CN112037231A - Medical image liver segmentation post-processing method based on feature classifier - Google Patents
Medical image liver segmentation post-processing method based on feature classifier Download PDFInfo
- Publication number
- CN112037231A CN112037231A CN202010751090.3A CN202010751090A CN112037231A CN 112037231 A CN112037231 A CN 112037231A CN 202010751090 A CN202010751090 A CN 202010751090A CN 112037231 A CN112037231 A CN 112037231A
- Authority
- CN
- China
- Prior art keywords
- tumor
- algorithm
- classifier
- segmentation
- post
- 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
- 230000011218 segmentation Effects 0.000 title claims abstract description 34
- 210000004185 liver Anatomy 0.000 title claims abstract description 21
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000012805 post-processing Methods 0.000 title claims abstract description 18
- 206010028980 Neoplasm Diseases 0.000 claims abstract description 98
- 230000000877 morphologic effect Effects 0.000 claims abstract description 16
- 238000013135 deep learning Methods 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims description 12
- 208000007660 Residual Neoplasm Diseases 0.000 claims description 3
- 238000003066 decision tree Methods 0.000 claims description 3
- 238000007477 logistic regression Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 10
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 238000002591 computed tomography Methods 0.000 description 25
- 238000013473 artificial intelligence Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses a medical image liver segmentation post-processing method based on a feature classifier, which judges whether a corresponding tumor block is a real tumor or not by extracting morphological features of the tumor block identified by an algorithm and inputting the morphological features into the classifier, and comprises the following specific steps: s1, extracting morphological characteristics of the interior of the tumor region after algorithm segmentation; s2, training a corresponding classifier; and S3, performing post-processing on the result of the deep learning algorithm recognition by using the trained classifier. The method can be effectively used in the post-processing of the deep learning algorithm on the semantic segmentation recognition of the tumor, and filters out the wrong tumor, thereby improving each index of the recognition result, and simultaneously improving the effect of the algorithm segmentation on the auxiliary diagnosis.
Description
Technical Field
The invention belongs to the technical field of energy storage, and particularly relates to a medical image liver segmentation post-processing method based on a feature classifier.
Background
The identification of diseases is often done medically by means of medical images, and the diagnosis assisted by medical images based on artificial intelligence method can assist doctors in the diagnosis of diseases by an algorithm model trained on a large amount of labeled data. The semantic segmentation means that a target region is segmented from an image in a pixel point classification mode, the current artificial intelligence methods such as deep learning are widely applied to semantic segmentation of medical images, and a liver or a pathological change region in the medical images can be extracted in a mode of training a neural network.
However, in the existing deep learning method, some wrong segmentation regions may exist after prediction, that is, some positions which are not tumors are mistakenly identified as tumors. When the number of wrong tumors identified by the algorithm is too large, the effect of the algorithm on the auxiliary identification of doctors is greatly influenced. Improving the recognition accuracy of the deep learning algorithm can reduce the number of recognized error tumors, but this method needs time and labor consuming to conceive a better model and needs stronger calculation for training.
Disclosure of Invention
In order to solve the above problems, the present invention provides a medical image liver segmentation post-processing method based on a feature classifier, which comprises the steps of extracting morphological features of tumor blocks identified by an algorithm, inputting the morphological features into the classifier to judge whether the corresponding tumor blocks are real tumors, and specifically comprises the following steps:
s1, extracting morphological characteristics of the interior of the tumor region after algorithm segmentation;
s2, training a corresponding classifier;
and S3, performing post-processing on the result of the deep learning algorithm recognition by using the trained classifier.
Preferably, the step S1 includes:
s101, training a semantic segmentation deep learning algorithm to identify a liver region and a tumor region in a CT image;
s102, predicting the data of the training set once to obtain a tumor label of the training set, namely whether the tumor is a wrong prediction or not;
s103, extracting the sizes of the tumors in the residual tumor blocks and the corresponding characteristics of all the tumors in the CT;
and S104, if the multi-stage CT exists, aligning the multi-stage CT by using an alignment algorithm, extracting the CT distribution characteristics of the position of the tumor block in each stage, and splicing the characteristics to obtain the characteristic vector of each tumor. And the morphological characteristics of multiple stages are fused and input into a classifier to identify whether the tumor block is a true tumor, and better classification effect can be obtained by using the characteristic data of multiple stages.
Preferably, the corresponding features include CT distribution features such as an average value, a maximum value, a minimum value, a standard deviation, and a skewness of CT pixel values inside the tumor block, an average value of distances from each layer of the tumor block to left, right, upper, and lower boundaries of the liver, and the number of layers corresponding to the upper and lower boundaries of the liver and the number of layers corresponding to the tumor block. The distribution of the CT values of the real tumor generally falls within the corresponding organ, and the CT values within the block contain a certain rule, so that the shape, size and position of the tumor can be used to distinguish whether the tumor is the real tumor.
Preferably, the specific step of S2 is:
s201, classifying the tumor blocks obtained in the S1 and the corresponding labels thereof by using an xgboost algorithm; the xgboost algorithm trains the corresponding model by minimizing the loss function, and the minimizing loss function of the xgboost model is defined as the weighted sum of the loss between the predicted label and the real label of the model on the multiple base classifiers, and the specific formula is as follows:
wherein: penalty term
T is the number of leaf nodes of the decision tree, ωIs the weight of each node of the tree, and
then, the difference loss function between the predicted label and the real label is calculated, and here, we select a binary logistic regression loss function:
s202, increasing the scale _ pos _ weight parameter of the algorithm to increase the weight for correctly classifying the positive samples, wherein the specific formula is as follows:
the main influence of the parameter on the operation result is to increase the weight of the positive sample after prediction, thereby increasing the influence of the positive sample during the gradient replay; the recall rate of the classifier can be improved as much as possible by increasing the parameters, so that the classifier retains more tumors with correct segmentation.
Preferably, the S3 includes: and (3) extracting corresponding characteristics of each tumor block in the segmentation result output by the algorithm by using a trained classifier, judging whether the tumor block is wrongly segmented, and performing post-processing to remove the tumor block if the tumor block is wrongly segmented to obtain a final segmentation result.
Compared with the prior art, the invention has the beneficial effects that: the method can be effectively used in the post-processing of the deep learning algorithm on the semantic segmentation recognition of the tumor, and the wrong tumor is filtered, so that various indexes of the recognition result such as the dice score are improved, and meanwhile, the effect of the algorithm segmentation on the auxiliary diagnosis can be improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a detailed algorithmic flow chart of the present invention;
FIG. 3 is a sample effect of the mis-segmentation identification of the present invention FIG. 1;
FIG. 4 is a sample effect of the miscut identification of the present invention FIG. 2;
FIG. 5 is a sample effect of the miscut identification of the present invention FIG. 3;
FIG. 6 is a sample effect of the segmentation error recognition of the present invention in FIG. 4.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention.
As shown in fig. 1 to 6, the present embodiment provides a method for processing after medical image liver segmentation based on a feature classifier, which extracts morphological features of a tumor block identified by an algorithm, inputs the morphological features into the classifier to determine whether the corresponding tumor block is a real tumor, and specifically includes the following steps:
s1, extracting morphological characteristics of the interior of the tumor region after algorithm segmentation;
s2, training a corresponding classifier;
s3, performing post-processing on the result of the deep learning algorithm recognition by using the trained classifier; and (3) extracting corresponding characteristics of each tumor block in the segmentation result output by the algorithm by using a trained classifier, judging whether the tumor block is wrongly segmented, and performing post-processing to remove the tumor block if the tumor block is wrongly segmented to obtain a final segmentation result. Fig. 3 to 6 show the effect diagrams of the cases of the erroneous segmentation recognition near the liver boundary.
The specific method for implementing step S1 in this embodiment includes:
s101, training a semantic segmentation deep learning algorithm to identify a liver region and a tumor region in a CT image;
s102, predicting the data of the training set once to obtain a tumor label of the training set, namely whether the tumor is a wrong prediction or not;
s103, extracting the sizes of the tumors in the residual tumor blocks and the corresponding characteristics of all the tumors in the CT;
and S104, if the multi-stage CT exists, aligning the multi-stage CT by using an alignment algorithm, extracting the CT distribution characteristics of the position of the tumor block in each stage, and splicing the characteristics to obtain the characteristic vector of each tumor.
In order to improve the effect of deep learning and identification and remove redundant erroneous tumors, this task may be regarded as a classification task, and for each tumor, a label of 1 in this embodiment indicates that the tumor is a true tumor, and a label of 0 indicates that the tumor is an erroneously identified tumor. There are many methods to process the classification task, and because the tumors identified by the deep learning algorithm are different in size, this point brings great difficulty to accurately judge whether the tumor block is true, so this embodiment adopts the feature extraction algorithm to classify, and reduces the influence of the tumor size by extracting the morphological statistical features in the region. Under the condition that multi-stage CT (computed tomography) such as a flat scanning stage, a venous stage and an arterial stage exists, a corresponding pixel block in the multi-stage CT can be extracted aiming at a tumor block by combining a multi-stage CT alignment algorithm, the morphological characteristics of the multi-stage CT are fused, and the fused pixel block is input to a classifier to identify whether the tumor block is a true tumor.
