CN107368695A - The construction method of GISTs malignant potential disaggregated model based on radiation group - Google Patents
The construction method of GISTs malignant potential disaggregated model based on radiation group Download PDFInfo
- Publication number
- CN107368695A CN107368695A CN201710734667.8A CN201710734667A CN107368695A CN 107368695 A CN107368695 A CN 107368695A CN 201710734667 A CN201710734667 A CN 201710734667A CN 107368695 A CN107368695 A CN 107368695A
- Authority
- CN
- China
- Prior art keywords
- radiation group
- feature
- group
- radiation
- disaggregated model
- 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; CALCULATING OR 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; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
-
- 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
- G06T2207/30092—Stomach; Gastric
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (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)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention belongs to oncology, iconography, computer-aided medical science field, is related to a kind of construction method of the gastrointestinal stromal tumor malignant potential disaggregated model based on radiation group.Construction method of the present invention includes：Data acquisition：Belly strengthens phase thin-layer CT IMAQ；Extract radiation group feature；Statistical analysis；Feature selecting is built with radiation group characteristic model；Radiation group characteristic model is verified and calibration.The GISTs malignant potential disaggregated model based on radiation group constructed by the method for the invention accurately can be classified the GIST with different malignant potentials, belong to a kind of atraumatic technique, and the extra charge of patient will not be increased, operation is simple, is easy to clinical expansion to use.
Description
Technical field
The invention belongs to oncology, iconography, computer-aided medical science field, is related to a kind of stomach and intestine based on radiation group
The construction method of road mesenchymoma (Gastrointestinal Stromal Tumors, GIST) malignant potential disaggregated model.
Background technology
Gastrointestinal stromal tumor (Gastrointestinal Stromal Tumors, GIST) is that one kind originates from intestines and stomach
The tumour of interstitial tissue, account for the major part of leaf tumour between alimentary canal.SABC detection is often expressed as CD117, shows Ka Haer
Cell (Cajal cell) breaks up, and most of cases have c-kit or PDGFRA Activating mutations.In recent years, GIST is in China
The incidence of disease is in the trend rapidly risen.Radical cure GIST unique method is surgery excision, but GIST high-risk patients postoperative recurrence turns
Shifting rate may be up to 55%~90%.And the tumour to recurring potential with height carries out small molecule targeted drug --- Imatinib
Auxiliary treatment, the recurrence of tumour and transfer probability can be significantly reduced.Therefore, clinically most of GIST patients are after surgery
Receive the auxiliary treatment of Imatinib.But, if there is patient to be only just cured by operation in factWhether have again
GIST high-risk patients cause death because selection is let slip a golden opportunity without treatment with imatinibIt is how accurate in face of above mentioned problem
Really differentiating and distinguish the GIST of different malignant potentials turns into the key point of clinical diagnosis and treatment.
At present clinically, GIST sorting technique mainly includes two kinds：Improve NIH and AFIP.It is equal to improve NIH and AFIP
Three tumour maximum diameter, Mitotic count and tumor locus parameters are introduced, wherein improvement NIH also introduces tumour and ruptures this
Individual parameter.But due to AFIP not as NIH it is directly perceived, therefore, limit its extensive use to a certain extent.
However, and not all GIST biological behaviour and Clinical Outcome can be solved with these foregoing sorting techniques
Release, for example the GIST of a part of very little with rapid progression and can develop into hepatic metastasis；Also it is no lack of a part of big GIST patient
Even if do not receive postoperative adjuvant therapy still keeps disease-free survival for a long time.Meanwhile unlike plain edition GIST, Mitotic index
SDH deficiencies GIST risk factor can not be assessed.SDH deficiency GIST patient's Mitotic index it is few can liver metastasis, core point
Split as more can not but shift, and the intermittent phase shifted is longer, it is necessary to long term follow-up.Many plain editions shifted
GIST patient is dead generally in 1~2 year, and SDH deficiencies GIST patient can often survive 5 years after TKI inhibitor for treating or
More than.In summary, though existing GIST sorting techniques can provide some clues and direction for the diagnosis and treatment of clinician, but still have
Larger lifting and improved space.
The content of the invention
It is an object of the invention to provide a kind of gastrointestinal stromal tumor (GIST) malignant potential classification based on radiation group
The method of model, it extracts characteristic parameter therein, then pass through meter according to the iconography image of GIST patient by computer
Calculation machine function is classified the GIST with different malignant potentials, so as to establish a kind of Gastrointestinal Stromal based on radiation group
Knurl malignant potential disaggregated model, to help clinician to make more preferable clinical decision.
The construction method of GISTs malignant potential disaggregated model of the present invention based on radiation group, including with
Lower step：
A. data acquisition：Belly strengthens phase thin-layer CT IMAQ；
B. radiation group feature is extracted：ROI is sketched out using ITK-SNAP softwares, tumor's profiles are successively sketched out,
Then two-dimentional ROI is subjected to Multi-slice spiral CT to generate VOI, characteristic is therefrom extracted using Matlab 2014b softwares,
Including textural characteristics and non-grain feature；
C. statistical analysis：Using statistical method, assess in main queue with examining tumor size, nuclear fission in queue
As the single factor test between clinical risk factors and radiation group feature such as, gene expression characteristicses associates；
D. feature selecting is built with radiation group characteristic model：
1) model is established using LASSO algorithms：High dimensional data is subjected to dimensionality reduction, concentrates screening most useful from initial data
The category feature of two category features, respectively Global and GLSZM two, the anticipation function of the linear combination of feature selected by foundation：
Rad score=0.1335-0.2209 × GLSZM_RLV (r=0.5, s=4, a=Equal, n=64)
- 0.0061 × Global_Variance (r=2/3, s=4, a=Lloyd, n=8)；
2) data analysis, including training group and validation group are carried out to each patient by the anticipation function, establishes DICOM
Database, design disaggregated model；
E. the checking of radiation group characteristic model and calibration：Calibration curve is drawn to assess the characteristic model of gained, is carried out
The Hosmer-Lemenshow goodness of fit is tested, and the C- indexes of relative calibration is calculated, to carry out system calibration；Wherein, training group
C- indexes be 0.8161 (95%CI, 0.7453 to 0.8988), the C- indexes of test group for 0.8079 (95%CI, 0.7279
To 0.9042).
The construction method of GISTs malignant potential disaggregated model of the present invention based on radiation group have with
Lower beneficial effect：
(1) radiation group can disclose the predictive signal of tumour, can capture intra-tumor heterogeneity；
(2) model establishes data source in the preoperative conventional CT images of patient in the present invention, avoids Conventional diagnostic method
Damage to patient body, belong to a kind of noninvasive method, and the extra charge of patient will not be increased.
(3) present invention be based on the present computer technology, accurate, efficient and substantially reduce fault rate；The present invention can be based on
Exercisable software is write, beneficial to the application of clinician.
(4) present invention reduces influence of the other influences factor to this model by LASSO simplified models, improve result with
The degree of association of Clinical symptoms.
Using the GIST classification based on radiation group constructed by the method for the invention estimate model be look for another way from
The Features of GIST patient are started with, and are a kind of modes of non-invasive.By being carried with high throughput from the CT images of patient
Take and analyze a large amount of advanced, quantitative Features, using substantial amounts of automation data characterization algorithm, by area-of-interest
(ROI:Region of interest) image data be converted into feature space data that are high-resolution, can excavating,
So as to classify to the GSIT with different malignant potentials, the efficiency and effect of the diagnosis and treatment of clinician are improved.
Brief description of the drawings
Fig. 1 shows Multi-slice spiral CT generation VOI, wherein, A is the segmentation of patient CT tumor boundaries；B is three-dimensional volume
Rebuild the VOI of generation.
Fig. 2 shows the span of lambda values in feature extraction, and position corresponding to the stringer cut-off rule of figure Green is
Final preferred feature numerical value.
Fig. 3 shows scoring distribution of the anticipation function to training group patient；Rad-score(Radiomics signature
Score, the scoring of radiation group anticipation function).
Fig. 4 shows scoring distribution of the anticipation function to test group patient；Rad-score(Radiomics signature
Score radiation groups anticipation function scores).
Fig. 5 shows that the gastro-intestinal stromal disaggregated model based on radiation group uses interface；Image is image input windows；AUC
(area under area under Receiver Operating Characteristic Receiver operating curves)；
RadScore exports for radiation group anticipation function appraisal result.
Embodiment
1. data acquisition：
(1) standard is formulated, the GIST patient (b) that such as (a) on patient did not carried out Imatinib auxiliary treatment passes through
The GIST (c) for completing excision is less than the belly enhanced CT of preoperative 15 days.Exclude influence of the other factors to this experiment.
(2) abdominal CT images for the GIST patient for not carrying out Imatinib auxiliary treatment are obtained, are divided into training group and pre-
Survey group.
(3) made between each case of inspection on underlying factors such as age-sex, tumour original site, histological grades
Into difference it is whether statistically significant.
2. extract radiation group feature：
ROI is sketched out using ITK-SNAP softwares, tumor's profiles are successively sketched out, two-dimentional ROI is then carried out three
Volume reconstruction is tieed up to generate VOI (as shown in Figure 1), characteristic, including texture are therefrom extracted using Matlab 2014b softwares
Feature and non-grain feature.
Note：ITK-SNAP softwares are for the application software to three-dimensional segmentation in medical image, can be provided using master
The wide method of driving wheel, and the function such as manual semi-automatic segmentation delimited with image-guidance (http://www.itksnap.org/ pmwiki/pmwiki.php)。
Non-grain feature：Diameter, capacity, eccentricity, hardness.
Textural characteristics：Totally 9690, typonym is as shown in table 1.
Table 1：Textural characteristics
3. statistical analysis：
Using statistical method, assess in main queue with examining tumor size, Mitotic index, gene expression characteristicses etc. in queue
Single factor test between clinical risk factors and radiation group feature associates.
4. feature selecting is built with radiation group characteristic model：
1) model is established using LASSO algorithms (lasso trick algorithm)：High dimensional data is subjected to dimensionality reduction, concentrates and sieves from initial data
Two most useful features (as shown in Figure 2) are selected, are belonging respectively to Global and GLSZM two types, feature selected by foundation
The anticipation function of linear combination：
Rad score=0.1335-0.2209 × GLSZM_RLV (r=0.5, s=4, a=Equal, n=64)
- 0.0061 × Global_Variance (r=2/3, s=4, a=Lloyd, n=8)；
2) data analysis, including training group and validation group (such as Fig. 3,4 institutes are carried out to each patient by the anticipation function
Show), DICOM databases are established, design disaggregated model.
5. radiation group characteristic model is verified and calibration：
Calibration curve is drawn to assess the characteristic model of gained, carrying out Hosmer-Lemenshow tests, (one kind fitting is excellent
Degree test), the C- indexes of relative calibration are calculated, to carry out system calibration；Wherein, the C- indexes of training group are 0.8161 (95%
CI, 0.7453 to 0.8988), the C- indexes of test group are 0.8079 (95%CI, 0.7279 to 0.9042).
In addition, the image department doctor that selection Hospital of Southern Medical University two has more than 5 years abdominal CT diagnostic experiences
Teacher A and B carries out diagnosis diagosis to qualified 20 random cases simultaneously respectively, and it is poor to both results that applied statistics is gained knowledge
Different carry out conspicuousness comparison.As a result show, there was no significant difference for both.After a period of time, doctor A in the case of not being apprised of,
Diagnosis diagosis is carried out to above-mentioned 20 random cases, statistical analysis is carried out to two times result, draws two times result without conspicuousness
Difference.
Function is drawn by above procedure, and weaves into the software (as shown in Figure 5) of clinician's convenient use, is tried in clinic
Exercise and use, and carry out perspective checking and more external certificates.The software is to be based on the present invention by inventor and voluntarily independently opened
The software of hair.
The GISTs malignant potential disaggregated model based on radiation group constructed by the method for the invention can be accurate
GIST with different malignant potentials is classified, belongs to a kind of atraumatic technique, and the extra charge of patient will not be increased, is grasped
Make simply, to be easy to clinical expansion to use.
Claims (1)
- A kind of 1. construction method of the GISTs malignant potential disaggregated model based on radiation group, it is characterised in that including Following steps：A. data acquisition：Belly strengthens phase thin-layer CT IMAQ；B. radiation group feature is extracted：ROI is sketched out using ITK-SNAP softwares, tumor's profiles are successively sketched out, then Two-dimentional ROI is subjected to Multi-slice spiral CT to generate VOI, characteristic is therefrom extracted using Matlab 2014b softwares, including Textural characteristics and non-grain feature；C. statistical analysis：Using statistical method, assess main queue with examine tumor size in queue, Mitotic index, Single factor test between the clinical risk factors such as gene expression characteristicses and radiation group feature associates；D. feature selecting is built with radiation group characteristic model：1) model is established using LASSO algorithms：High dimensional data is subjected to dimensionality reduction, is concentrated from initial data and screens two most useful classes The category feature of feature, respectively Global and GLSZM two, the anticipation function of the linear combination of feature selected by foundation：Rad score=0.1335-0.2209 × GLSZM_RLV (r=0.5, s=4, a=Equal, n=64) -0.0061 × Global_Variance (r=2/3, s=4, a=Lloyd, n=8)；2) data analysis, including training group and validation group are carried out to each patient by the anticipation function, establishes DICOM data Storehouse, design disaggregated model；E. the checking of radiation group characteristic model and calibration：Calibration curve is drawn to assess the characteristic model of gained, carries out Hosmer- The Lemenshow goodness of fit is tested, and the C- indexes of relative calibration is calculated, to carry out system calibration；Wherein, the C- indexes of training group For 0.8161 (95%CI, 0.7453 to 0.8988), the C- indexes of test group for 0.8079 (95%CI, 0.7279 to 0.9042)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710734667.8A CN107368695A (en) | 2017-08-24 | 2017-08-24 | The construction method of GISTs malignant potential disaggregated model based on radiation group |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710734667.8A CN107368695A (en) | 2017-08-24 | 2017-08-24 | The construction method of GISTs malignant potential disaggregated model based on radiation group |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107368695A true CN107368695A (en) | 2017-11-21 |
Family
ID=60310605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710734667.8A Pending CN107368695A (en) | 2017-08-24 | 2017-08-24 | The construction method of GISTs malignant potential disaggregated model based on radiation group |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107368695A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108038845A (en) * | 2017-12-01 | 2018-05-15 | 南方医科大学南方医院 | A kind of topological analysis householder method for abdominal CT images three-dimensional reconstruction |
| CN108847286A (en) * | 2018-06-04 | 2018-11-20 | 暨南大学附属第医院（广州华侨医院） | Tumour prediction technique based on image feature data |
| CN109259780A (en) * | 2018-07-24 | 2019-01-25 | 南方医科大学南方医院 | The aided assessment system and method that gastric cancer prognosis and chemotherapy based on enhancing CT images group benefit |
| CN109583447A (en) * | 2018-12-11 | 2019-04-05 | 山东省肿瘤防治研究院（山东省肿瘤医院） | A kind of image group credit analysis model building method and analysis method |
| CN109598266A (en) * | 2018-10-24 | 2019-04-09 | 深圳大学 | Lower-limb deep veins thrombus efficiency of thrombolysis prediction technique and system based on machine learning |
| CN112071418A (en) * | 2020-05-26 | 2020-12-11 | 南方医科大学南方医院 | Gastric cancer peritoneal metastasis prediction system and method based on enhanced CT (computed tomography) imaging omics |
-
2017
- 2017-08-24 CN CN201710734667.8A patent/CN107368695A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108038845A (en) * | 2017-12-01 | 2018-05-15 | 南方医科大学南方医院 | A kind of topological analysis householder method for abdominal CT images three-dimensional reconstruction |
| CN108847286A (en) * | 2018-06-04 | 2018-11-20 | 暨南大学附属第医院（广州华侨医院） | Tumour prediction technique based on image feature data |
