WO2022011616A1 - 一种通过影像组学特征判断癌症治疗反应的方法及系统 - Google Patents

一种通过影像组学特征判断癌症治疗反应的方法及系统 Download PDF

Info

Publication number
WO2022011616A1
WO2022011616A1 PCT/CN2020/102206 CN2020102206W WO2022011616A1 WO 2022011616 A1 WO2022011616 A1 WO 2022011616A1 CN 2020102206 W CN2020102206 W CN 2020102206W WO 2022011616 A1 WO2022011616 A1 WO 2022011616A1
Authority
WO
WIPO (PCT)
Prior art keywords
radiomics
risk
features
metastasis
treatment
Prior art date
Application number
PCT/CN2020/102206
Other languages
English (en)
French (fr)
Inventor
张艺宝
黄宇亮
李晨光
吴昊
刘宏嘉
Original Assignee
北京肿瘤医院(北京大学肿瘤医院)
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京肿瘤医院(北京大学肿瘤医院) filed Critical 北京肿瘤医院(北京大学肿瘤医院)
Priority to PCT/CN2020/102206 priority Critical patent/WO2022011616A1/zh
Priority to CN202080001312.8A priority patent/CN112262440A/zh
Publication of WO2022011616A1 publication Critical patent/WO2022011616A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • 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
    • G06T2207/10104—Positron emission tomography [PET]
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30061—Lung
    • 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

Definitions

  • the invention relates to the field of radiotherapy equipment, in particular to a method and a system for judging cancer treatment response by radiomics features.
  • Radiotherapy is one of the main treatment methods for cancer, and there is a contradiction in radiation dosimetry between tumor control and organ damage.
  • the radiation effect of cancer is positively correlated with the radiation dose in the target area, but the increased radiation damage, such as the risk of pneumonia (lung radiation) in lung cancer, is also a key constraint to further increase the target dose, and it is also the main reason for the interruption or even failure of radiation therapy.
  • Lung radiotherapy is an inflammatory reaction that occurs after normal lung tissue is damaged by radiation.
  • the main clinical manifestation is diffuse alveolar damage. In severe cases, it may develop into radioactive pulmonary fibrosis.
  • the probability of tumor control and the risk of lung radiotherapy are both positively related to the dose.
  • Individualized early dynamic monitoring is not only conducive to the prevention and intervention of lung radiotherapy, but also to better radiotherapy effects with lower risks.
  • the critical dose thresholds for tumor control and lung radiotherapy vary greatly among individuals, and it is difficult to accurately grasp the risk of individualized lung radiotherapy while striving for a higher target dose and tumor control probability.
  • the main purpose of the present invention is to overcome the above-mentioned defects of the prior art, and to provide a method and system for judging cancer treatment response by radiomics features, so as to solve the risk prediction of tumor recurrence and metastasis, and the risk of radiation damage that cannot be solved by the prior art. Prediction problems.
  • One of the embodiments of the present invention provides a system for judging cancer treatment response based on radiomics features, which is characterized by comprising: a synthesis unit for matching a patient positioning image with a time series image in a course of treatment, The data is mapped to the time series images in the course of treatment, and the selected parts on the time series images are divided into corresponding regions of interest; the acquisition unit is used to extract the stable radiomics features in the patient images; the analysis unit is used to With the progress of the treatment, the patient's treatment response is evaluated according to the change trend of the stable radiomics characteristic parameter in the region of interest.
  • the stable radiomics features include temporally stable radiomics features; the temporally stable radiomics features refer to the same location at different times based on the same conditions
  • the radiomics features in the collected images are consistent, and are radiomics features with temporal stability.
  • the radiomic features with stability further include radiomic features with cross-modal equivalence; the radiomic features with cross-modal equivalence refer to the same object The same radiomics features of different image modalities at the same location are consistent, and they are radiomics features with cross-modal equivalence.
  • the radiomics signature with stability has a characteristic that the trend of change is associated with the gradually accumulated radiation dose.
  • the patient positioning images are matched with the time-series images in the session, including morphing registration method matching, or manual labeling method matching, or rigid registration method matching.
  • evaluating the patient's treatment response according to the change trend of the radiomics feature parameter in the region of interest along with the treatment progress including analyzing the risk of tumor recurrence and metastasis, when the risk value is higher than the threshold, the patient is currently in the current state. conditions with a high risk of cancer recurrence and metastasis.
  • the analyzing the risk of tumor recurrence and metastasis includes inputting stable radiomics features into a prediction model to obtain a result of whether the tumor recurrence and metastasis;
  • the patient data is used as the training set, and the relationship between the changing trend of radiomics features and tumor recurrence and metastasis is fitted to obtain the threshold of tumor recurrence and metastasis risk. get.
  • the treatment response of the patient is evaluated according to the change trend of the radiomics feature parameter in the region of interest along with the treatment progress, including analyzing the risk of radiation damage.
  • the risk value is higher than a threshold value, the patient is in the current condition The risk of radiation damage is higher.
  • the analyzing the risk of radiation injury includes inputting the stable radiomics features into a prediction model to obtain a result of whether the risk of radiation injury occurs;
  • the prediction model uses the patients with and without radiation injury by using The time series data of radiomics features in different regions of interest are used as the training set, and the relationship between the changing trend of radiomics features and the risk of radiation damage is fitted to obtain the threshold of radiation damage risk;
  • the prediction model takes the radiomics features as the input, and determines whether The risk of radiation damage is the output target, which is obtained by training.
  • One of the embodiments of the present invention provides a method for judging cancer treatment response by radiomics features, including: matching a patient positioning image with a time series image in a course of treatment, and mapping data on the positioning image to the time series image in the course of treatment , segment the selected part on the time series image into corresponding regions of interest; extract the stable radiomics features in the patient images; with the progress of the treatment, according to the stable radiomics feature parameters Regional trends were used to assess patient response to treatment.
  • the stable radiomics features include temporally stable radiomics features; the temporally stable radiomics features refer to the same location at different times based on the same conditions
  • the radiomics features in the collected images are consistent, and are radiomics features with temporal stability.
  • the radiomic features with stability further include radiomic features with cross-modal equivalence; the radiomic features with cross-modal equivalence refer to the same object The same radiomics features of different image modalities at the same location are consistent, and they are radiomics features with cross-modal equivalence.
  • the radiomics signature with stability has a characteristic that the trend of change is associated with the gradually accumulated radiation dose.
  • the patient positioning images are matched with the time-series images in the session, including morphing registration method matching, or manual labeling method matching, or rigid registration method matching.
  • the patient's treatment response is evaluated according to the change trend of the radiomics feature parameter in the region of interest, including analyzing the risk of tumor recurrence and metastasis.
  • the risk value is higher than the threshold, the patient is in the current condition. There is a high risk of cancer recurrence and metastasis.
  • the analyzing the risk of tumor recurrence and metastasis includes inputting stable radiomics features into a prediction model to obtain a result of whether the tumor recurrence and metastasis;
  • the patient data is used as the training set, and the relationship between the changing trend of radiomics features and tumor recurrence and metastasis is fitted to obtain the threshold of tumor recurrence and metastasis risk. get.
  • the treatment response of the patient is evaluated according to the change trend of the radiomics feature parameter in the region of interest along with the treatment progress, including analyzing the risk of radiation damage.
  • the risk value is higher than a threshold value, the patient is in the current condition The risk of radiation damage is higher.
  • the analyzing the risk of radiation injury includes inputting the stable radiomics features into a prediction model to obtain a result of whether the risk of radiation injury occurs; the prediction model uses the patients with and without radiation injury by using The time series data of radiomics features in different regions of interest are used as the training set, and the relationship between the changing trend of radiomics features and the risk of radiation damage is fitted, and the threshold value reflecting the risk of radiation damage is obtained. Whether the risk of radiation damage occurs is the output target, which can be obtained by training.
  • One of the embodiments of the present invention provides a method for constructing a risk prediction model for tumor recurrence and metastasis based on radiomics features, collecting data of patients with and without tumor recurrence and metastasis as a training set; using a deep learning model to fit changes in radiomics features The relationship between the trend and tumor recurrence and metastasis was trained to obtain a prediction model with radiomic features as input and tumor recurrence and metastasis as output target.
  • One of the embodiments of the present invention provides a method for constructing a radiation injury risk prediction model by using radiomics features, collecting time series data of radiomics features in different regions of interest of patients with and without radiation damage as a training set; using a deep learning model Fit the relationship between the trend of radiomics features and the risk of radiation damage, and train a predictive model that takes radiomics features as the input and whether the risk of radiation damage occurs as the output target.
  • FIG. 1 is a schematic diagram of a system for judging cancer treatment response by radiomics features according to some embodiments of the present invention
  • FIG. 2 is a schematic diagram of a method for judging cancer treatment response by radiomics features according to some embodiments of the present invention
  • FIG. 3 is a schematic diagram of a method for constructing a risk prediction model for tumor recurrence and metastasis by using radiomics features according to some embodiments of the present invention
  • FIG. 4 is a schematic diagram of a method for constructing a radiation damage risk prediction model based on radiomics features according to some embodiments of the present invention.
  • system is a method used to distinguish different components, elements, parts, parts or assemblies at different levels. However, other words may be replaced by other expressions if they serve the same purpose.
  • system and unit may be implemented by software or hardware, and may be a physical or virtual name having the functional part.
  • a system for judging cancer treatment response by radiomics features includes: a synthesis unit for matching patient positioning images with time-series images in a course of treatment, and mapping the data on the positioning images to the course of treatment On the time-series images, the selected parts on the time-series images are divided into corresponding regions of interest; the acquisition unit is used to extract the stable radiomics features in the patient images; the analysis unit is used to follow the progress of the treatment, according to the The trend of radiomics characteristic parameters in the region of interest was used to evaluate the patient's treatment response.
  • Time-series images include time-series CBCT images, or other imaging with similar functions.
  • the advantage of time-series CBCT image series is that it can provide multi-dimensional correlation information such as changes in radiomics features and cumulative radiation dose throughout the entire radiotherapy process.
  • a course of treatment refers to the entire course of treatment from the beginning of the treatment to the end of the treatment, including during the specific real-time radiation therapy, and after one or more radiation treatments.
  • Image-guided radiotherapy (IGRT) clinical practice has accumulated a large amount of multimodal medical data, including diagnostic CT (CT) before radiotherapy, localization CT, localization MR, localization PETCT, and time series cone beam CT (CBCT) acquired regularly during radiotherapy. series, as well as CT, MR, etc. for follow-up and evaluation after radiotherapy.
  • CT diagnostic CT
  • CBCT time series cone beam CT
  • series as well as CT, MR, etc. for follow-up and evaluation after radiotherapy.
  • CBCT imaging as a setup guidance modality, not only provides a large amount of historical data for the establishment of artificial intelligence models, but also contains more dimensional information such as space, time, dose, and biological effects; among them, the spatial dimension is reflected in the tumor target area.
  • the multimodal images before radiotherapy and the time-series CBCT acquired periodically during radiotherapy were used for annotation and dose reconstruction.
  • multimodal image data before treatment, all time series CBCT image series during treatment, and regular follow-up data such as regular chest enhanced CT at the end of treatment and after the end of treatment are collected.
  • Multi-modal image collection method due to the huge amount of multi-modal image data, in order to further improve the analysis efficiency and reduce human errors, including but not limited to the following implementation methods to collect data, for data in standard DICOM format, use the API interface provided by the planning system Automatically export, organize and analyze data in batches, such as converting patient isodose lines into structure files, batch calculation and accumulation of doses after deformation registration, etc.
  • the crawler tool is used to automatically capture clinical information such as tumor recurrence and metastasis, lung discharge diagnosis report, etc. in the patient's electronic medical record from the web page of the hospital information system (HIS), so as to avoid directly exporting electronic data from the database. Medical records, reducing risk and improving efficiency.
  • the patient positioning images are matched with the time-series images in the course of treatment, based on the deformation registration technology, or the manual labeling method or the rigid registration method.
  • Regions of interest (ROI) such as site tissue.
  • the deformation registration technology is not only used to map the labeled data of a single imaging modality to other modalities, but also used for the reconstruction and accumulation of the actual treatment dose of the patient.
  • the deformation registration parameters will be selected, and necessary manual confirmation and adjustment will be carried out on the basis of automatic evaluation to ensure the quality control requirements of deformation registration and the accuracy of the prediction model.
  • One of the embodiments of the present invention adopts the B-spline registration algorithm based on mutual information to adapt to the difference of pixel value distribution between multimodal image data, and other algorithms with similar functions can also be used.
  • the research will generate three-dimensional virtual deformation as a reference value by random number method, and apply it to the CBCT image of the phantom.
  • Deformation registration is performed between the CBCT image after the action and the CT image of the phantom, and the error between the calculated deformation vector and the known reference value is compared.
  • the error mean square is used as the evaluation index of the deformation registration accuracy, and the parameters of the deformation registration (such as gray level, resolution, iteration number, optimization algorithm, etc.) are fed back and adjusted, and the error mean square is reduced as much as possible through continuous iteration.
  • an acquisition unit for extracting stable radiomics features in patient images an embodiment is to extract CBCT radiomics features for tumor target areas and normal lung tissue respectively, and the specific method includes: After preprocessing such as data resampling, Gaussian smoothing, histogram homogenization, and resolution resampling, the radiomics features are extracted on the region of interest (ROI), including but not limited to extraction of volume and shape features, first-order statistical features, texture Features and wavelet analysis features and other four categories.
  • ROI region of interest
  • the treatment response of patients was evaluated according to the changing trend of radiomics characteristic parameters in the region of interest.
  • the treatment response refers to whether there is a risk of recurrence and metastasis of lung cancer or a risk of radiation damage with the increase of radiation dose.
  • To achieve early prediction of radiation damage or tumor recurrence and metastasis. Provide evidence-based support and scientific basis for major clinical decisions such as adaptive radiotherapy.
  • Embodiment 2 is a diagrammatic representation of Embodiment 1:
  • the stable radiomics features include temporally stable radiomics features; the temporally stable radiomics features refer to images in images collected at the same location under the same conditions at different times. Consistent omics features are radiomics features with temporal stability.
  • CCC consistency correlation coefficient
  • the present invention optimizes the preprocessing method.
  • Pyradiomics software is used to realize automatic calculation of radiomics features.
  • the stable radiomics features include one or a combination of analysis features such as volume and shape features, first-order statistical features, texture features, and wavelets.
  • the above-mentioned temporally stable radiomic feature screening method can also be used for the screening of cross-modal equivalence radiomic features.
  • CT computed tomography
  • MR nuclear magnetic resonance
  • PET electron emission computed tomography
  • Radiomics features extracting and validating the most robust and reproducible features.
  • the radiomics feature is considered to have stable repeatability and consistency.
  • the results of CCC will be used in part to support the selection and optimization of preprocessing methods to obtain as many candidate radiomics features as possible with robustness.
  • the feature of multimodal consistency is not considered, only temporal stability can be considered.
  • the time series CBCT series further screened out the eigenvalues with obvious changing trends among different radiotherapy fractions, corresponding to the gradually accumulated radiation dose.
  • the treatment response of the patient is evaluated according to the change trend of the radiomics characteristic parameter in the region of interest with the progress of the treatment, including analyzing the risk of tumor recurrence and metastasis.
  • the risk value is higher than the threshold value, the patient has a high recurrence of cancer under the current conditions. transfer risk.
  • the risk of tumor recurrence and metastasis was analyzed, and the risk value of tumor recurrence and metastasis was determined, which was compared with the risk threshold of tumor recurrence and metastasis.
  • the acquisition method of the risk value and the threshold value is not limited in the present invention.
  • One embodiment is to obtain the risk value and the threshold value through a prediction model to obtain whether the tumor recurs or metastasizes.
  • the stable radiomics features are input into the prediction model to obtain the result of whether the tumor has recurrence or metastasis.
  • the prediction model uses the data of patients with and without tumor recurrence and metastasis as a training set, and fits the relationship between the change trend of radiomics features and tumor recurrence and metastasis, and obtains a threshold for the risk of tumor recurrence and metastasis; when the radiomics features are input to predict After the model, a value at risk is obtained and the value at risk is compared with a threshold. The output is the result of whether the tumor recurs or metastasizes.
  • One of the embodiments of the present invention uses Scikit-Feature to select radiomics features, and the Scikit-Learn software package provides the selection of multiple prediction models, and conducts prediction model modeling for patient radiomics features and dose distribution data. Fold cross-validation, traverse a variety of feature selection algorithms and model combinations, and measure the prediction effect and robustness of the model by the AUC average and variance. The higher the AUC average, the better, and the lower the variance, the better.
  • the data of patients with and without tumor recurrence and metastasis were collected as a training set, and LASSO regression was used in the training set to fit the relationship between the change trend of radiomics characteristics and tumor recurrence and metastasis. Due to the L1 regularization of LASSO regression, variables with poor predictability will be excluded from the model, and the final output of the model is a linear combination of a few variables, which can be used as a discriminant function for tumor recurrence and metastasis. Calculate the value of the discriminant function in the training set of two groups of patients with and without recurrence and metastasis, compare their histograms, and take the discriminant function corresponding to the intersection of the two histograms as the threshold.
  • the discriminant function (risk value) is higher than this threshold, it is considered that the patient has a high risk of recurrence and metastasis under the current conditions, and further adaptive radiotherapy strategies such as increasing the prescribed dose to the target area should be adopted for timely intervention.
  • the prediction model includes but is not limited to the above-mentioned models, and can also include other models that can achieve the purpose of the present invention.
  • Prediction model selection method calculate the ROC curve of each model, measure the effectiveness of the model by the area under the ROC curve, and compare the feature selection algorithm and classifier combination with the most predictive effectiveness. Based on the best model, we find the critical threshold of radiomics for predicting whether tumor recurrence and metastasis, and whether radiation damage such as lung radiation occurs.
  • the methods of radiation damage risk analysis and model establishment are the same as those in the second embodiment, except that the composition of the training set is different.
  • the prediction model uses time series data of radiomics features of different regions of interest of patients with and without radiation damage as a training set.
  • a specific implementation is: using isodose lines to divide the normal lung tissue on the CBCT image series of each treatment grade of the patient into sub-regions, based on different combinations of various feature selection algorithms and various classifiers, using lung release to generate Follow-up data such as whether or not are associated with radiomic features to establish a machine learning prediction model.
  • the predictive model obtains radiation damage, such as the occurrence of radiation pneumonitis, which can guide clinical decisions such as the timing and spatial positioning of adaptive radiotherapy.
  • radiation damage such as the occurrence of radiation pneumonitis
  • the relatively low-risk lung sub-regions are used to "share" high-risk sub-regions. Part of the dose, thereby effectively reducing the overall risk of lung radiotherapy without sacrificing the target dose and tumor control rate.