The corresponding characteristics comprise CT distribution characteristics such as the average value, the maximum value, the minimum value, the standard deviation, the skewness and the like of CT pixel values in the tumor block, the distance average value of each layer of the tumor block from the left, right and upper boundaries of the liver, the number of layers corresponding to the upper and lower boundaries of the liver and the number of layers corresponding to the tumor block. The distribution of the CT values of the real tumor generally falls within the corresponding organ, and the CT values within the block contain a certain rule, so that the shape, size and position of the tumor can be used to distinguish whether the tumor is the real tumor. Under the condition of multi-stage CT, after alignment, CT characteristics on corresponding positions are extracted from the three stages simultaneously, so that whether the tumor is a real tumor or not can be better identified.
The method adopted by the invention extracts the morphological characteristics of the tumor blocks identified by the algorithm and inputs the morphological characteristics into the xgboost classifier to judge whether the corresponding tumor blocks are real tumors.
The specific operation of training the corresponding classifier in this embodiment is as follows:
s201, classifying the tumor blocks obtained in the S1 and the corresponding labels thereof by using an xgboost algorithm; the xgboost algorithm trains the corresponding model by minimizing the loss function, and the minimizing loss function of the xgboost model is defined as the weighted sum of the loss between the predicted label and the real label of the model on the multiple base classifiers, and the specific formula is as follows:
wherein: penalty term
T is the number of leaf nodes of the decision tree, ω is the weight of each node of the tree, and
then, the difference loss function between the predicted label and the real label is calculated, and here, we select a binary logistic regression loss function:
s202, increasing the scale _ pos _ weight parameter of the algorithm to increase the weight for correctly classifying the positive samples, wherein the specific formula is as follows:
the main influence of the parameter on the operation result is that the weight of the positive sample after prediction is increased, so that the influence of the positive sample during gradient replay is increased, the recall rate of the classifier can be improved as much as possible by increasing the parameter, and the classifier can keep more tumors with correct segmentation.
It is to be understood that the described embodiments are merely some embodiments of the invention and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Claims (5)
1. The medical image liver segmentation post-processing method based on the feature classifier is characterized in that morphological features of tumor blocks identified by an algorithm are extracted and input into the classifier to judge whether the corresponding tumor blocks are real tumors or not, and the specific steps comprise:
s1, extracting morphological characteristics of the interior of the tumor region after algorithm segmentation;
s2, training a corresponding classifier;
and S3, performing post-processing on the result of the deep learning algorithm recognition by using the trained classifier.
2. The method for post-processing liver segmentation of medical image based on feature classifier as claimed in claim 1, wherein the step S1 includes:
s101, training a semantic segmentation deep learning algorithm to identify a liver region and a tumor region in a CT image;
s102, predicting the data of the training set once to obtain a tumor label of the training set, namely whether the tumor is a wrong prediction or not;
s103, extracting the sizes of the tumors in the residual tumor blocks and the corresponding characteristics of all the tumors in the CT;
and S104, if the multi-stage CT exists, aligning the multi-stage CT by using an alignment algorithm, extracting the CT distribution characteristics of the position of the tumor block in each stage, and splicing the characteristics to obtain the characteristic vector of each tumor.
3. The method of claim 2, wherein the corresponding features include CT distribution features of mean, maximum, minimum, standard deviation and skewness of CT pixel values inside the tumor block, mean distance between each layer of the tumor block and left, right and upper and lower boundaries of the liver, and the number of layers corresponding to the upper and lower boundaries of the liver and the number of layers corresponding to the tumor block.