| CN109259780A (en) * | 2018-07-24 | 2019-01-25 | 南方医科大学南方医院 | The aided assessment system and method that gastric cancer prognosis and chemotherapy based on enhancing CT images group benefit |
| CN109598266A (en) * | 2018-10-24 | 2019-04-09 | 深圳大学 | Lower-limb deep veins thrombus efficiency of thrombolysis prediction technique and system based on machine learning |
| CN109583447A (en) * | 2018-12-11 | 2019-04-05 | 山东省肿瘤防治研究院（山东省肿瘤医院） | A kind of image group credit analysis model building method and analysis method |
| CN112071418A (en) * | 2020-05-26 | 2020-12-11 | 南方医科大学南方医院 | Gastric cancer peritoneal metastasis prediction system and method based on enhanced CT (computed tomography) imaging omics |
| CN112071418B (en) * | 2020-05-26 | 2023-08-01 | 南方医科大学南方医院 | Gastric cancer peritoneal metastasis prediction system and method based on enhanced CT image histology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107368695A (en) | The construction method of GISTs malignant potential disaggregated model based on radiation group | |
| CN108694718A (en) | The same period new chemoradiation therapy curative effect evaluation system and method before rectal cancer | |
| CN105005714A (en) | Non-small cell lung cancer prognosis method based on tumor phenotypic characteristics | |
| Zhang et al. | Lung nodule classification in CT images using 3D DenseNet | |
| CN112767407B (en) | CT image kidney tumor segmentation method based on cascade gating 3DUnet model | |
| Wang et al. | Computed-tomography-based radiomics model for predicting the malignant potential of gastrointestinal stromal tumors preoperatively: a multi-classifier and multicenter study | |
| CN108922602A (en) | The same period new chemoradiation therapy curative effect evaluation system and method before rectal cancer based on big data analysis MRI image | |
| Maaref et al. | Predicting the response to FOLFOX-based chemotherapy regimen from untreated liver metastases on baseline CT: a deep neural network approach | |
| Chen et al. | CT radiomics model for discriminating the risk stratification of gastrointestinal stromal tumors: a multi-class classification and multi-center study | |
| Zou et al. | Multi-task deep learning based on T2-Weighted Images for predicting Muscular-Invasive Bladder Cancer | |
| CN107463964A (en) | A kind of tumor of breast sorting technique based on features of ultrasound pattern correlation, device | |
| CN112258511A (en) | Partial differential equation-based patient lung lesion prediction method and system | |
| González et al. | Deep convolutional neural network to predict 1p19q co-deletion and IDH1 mutation status from MRI in low grade gliomas | |
| Sharma et al. | A systematic review for detecting cancer using machine learning techniques | |
| Likhitkar et al. | Automated detection of cancerous lung nodule from the computed tomography images | |
| Qiu et al. | Comparison of comprehensive morphological and radiomics features of subsolid pulmonary nodules to distinguish minimally invasive adenocarcinomas and invasive adenocarcinomas in CT Scan | |
| Pocė et al. | Pancreas segmentation in CT images: state of the art in clinical practice | |
| Whitney et al. | 7 ARTIFICIAL INTELLIGENCE IN MEDICAL IMAGING | |
| CN110163195A (en) | Liver cancer divides group's prediction model, its forecasting system and liver cancer to divide group's judgment method | |
| CN113850788A (en) | System for judging bladder cancer muscle layer infiltration state and application thereof | |
| CN112396616A (en) | Osteosarcoma recurrence risk prediction model based on tissue morphology analysis | |
| CN114298957A (en) | Classification method, system and storage medium for breast molybdenum target image lesions | |
| Tang et al. | The value of combined PET/MRI, CT and clinical metabolic parameters in differentiating lung adenocarcinoma from squamous cell carcinoma | |
| Che | Improved nonalcoholic fatty liver disease diagnosis from ultrasound data based on deep learning | |
| Jaberipour et al. | Predicting local failure after stereotactic radiation therapy in brain metastasis using quantitative CT and machine learning |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171121 |
|
| WD01 | Invention patent application deemed withdrawn after publication |