  • Embodiment 4 is a diagrammatic representation of Embodiment 4:
  • a method for judging cancer treatment response by radiomics features corresponds one-to-one with a system for judging cancer treatment response by radiomics features, see the description of the system embodiment above.
  • Embodiment 5 is a diagrammatic representation of Embodiment 5:
  • a method for constructing a risk prediction model for tumor recurrence and metastasis through radiomics features collecting data of patients with and without tumor recurrence and metastasis as a training set; using a deep learning model to fit the relationship between changes in radiomics features and tumor recurrence and metastasis , trained to obtain a prediction model with radiomic features as input and tumor recurrence and metastasis as output target.
  • the method can be applied to a device, or a system.
  • Embodiment 6 is a diagrammatic representation of Embodiment 6
  • a method for constructing a radiation injury risk prediction model based on radiomics features collecting time series data of radiomics features in different regions of interest of patients with and without radiation damage as a training set; using a deep learning model to fit changes in radiomics features
  • the relationship between the trend and the risk of radiation damage is trained to obtain a prediction model with radiomics features as input and whether radiation damage risk occurs as output target.
  • the method can be applied to a device, or a system.
  • Embodiments 5 and 6 are the same, except that the composition of the training set and the input and output of the model are different. Taking the metastasis and radiation damage prediction of lung cancer as an example, other cancers can refer to this method:
  • Multi-modal imaging such as CT and CBCT were performed on the chest simulation phantom, and the lung area was marked after rigid registration or deformation registration or manual labeling of the images.
  • the discriminant function value in the validation set was calculated by the prediction model of tumor recurrence and metastasis and lung radiotherapy, respectively, and the validation set was divided into two groups according to whether it was greater than the critical threshold. Meier curve, log-rank test was used to check whether the difference between the two curves for each model was statistically significant. If there is, it proves that the model performs well, and further Cox regression analysis can be performed to predict the change of tumor recurrence rate or the incidence of lung radiotherapy over time.
  • Iterative improvement of the prediction model Due to differences in anatomical structures such as changes in the volume, shape, and contents of cavity organs and localized CT, the actual dose of the patient and the dose distribution of the radiotherapy plan may exist due to the patient's weight loss during the course of radiotherapy, the reduction of the target volume, and the change in the volume, shape, and content of the cavity. different.
