |
| CN107665492B (en) | Colorectal panoramic digital pathological image tissue segmentation method based on depth network | |
| JP6900581B1 (en) | Focus-weighted machine learning classifier error prediction for microscope slide images | |
| CN108830149B (en) | Target bacterium detection method and terminal equipment | |
| CN110736748A (en) | Immunohistochemical nuclear plasma staining section diagnosis method and system | |
| CN110763678A (en) | Pathological section interpretation method and system | |
| CN111488921A (en) | Panoramic digital pathological image intelligent analysis system and method | |
| CN110738658B (en) | Image quality evaluation method | |
| CN109903282B (en) | Cell counting method, system, device and storage medium | |
| WO2017221592A1 (en) | Image processing device, image processing method, and image processing program | |
| CN110647875A (en) | Method for segmenting and identifying model structure of blood cells and blood cell identification method | |
| Ghazali et al. | Automated system for diagnosis intestinal parasites by computerized image analysis | |
| CN110807757A (en) | Image quality evaluation method and device based on artificial intelligence and computer equipment | |
| CN113781455B (en) | Cervical cell image anomaly detection method, device, equipment and medium | |
| CN113344958B (en) | Microscopic imaging scanning method and scanning system | |
| CN109886170B (en) | Intelligent detection, identification and statistics system for oncomelania | |
| CN110763677A (en) | Thyroid gland frozen section diagnosis method and system | |
| CN112464802B (en) | Automatic identification method and device for slide sample information and computer equipment | |
| CN110728666A (en) | Typing method and system for chronic nasosinusitis based on digital pathological slide | |
| CN113237881B (en) | Detection method and device for specific cells and pathological section detection system | |
| CN114419619B (en) | Erythrocyte detection and classification method and device, computer storage medium and electronic equipment | |
| CN111339899B (en) | Catheter feature acquisition method, device, equipment, medium and intelligent microscope | |
| CN115294191B (en) | Marker size measuring method, device, equipment and medium based on electronic endoscope | |
| CN115359031A (en) | Digital pathological image slice quality evaluation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |