CN110288542A - A kind of liver's pathological image sample Enhancement Method based on stochastic transformation - Google Patents
A kind of liver's pathological image sample Enhancement Method based on stochastic transformation Download PDFInfo
- Publication number
- CN110288542A CN110288542A CN201910526943.0A CN201910526943A CN110288542A CN 110288542 A CN110288542 A CN 110288542A CN 201910526943 A CN201910526943 A CN 201910526943A CN 110288542 A CN110288542 A CN 110288542A
- Authority
- CN
- China
- Prior art keywords
- image
- liver
- sample
- width
- method based
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/77—Retouching; Inpainting; Scratch removal
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
- G06T2207/30004—Biomedical image processing
- G06T2207/30056—Liver; Hepatic
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
Liver's pathological image sample Enhancement Method based on stochastic transformation that the present invention relates to a kind of comprising following steps: 1) block division is carried out to liver pathological image, obtains several image fritters；2) stochastic transformation is carried out to each image fritter, forms exptended sample；The stochastic transformation includes one or more of horizontal mirror image switch, vertical mirror overturning, cutting, brightness adjustment, saturation degree adjustment and adjustment of color；3) exptended sample input deep learning model is trained liver's pathological image, and carries out corresponding enhancing with liver pathological image, obtain the enhancing sample of liver's pathological image.The present invention can effectively expand original pathology sample, solve the problems, such as that sample size deficiency and sample distribution are uneven to a certain extent, meet the requirement of deep learning model large sample size, the model trained can be effectively avoided over-fitting occur, the situation of generalization ability deficiency improves the reliability of assistant analysis result.
Description
Technical Field
The invention relates to the technical field of digital pathological image processing, in particular to a liver tissue pathological image segmentation method based on an OTSU threshold.
Background
Liver cancer (liver cancer) refers to malignant tumor occurring in liver, including primary liver cancer and metastatic liver cancer, and most of the liver cancer is the primary liver cancer. Primary liver cancer is one of the most common malignant tumors in clinic, and according to the latest statistics, about sixty-hundred thousand new liver cancer patients all year round live in the fifth part of the malignant tumors. There are different effective treatment modes aiming at different cancer cell types and the spread degree of the cancer cells. Currently, in clinical diagnosis, particularly in most of cancer diagnoses, pathological examination results obtained by observing and analyzing a biopsy of a lesion region of a patient in a microscopic field of view through biopsy, exfoliation, fine needle puncture cytology, and the like are important and final evidences for disease diagnosis by doctors, and diagnosis based on pathological images is regarded as "gold standard" for cancer diagnosis.
At present, the diagnosis of pathological images is mainly performed manually by professional doctors, and with the increasing number of patients and the increasing requirements for the accuracy of disease diagnosis, the number of pathological images to be analyzed is multiplied, so that more personnel and equipment are required to be added to meet the requirements of a larger amount of histopathological analysis. However, according to the current state of China, the number of experienced doctors is rare, the levels are irregular, the digital level of a pathology department is generally low, and digital pathology equipment is seriously deficient, which brings great difficulty to the further development of pathological diagnosis. In consideration of the convenience of storage and remote transmission, the digital histopathology images are more and more emphasized, and more importantly, the digital histopathology images provide possibility for introducing intelligent auxiliary analysis to reduce the burden of doctors, and have important significance for relieving the tension situation of medical resources.
The current mainstream intelligent auxiliary analysis means comprises: image feature extraction, deep learning models, etc., and these auxiliary analysis methods often require a large number of data samples. Taking deep learning as an example, firstly, there are training samples as many as possible, and secondly, it is to ensure that the samples are distributed uniformly, but in actual conditions, the situation that the sample amount is not enough may often be encountered, and the training requirements of the deep learning model cannot be met.