  • the dose reconstruction is performed by techniques such as deformation registration with the localization CT, and the difference and position distribution of the actual dose and the planned dose of the patient are analyzed, and the change trend and outcome of the radiomics characteristics are correlated. May provide a dosimetric explanation for the predictive results of the radiomics model. At the same time, this data can also be used as input to further iterate and improve the predictive accuracy and power of the model.
  • the possible beneficial effects of the embodiments of the present invention include, but are not limited to, guiding the clinical timely discovery of cases and accurate locations in which tumor target areas and/or normal lung tissues are close to the threshold warning range.
  • the probability of tumor control can be improved by increasing the number of treatments and the prescribed dose, and at the same time, the dynamic changes of radiation damage characteristic values can be strictly monitored.
  • adaptive radiotherapy can be used. Using the technology, other normal tissues with relatively low risk can be used to share part of the dose in time, so as to obtain better target dose and curative effect with an overall lower radiation risk.
  • This method can also be applied to various cancer types such as lung cancer, providing new scientific guidelines and technical means for individualized precision radiotherapy, and at the same time forming evidence-based medical evidence for Chinese people that can be radiated and promoted to the grassroots level, and promote wider medical equity and access to quality medical resources. It is beneficial to clinical promotion and industrial application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biomedical Technology (AREA)
  • Theoretical Computer Science (AREA)
  • Medical Informatics (AREA)
  • General Physics & Mathematics (AREA)
  • Public Health (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Pathology (AREA)
  • Data Mining & Analysis (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Radiation-Therapy Devices (AREA)

Abstract

一种通过影像组学特征判断癌症治疗反应的系统，包括：合成单元，用于将患者定位图像和疗程中的时序图像进行匹配，将定位图像上的数据映射到疗程中的时序图像上，将时序图像上选定的部位分割成相应多个感兴趣区；获取单元，用于提取患者影像中具备稳定性的影像组学特征；分析单元，用于随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。这一系统利用稳定性好、可重复性高、随治疗分次和累积剂量变化趋势明显的时序影像组学特征值；获得肿瘤复发转移或正常组织辐射损伤发生的早期个体化预测，提高肿瘤控制概率的同时降低正常组织辐射损伤风险和肿瘤复发转移风险。

Description

一种通过影像组学特征判断癌症治疗反应的方法及系统 技术领域
本发明涉及放射治疗设备领域，特别涉及一种通过影像组学特征判断癌症治疗反应的方法及系统。
背景技术
放疗是癌症的主要治疗手段之一，肿瘤控制和器官损伤存在放射剂量学矛盾。癌症的放疗效果与靶区辐射剂量正相关，但随之增加的放射性损伤，比如肺癌中的肺炎(放肺)风险同是进一步提高靶区剂量的关键制约因素，也是导致放疗中断甚至失败的主要原因之一。放肺是正常肺组织受到辐射损伤后出现的炎症反应，临床主要表现为弥漫性肺泡损伤，严重的有可能会发展为放射性肺纤维化，在肺癌、食管癌等胸部肿瘤放疗患者中的发病率和死亡率较高。肿瘤的控制概率和放肺的发生风险均与剂量正相关，个体化早期动态监控不仅有利于放肺的预防和干预，而且有利于以更低的风险获得更好的放疗效果。但肿瘤控制和放肺发生的临界剂量阈值的个体差异很大，临床难以准确把握个体化放肺风险的同时争取更高的靶区剂量和肿瘤控制概率。