4. The method for post-processing medical image liver segmentation based on the feature classifier as claimed in claim 1, wherein the step S2 comprises the steps of:
s201, classifying the tumor blocks obtained in the S1 and the corresponding labels thereof by using an xgboost algorithm: the xgboost algorithm trains the corresponding model by minimizing the loss function, and the minimizing loss function of the xgboost model is defined as the weighted sum of the loss between the predicted label and the real label of the model on the multiple base classifiers, and the specific formula is as follows:
wherein: penalty term
T is the number of leaf nodes of the decision tree, ω is the weight of each node of the tree, and
then, the difference loss function of the predicted label and the real label is calculated, and a binary logistic regression loss function is selected:
s202, increasing the scale _ pos _ weight parameter of the algorithm to increase the weight for correctly classifying the positive sample, wherein the main influence of the parameter on the operation result is to increase the weight of the positive sample after prediction, so that the influence of the positive sample during gradient replay is increased, and the specific formula is as follows:
5. the method for post-processing liver segmentation of medical image based on feature classifier as claimed in claim 1, wherein the step S3 includes: and (3) extracting corresponding characteristics of each tumor block in the segmentation result output by the algorithm by using a trained classifier, judging whether the tumor block is wrongly segmented, and performing post-processing to remove the tumor block if the tumor block is wrongly segmented to obtain a final segmentation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010751090.3A CN112037231A (en) | 2020-07-30 | 2020-07-30 | Medical image liver segmentation post-processing method based on feature classifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010751090.3A CN112037231A (en) | 2020-07-30 | 2020-07-30 | Medical image liver segmentation post-processing method based on feature classifier |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112037231A true CN112037231A (en) | 2020-12-04 |
Family
ID=73583579
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010751090.3A Pending CN112037231A (en) | 2020-07-30 | 2020-07-30 | Medical image liver segmentation post-processing method based on feature classifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112037231A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109493362A (en) * | 2018-09-03 | 2019-03-19 | 李磊 | A kind of human body foreground segmentation algorithm of neural network jointing edge detection |
| CN110188788A (en) * | 2019-04-15 | 2019-08-30 | 浙江工业大学 | The classification method of cystic Tumor of Pancreas CT image based on radiation group feature |
| CN110837844A (en) * | 2019-09-26 | 2020-02-25 | 浙江工业大学 | Pancreatic cystic tumor benign and malignant classification method based on CT image dissimilarity characteristics |
| US20200074632A1 (en) * | 2017-04-12 | 2020-03-05 | Kheiron Medical Technologies Ltd | Assessment of density in mammography |
| KR20200082660A (en) * | 2018-12-31 | 2020-07-08 | 주식회사 메디픽셀 | Pathological diagnosis method and apparatus based on machine learning |
-
2020
- 2020-07-30 CN CN202010751090.3A patent/CN112037231A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200074632A1 (en) * | 2017-04-12 | 2020-03-05 | Kheiron Medical Technologies Ltd | Assessment of density in mammography |
| CN109493362A (en) * | 2018-09-03 | 2019-03-19 | 李磊 | A kind of human body foreground segmentation algorithm of neural network jointing edge detection |
| KR20200082660A (en) * | 2018-12-31 | 2020-07-08 | 주식회사 메디픽셀 | Pathological diagnosis method and apparatus based on machine learning |
| CN110188788A (en) * | 2019-04-15 | 2019-08-30 | 浙江工业大学 | The classification method of cystic Tumor of Pancreas CT image based on radiation group feature |
| CN110837844A (en) * | 2019-09-26 | 2020-02-25 | 浙江工业大学 | Pancreatic cystic tumor benign and malignant classification method based on CT image dissimilarity characteristics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6941123B2 (en) | Cell annotation method and annotation system using adaptive additional learning | |
| Andrade et al. | Recent computational methods for white blood cell nuclei segmentation: A comparative study | |
| CN109345527B (en) | Bladder tumor detection method based on MaskRcnn | |
| CN110796186A (en) | Dry and wet garbage identification and classification method based on improved YOLOv3 network | |
| BenTaieb et al. | Multi-loss convolutional networks for gland analysis in microscopy | |
| JP5394959B2 (en) | Discriminator generating apparatus and method, and program | |
| CN107122735B (en) | Multi-target tracking method based on deep learning and conditional random field | |
| CN108073918B (en) | Method for extracting blood vessel arteriovenous cross compression characteristics of fundus retina | |
| CN109871885A (en) | A kind of plants identification method based on deep learning and Plant Taxonomy | |
| CN111597887B (en) | Pedestrian re-identification method and system | |
| CN111598875A (en) | Method, system and device for building thyroid nodule automatic detection model | |
| CN109583375B (en) | Multi-feature fusion face image illumination identification method and system | |
| Stanley et al. | Data-driven homologue matching for chromosome identification | |
| CN112330686A (en) | Method for segmenting and calibrating lung bronchus | |
| CN114596467A (en) | Multimode image classification method based on evidence deep learning | |
| CN111241987B (en) | Multi-target model visual tracking method based on cost-sensitive three-branch decision | |
| CN114997169B (en) | Entity word recognition method and device, electronic equipment and readable storage medium | |
| CN115862045A (en) | Case automatic identification method, system, equipment and storage medium based on image-text identification technology | |
| CN115880704A (en) | Automatic case cataloging method, system, equipment and storage medium | |
| CN113658199B (en) | Regression correction-based chromosome instance segmentation network | |
| Zhang et al. | Whole slide image classification via iterative patch labelling | |
| CN111612749B (en) | Focus detection method and device based on lung image | |
| CN112037231A (en) | Medical image liver segmentation post-processing method based on feature classifier | |
| Mia et al. | CompTRNet: An U-Net Approach for White Blood Cell Segmentation Using Compound Loss Function and Transfer Learning with Pre-trained ResNet34 Network | |
| CN111027602B (en) | Method and system for detecting target with multi-level structure |
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 |
Application publication date: 20201204 |
|
| RJ01 | Rejection of invention patent application after publication |