Disclosure of Invention
The invention aims to provide a liver pathology image sample enhancement method based on random transformation, which is reasonable in design and can solve the problems of overlarge pathological section images and insufficient sample size in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a liver pathology image sample enhancement method based on random transformation comprises the following steps:
1) carrying out block division on the liver pathological image to obtain a plurality of image small blocks;
2) randomly transforming each image small block to form an extended sample; the random transformation comprises more than one of horizontal mirror image turning, vertical mirror image turning, clipping, brightness adjustment, saturation adjustment and hue adjustment;
3) and inputting the extended sample into a deep learning model to train the liver pathological image, and correspondingly enhancing the liver pathological image to obtain an enhanced sample of the liver pathological image.
Preferably, the size of the image patch in step 1) is 320 × 320 pixels.
Preferably, the method for turning over the horizontal mirror image and the vertical mirror image in the step 2) comprises the following steps:
let the width of the image small block be width, the length be height, and the original coordinate before the transformation of a certain point in the image small block be (x)0，y0) The transformed coordinates are (x, y), then,
the horizontal mirror image is turned over, and needs to be satisfied:
x＝width-x0-1
y＝y0；
the vertical mirror image is turned over, and needs to be satisfied:
preferably, the cutting method in step 2) is as follows: and keeping the resolution of the image small blocks unchanged, and setting the width of the image small blocks as width, the length as height, and the clipping coefficients of the width and the length as a and b respectively, so that the width of the clipped image is width a, and the length is height b.
Preferably, the method for adjusting brightness, saturation and hue in step 2) comprises:
converting the RGB pixel values of the image small blocks into HSI, wherein the conversion formula is as follows:
wherein,
the saturation component is given by:
the intensity component is given by:
and setting the change coefficients of the three HSIs to be 0.1, randomly changing the HSI of the image small block to be 0.9 or 1.1 times of the original HSI, and finishing the adjustment of the brightness, the saturation or the hue of the image small block.
In order to ensure the sufficient training and effective convergence of the deep learning model, a large number of sample pictures need to be collected before training as input during model training, and generally, labeled pathological section pictures which can be provided by doctors are very limited and often cannot meet the training requirements of the deep learning model.
The invention is established under the background of training a deep learning model by utilizing a pathological section image, and the background has two problems which make the pathological section image difficult to be directly used for training the model: firstly, the picture is too large to be directly used for training the model and needs to be cut into small blocks; secondly, the number of slice images is small, so the number of samples formed by direct blocking is small, and the training of the model is insufficient.
By adopting the technical scheme, the liver pathological image is divided into blocks to obtain a plurality of image small blocks, and each image small block is randomly transformed to form an expanded sample, so that the original pathological sample can be effectively expanded, the problems of insufficient sample quantity and uneven sample distribution are solved to a certain extent, the requirement of a deep learning model on large sample quantity is met, the situations of overfitting and insufficient generalization capability of the trained model can be effectively avoided, and the reliability of an auxiliary analysis result is improved. The core of the invention lies in that after random transformation is carried out on small image blocks, the small image blocks are applied to a picture preprocessing link before a deep learning model is trained, and the problems of overlarge pathological section picture and insufficient sample size are effectively solved.
Drawings
The invention will now be further elucidated with reference to the accompanying drawings:
FIG. 1 is a flow chart of a method for enhancing a liver pathology image sample based on stochastic transformation according to the present invention;
FIG. 2 is a schematic diagram of random variation of the liver pathology image sample enhancement method based on random transformation according to the present invention.
Detailed Description
As shown in fig. 1 or fig. 2, the liver pathology image sample enhancement method based on random transformation of the present invention includes the following steps:
1) carrying out block division on the liver pathological image to obtain a plurality of image small blocks;
2) randomly transforming each image small block to form an extended sample; the random transformation comprises more than one of horizontal mirror image turning, vertical mirror image turning, clipping, brightness adjustment, saturation adjustment and hue adjustment;
3) and inputting the extended sample into a deep learning model to train the liver pathological image, and correspondingly enhancing the liver pathological image to obtain an enhanced sample of the liver pathological image.