放疗中的严重损伤，比如放射性肺炎的预测成为放疗过程中亟待解决的技术问题。癌症难治愈，最大的风险是容易肿瘤复发转移，肿瘤复发转移早期预测具有非常重要的现实意义，也是亟待解决的问题。
发明内容
本发明的主要目的在于克服上述现有技术的缺陷，提供了一种通过影像组学特征判断癌症治疗反应的方法及系统，以解决现有技术无法解决的肿瘤复发转移风险预测，及放射性损伤风险预测问题。
本发明实施例之一提供一种通过影像组学特征判断癌症治疗反应的系统，其特征在于，包括：合成单元，用于将患者定位图像和疗程中的时序图像进行匹配，将定位图像上的数据映射到疗程中的时序图像上，将时序图像上选定的部位分割成相应多个感兴趣区；获取单元，用于提取患者影像中具备稳定性的影像组学特征；分析单元，用于随治疗进度，根据所述具备稳定性的影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。
在一些实施例中，所述具备稳定性的影像组学特征，包括具有时间稳定性的影像组学特征；所述具有时间稳定性的影像组学特征指，在不同时间，基于相同条件相同位置采集的图像中的影像组学特征具备一致性的，为具备时间稳定性的影像组学特征。
在一些实施例中，所述具备稳定性的影像组学特征，还包括具有跨模态等价性的影像组学特征；所述具有跨模态等价性的影像组学特征指，同一对象相同位置不同图像模态的同一影像组学特征具备一致性的，为具有跨模态等价性的影像组学特征。
在一些实施例中，具备稳定性的影像组学特征具有变化趋势与逐渐累积的辐射剂量相关联的特点。
在一些实施例中，将患者定位图像和疗程中的时序图像进行匹配，包括形变配准方法匹配、或手工标注方法匹配、或刚性配准方法匹配。
在一些实施例中，所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析肿瘤复发转移风险，当风险值高于阈值时，该患者在当前条件下有癌症高复发转移风险。
在一些实施例中，所述分析肿瘤复发转移风险包括，将具备稳定性的影像组学特征输入预测模型，得出肿瘤是否复发转移的结果；所述预测模型通过发生和未发生肿瘤复发转移的患者数据为训练集，拟合影像组学特征变化趋势与肿瘤复发转移的关系，得出肿瘤复发转移风险的阈值；预测模型以影像组学特征为输入，以肿瘤是否复发转移为输出目标，训练得到。
在一些实施例中，所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析放射性损伤风险，当风险值高于阈值时，该患者在当前条件下放射性损伤风险较高。
在一些实施例中，所述分析放射性损伤风险包括，将具备稳定性的影像组学特征输入预测模型，得出是否发生放射性损伤风险的结果；所述预测模型通过利用发生和未发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集，拟合影像组学特征变化趋势与放射性损伤风险的关系，得出放射性损伤风险的阈值；预测模型以影像组学特征为输入，以是否发生放射性损伤风险为输出目标，训练得到。
本发明实施例之一提供一种通过影像组学特征判断癌症治疗反应的方法，包括：将患者定位图像和疗程中的时序图像进行匹配，将定位图像上的数据映射到疗程中的时序图像上，将时序图像上选定的部位分割成相应多个感兴趣区；提取患者影像中具备稳定性的影像组学特征；随治疗进度，根据所述具备稳定性的影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。
在一些实施例中，所述具备稳定性的影像组学特征，包括具有时间稳定性的影像组学特征；所述具有时间稳定性的影像组学特征指，在不同时间，基于相同条件相同位置采集的图像中的影像组学特征具备一致性的，为具备时间稳定性的影像组学特征。
在一些实施例中，所述具备稳定性的影像组学特征，还包括具有跨模态等价性的影像组学特征；所述具有跨模态等价性的影像组学特征指，同一对象相同位置不同图像模态的同一影像组学特征具备一致性的，为具有跨模态等价性的影像组学特征。
在一些实施例中，具备稳定性的影像组学特征具有变化趋势与逐渐累积的辐射剂量相关联的特点。
在一些实施例中，将患者定位图像和疗程中的时序图像进行匹配，包括形变配准方法匹配、或手工标注方法匹配、或刚性配准方法匹配。
在一些实施例中，随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析肿瘤复发转移风险，当风险值高于阈值时，该患者在当前条件下有癌症高复发转移风险。
在一些实施例中，所述分析肿瘤复发转移风险包括，将具备稳定性的影像组学特征输入预测模型，得出肿瘤是否复发转移的结果；所述预测模型通过发生和未发生肿瘤复发转移的患者数据为训练集，拟合影像组学特征变化趋势与肿瘤复发转移的关系，得出肿瘤复发转移风险的阈值；预测模型以影像组学特征为输入，以肿瘤是否复发转移为输出目标，训练得到。
在一些实施例中，所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析放射性损伤风险，当风险值高于阈值时，该患者在当前条件下放射性损伤风险较高。
在一些实施例中，所述分析放射性损伤风险包括，将具备稳定性的影像组学特征输入预测模型，得出是否发生放射性损伤风险的结果；所述预测模型通过利用发生和未发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集，拟合影像组学特征变化趋势与放射性损伤风险的关系，得出反映放射性损伤风险的阈值；预测模型以影像组学特征为输入，以是否发生放射性损伤风险为输出目标，训练得到。
本发明实施例之一提供一种通过影像组学特征构建肿瘤复发转移风险预测模型的方法，收集发生和未发生肿瘤复发转移的患者数据为训练集；利用深度学习模型拟合影像组学特征变化趋势与肿瘤复发转移的关系，训练得出以影像组学特征为输入，以肿瘤是否复发转移为输出目标的预测模型。
本发明实施例之一提供一种通过影像组学特征构建放射性损伤风险 预测模型的方法，收集发生和不发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集；利用深度学习模型拟合影像组学特征变化趋势与放射性损伤风险的关系，训练得出以影像组学特征为输入，以是否发生放射性损伤风险为输出目标的预测模型。
本发明的方案有益效果如下：