Preferably, the size of the image patch in step 1) is 320 × 320 pixels.
Preferably, the method for turning over the horizontal mirror image and the vertical mirror image in the step 2) comprises the following steps:
let the width of the image small block be width, the length be height, and the original coordinate before the transformation of a certain point in the image small block be (x)0，y0) The transformed coordinates are (x, y), then,
the horizontal mirror image is turned over, and needs to be satisfied:
x＝width-x0-1
y＝y0；
the vertical mirror image is turned over, and needs to be satisfied:
preferably, the cutting method in step 2) is as follows: and keeping the resolution of the image small blocks unchanged, and setting the width of the image small blocks as width, the length as height, and the clipping coefficients of the width and the length as a and b respectively, so that the width of the clipped image is width a, and the length is height b.
Preferably, the method for adjusting brightness, saturation and hue in step 2) comprises:
converting the RGB pixel values of the image small blocks into HSI, wherein the conversion formula is as follows:
wherein,
the saturation component is given by:
the intensity component is given by:
and setting the change coefficients of the three HSIs to be 0.1, randomly changing the HSI of the image small block to be 0.9 or 1.1 times of the original HSI, and finishing the adjustment of the brightness, the saturation or the hue of the image small block.
Examples
The invention relates to a liver pathology image sample enhancement method based on random transformation, which comprises the following steps:
step1, dividing image blocks
Pathological images are usually of very high resolution, making the calculation extremely inefficient if the whole image is directly manipulated. Therefore, firstly, according to the resolution of the full-scanning pathological image, the appropriate block length is set, and the whole pathological image is subjected to local blocking. For the liver pathology image, the selected block size is 320 × 320 pixels, all wn × hn small blocks are obtained, and the position of each small block is recorded at the same time of blocking.
Step 2, sample enhancement
(1) Mirror image turning (horizontal or vertical)
The principle of horizontal and vertical mirroring of image patches is as follows: setting the width of the image as width and the length as height; (x, y) is a coordinate converted from a certain point in the image, (x)0，y0) The original coordinates before transformation are obtained; then it is determined that,
the horizontal mirror image turning needs to meet the following requirements:
x＝width-x0-1
y＝y0
the vertical mirror image turning needs to meet the following requirements:
(2) random cutting
And keeping the resolution of the original image unchanged, and taking out a part of content in the original image, wherein the process of clipping the image small blocks is that the width of the image small blocks is width and the length of the image small blocks is height according to clipping coefficients a and b, so that the width of the clipped image is width a and the length of the clipped image is height b.
(3) Adjusting brightness, saturation and hue
Generally, Hue (Hue), Saturation (Saturation) and brightness (Intensity) of an image are collectively referred to as HSI, and for another description of an RGB image, adjusting image HIS requires converting pixel values of RGB of the image into HSI, and the conversion formula is as follows:
wherein,
the saturation component is given by:
the final intensity component is given by:
in the pathological picture of the liver, the change coefficients of the three HSI are all set to be 0.1, the HSI of the original image is randomly changed to be 0.9 or 1.1 times of the original HSI, and the adjustment of the brightness, the saturation and the tone of the image is completed.
The above description should not be taken as limiting the scope of the invention in any way.
Claims (5)
1. A liver pathology image sample enhancement method based on random transformation is characterized in that: which comprises the following steps:
1) carrying out block division on the liver pathological image to obtain a plurality of image small blocks;
2) randomly transforming each image small block to form an extended sample; the random transformation comprises more than one of horizontal mirror image turning, vertical mirror image turning, clipping, brightness adjustment, saturation adjustment and hue adjustment;
3) and inputting the extended sample into a deep learning model to train the liver pathological image, and correspondingly enhancing the liver pathological image to obtain an enhanced sample of the liver pathological image.
2. The liver pathology image sample enhancement method based on stochastic transformation according to claim 1, wherein: the size of the image patch in step 1) is 320 x 320 pixels.
3. The liver pathology image sample enhancement method based on stochastic transformation according to claim 1, wherein: the method for turning over the horizontal mirror image and the vertical mirror image in the step 2) comprises the following steps:
let the width of the image small block be width, the length be height, and the original coordinate before the transformation of a certain point in the image small block be (x)0，y0) The transformed coordinates are (x, y), then,
the horizontal mirror image is turned over, and needs to be satisfied:
x＝width-x0-1
y＝y0；
the vertical mirror image is turned over, and needs to be satisfied:
4. the liver pathology image sample enhancement method based on stochastic transformation according to claim 1, wherein: the cutting method in the step 2) comprises the following steps: and keeping the resolution of the image small blocks unchanged, and setting the width of the image small blocks as width, the length as height, and the clipping coefficients of the width and the length as a and b respectively, so that the width of the clipped image is width a, and the length is height b.
5. The liver pathology image sample enhancement method based on stochastic transformation according to claim 1, wherein: the method for adjusting brightness, saturation and hue in step 2) comprises the following steps:
converting the RGB pixel values of the image small blocks into HSI, wherein the conversion formula is as follows:
wherein,
the saturation component is given by:
the intensity component is given by:
and setting the change coefficients of the three HSIs to be 0.1, randomly changing the HSI of the image small block to be 0.9 or 1.1 times of the original HSI, and finishing the adjustment of the brightness, the saturation or the hue of the image small block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910526943.0A CN110288542A (en) | 2019-06-18 | 2019-06-18 | A kind of liver's pathological image sample Enhancement Method based on stochastic transformation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910526943.0A CN110288542A (en) | 2019-06-18 | 2019-06-18 | A kind of liver's pathological image sample Enhancement Method based on stochastic transformation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110288542A true CN110288542A (en) | 2019-09-27 |
Family
ID=68004191
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910526943.0A Pending CN110288542A (en) | 2019-06-18 | 2019-06-18 | A kind of liver's pathological image sample Enhancement Method based on stochastic transformation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110288542A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866893A (en) * | 2019-09-30 | 2020-03-06 | 中国科学院计算技术研究所 | Pathological image-based TMB classification method, system and TMB analysis device |
| CN111062956A (en) * | 2019-11-08 | 2020-04-24 | 哈尔滨工业大学（深圳） | Pathological image mass target segmentation method and device |
| CN113011468A (en) * | 2021-02-25 | 2021-06-22 | 上海皓桦科技股份有限公司 | Image feature extraction method and device |
| GB2591177A (en) * | 2019-11-21 | 2021-07-21 | Hsiao Ching Nien | Method and apparatus of intelligent analysis for liver tumour |
| CN116779170A (en) * | 2023-08-24 | 2023-09-19 | 济南市人民医院 | Pulmonary function attenuation prediction system and device based on self-adaptive deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPWO2012157201A1 (en) * | 2011-05-18 | 2014-07-31 | 日本電気株式会社 | Information processing system, information processing method, information processing apparatus, control method thereof, and control program |
| CN109740626A (en) * | 2018-11-23 | 2019-05-10 | 杭州电子科技大学 | Detection method of cancer regions in breast cancer pathological slices based on deep learning |
| CN109740669A (en) * | 2018-12-29 | 2019-05-10 | 大连大学 | A breast cancer pathological image classification method based on deep feature aggregation |
-
2019
- 2019-06-18 CN CN201910526943.0A patent/CN110288542A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPWO2012157201A1 (en) * | 2011-05-18 | 2014-07-31 | 日本電気株式会社 | Information processing system, information processing method, information processing apparatus, control method thereof, and control program |
| CN109740626A (en) * | 2018-11-23 | 2019-05-10 | 杭州电子科技大学 | Detection method of cancer regions in breast cancer pathological slices based on deep learning |
| CN109740669A (en) * | 2018-12-29 | 2019-05-10 | 大连大学 | A breast cancer pathological image classification method based on deep feature aggregation |
Non-Patent Citations (2)
| Title |
|---|
| 余胜威等: "《MATLAB图像滤波去噪分析及其应用》", 30 September 2015 * |
| 李琼等: "糖尿病性视网膜图像的深度学习分类方法", 《中国图象图形学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866893A (en) * | 2019-09-30 | 2020-03-06 | 中国科学院计算技术研究所 | Pathological image-based TMB classification method, system and TMB analysis device |
| CN110866893B (en) * | 2019-09-30 | 2021-04-06 | 中国科学院计算技术研究所 | Pathological image-based TMB classification method, system and TMB analysis device |
| US11468565B2 (en) | 2019-09-30 | 2022-10-11 | Institute Of Computing Technology, Chinese Academy Of Sciences | TMB classification method and system and TMB analysis device based on pathological image |
| CN111062956A (en) * | 2019-11-08 | 2020-04-24 | 哈尔滨工业大学（深圳） | Pathological image mass target segmentation method and device |
| GB2591177A (en) * | 2019-11-21 | 2021-07-21 | Hsiao Ching Nien | Method and apparatus of intelligent analysis for liver tumour |
| CN113011468A (en) * | 2021-02-25 | 2021-06-22 | 上海皓桦科技股份有限公司 | Image feature extraction method and device |
| CN116779170A (en) * | 2023-08-24 | 2023-09-19 | 济南市人民医院 | Pulmonary function attenuation prediction system and device based on self-adaptive deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110288542A (en) | A kind of liver's pathological image sample Enhancement Method based on stochastic transformation | |
| CN113256641B (en) | A deep learning-based image segmentation method for skin lesions | |
| US20040264749A1 (en) | Boundary finding in dermatological examination | |
| DE69426745T2 (en) | System and method for processing images of living tissue | |
| CN101706843B (en) | Interactive film Interpretation method of mammary gland CR image | |
| CN108596887A (en) | A kind of abdominal CT sequence image liver neoplasm automatic division method | |
| CN113570619B (en) | Computer-aided pancreatic pathology image diagnosis system based on artificial intelligence | |
| CN101576997A (en) | Abdominal organ segmentation method based on secondary three-dimensional region growth | |
| CN110390678B (en) | A tissue type segmentation method for colorectal cancer IHC stained images | |
| JP2002165757A (en) | Diagnostic supporting system | |
| CN108629761A (en) | A kind of breast cancer image-recognizing method, device and user terminal | |
| CN106780453A (en) | A kind of method realized based on depth trust network to brain tumor segmentation | |
| CN111583246A (en) | Method for classifying liver tumors by utilizing CT (computed tomography) slice images | |
| JP2016507229A (en) | Method and system for cell level FISH dot count | |
| CN110728239A (en) | Gastric cancer enhanced CT image automatic identification system utilizing deep learning | |
| JP2018517427A (en) | Tissue sample analysis technology | |
| CN103236062B (en) | Based on the magnetic resonance image (MRI) blood vessel segmentation system in human brain tumour's nuclear-magnetism storehouse | |
| CN113689442A (en) | Method for realizing lung organ segmentation based on three-dimensional image | |
| CN112617799A (en) | Network model for endometrial tumor segmentation and segmentation method | |
| Gangwar | Study on different Skin lesion Segmentation techniques and their comparisons | |
| Nowak et al. | Texture analysis for dermoscopic image processing | |
| CN117274606A (en) | Nuclear magnetic resonance brain tumor region extraction method | |
| CN113344938A (en) | Liver tumor image segmentation model training method | |
| CN112396616A (en) | Osteosarcoma recurrence risk prediction model based on tissue morphology analysis | |
| CN118587238B (en) | Identification analysis construction and image processing method for special lung pathology image |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190927 |