从大量时序图像系列中筛选出稳定性好、可重复性高、随累积剂量变化趋势明显的影像组学特征，利用深度学习模型拟合影像组学特征变化趋势与放射性损伤风险及肿瘤复发转移风险的关系。临床结局进行关联并建立早期预测模型，从时序影像组学动态变化中寻找并验证肿瘤控制和放射损伤的个体化判别规则。通过阈值与剂量的对应关系指导自适应精确放疗等临床实践，以更低辐射损伤风险获得更好的放疗疗效。为自适应放疗等重大临床决策提供循证支持和科学依据。
附图说明
本发明将以示例性实施例的方式进一步说明，这些示例性实施例将通过附图进行详细描述。这些实施例并非限制性的，在这些实施例中，相同的编号表示相同的结构，其中：
图1是根据本发明一些实施例所示的通过影像组学特征判断癌症治疗反应的系统示意图；
图2是根据本发明一些实施例所示的通过影像组学特征判断癌症治疗反应的方法示意图；
图3是根据本发明一些实施例所示的一种通过影像组学特征构建肿瘤复发转移风险预测模型的方法示意图；
图4是根据本发明一些实施例所示的一种通过影像组学特征构建放射性损伤风险预测模型的方法示意图。
具体实施方式
为了更清楚地说明本发明实施例的技术方案，下面将对实施例描述中所需要使用的附图作简单的介绍。显而易见地，下面描述中的附图仅仅是本发明的一些示例或实施例，各个实施例的技术特征之间可以相互组合，构成实现发明目的的实际方案，对于本领域的普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图将本发明应用于其它类似情景。除非从语言环境中显而易见或另做说明，图中相同标号代表相同结构或操作。
应当理解，本文使用的“系统”、“单元”是用于区分不同级别的不同组件、元件、部件、部分或装配的一种方法。然而，如果其他词语可实现相同的目的，则可通过其他表达来替换所述词语。并且，“系统”、“单元”可以由软件或者硬件实施，可以是实体或虚拟的具有该功能部分的称呼。
本发明中使用了流程图用来说明根据本发明的实施例的系统所执行的操作。应当理解的是，前面或后面操作不一定按照顺序来精确地执行。相反，可以按照倒序或同时处理各个步骤。同时，也可以将其他操作添加到这些过程中，或从这些过程移除某一步或数步操作。各个实施例中的技术方案可以相互组合实现本发明的目的。
实施例一：
一种通过影像组学特征判断癌症治疗反应的系统，如图1所示，包括：合成单元，用于将患者定位图像和疗程中的时序图像进行匹配，将定位图像上的数据映射到疗程中的时序图像上，将时序图像上选定的部位分割成相应多个感兴趣区；获取单元，用于提取患者影像中具备稳定性的影像组学特征；分析单元，用于随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。
时序图像包括时序CBCT图像、或者其它具有类似功能的成像等，时序CBCT图像系列的优势在于可提供放疗全程的影像组学特征变化和辐射累积剂量等多维度关联信息。
疗程中指的是从治疗开始到治疗结束的整个疗程，包括具体的实时放射治疗过程中，及一次或几次放射治疗后。
图像引导放疗(IGRT)临床实践中积累了大量多模态医疗数据，包括放疗前的诊断CT(CT)、定位CT、定位MR、定位PETCT，放疗过程中定期获取的时序锥束CT(CBCT)系列，以及放疗后用于随访和评效的CT、MR等。其中，CBCT成像作为摆位引导模态，不仅可为人工智能模型的建立提供大量历史数据，而且包含了空间、时间、剂量、生物效应等更多维度信息；其中，空间维度体现于肿瘤靶区和正常器官等解剖结构的形状、大小、相对位置等在放疗疗程中的变化情况，相较治疗前的定位CT等图像更加接近患者治疗当天的实际状态；时间和剂量等维度信息则主要体现于CBCT所反映的人体组织与射线的相互作用过程、生物学效应发生的时间延迟性及其与累积辐射剂量的正相关性等动态变化。CBCT的部分影像组学特征(如GLCM特征)具有稳定的可重复性和跨模态等价性，能补充甚至替代CT等传统数据用于影像组学分析，对于肺癌患者的生存具有预测 价值。相较治疗前的CT图像，CBCT影像组学特征对于立体定向放疗中的放肺风险或者其它放疗风险具有更好的预测效果。
如果治疗前的图像不是CBCT影像，以放疗前的多模态影像与患者放疗过程中定期获取的时序CBCT进行标注和剂量重建。比如，肺癌放疗患者搜集治疗前的多模态影像数据，治疗过程的全部时序CBCT影像系列，以及治疗结束时、结束后的定期胸部增强CT等常规随访数据。多模态影像搜集方式，由于多模态影像数据量庞大，为进一步提高分析效率并减少人为误差，包括不限于采用以下实施方式搜集数据，对于标准DICOM格式的数据，利用计划系统提供的API接口对数据进行批量自动导出、整理和分析，比如将患者等剂量线转换为结构文件、形变配准后的批量剂量计算及累积等。对于非标准化多模态医学信息，通过爬虫工具从医院信息系统(HIS)的Web网页中自动抓取患者电子病历中的肿瘤复发转移、放肺诊断报告等临床信息，从而避免从数据库直接导出电子病历，降低风险的同时也提高了效率。
将患者定位图像和疗程中的时序图像进行匹配，基于形变配准技术、或手工标注方法或刚性配准方法匹配。将定位CT图像上的专家标注信息、治疗计划中提取的等剂量线梯度等数据映射至CBCT图像，并利用自动分割技术标注CBCT图像上的正常部位组织，从而确定靶区、高剂量区和正常部位组织等感兴趣区(ROI)。
形变配准技术在本发明中不仅用于将单一影像模态的标注数据映射到其他模态，也用于患者实际治疗剂量的重建和累积。为达到优化的配准效果，将精选形变配准参数，并在自动评估的基础上进行必要的人工确认和调整，以保证形变配准的质控需求和预测模型的准确性。本发明实施例之一采用基于互信息的B样条配准算法以适应多模态影像数据间像素值分布的差异性，也可以其它采用具有类似功能的算法。为定量评估形变配准精度并为参数调试提供依据，研究将通过随机数方法生成三维虚拟形变作为参考值，并将其作用到模体的CBCT图像上。将作用后的CBCT图像与模体的CT图像进行形变配准，比较计算得到的形变向量与已知参考值的误差。将误差均方作为形变配准精度的评价指标，反馈调试形变配准的参数(如灰度级数、分辨率、迭代次数、优化算法等)，通过不断迭代以尽量缩小误差均方。
获取单元，用于提取患者影像中具备稳定性的影像组学特征；一种实施例为，分别对于肿瘤靶区和正常肺组织进行CBCT影像组学特征的提取，具体 方法包括：经过灰度级数重采样、高斯平滑、直方图均匀化、分辨率重采样等预处理后，在感兴趣区(ROI)上提取影像组学特征，包括不限于提取体积及形状特征、一阶统计特征、纹理特征和小波分析特征等四大类。
根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。所述的治疗反应指的是随着放射剂量的增加，是否存在肺癌放疗复发转移风险或者放射性损伤风险。做到放射性损伤或肿瘤复发转移提前预测。为自适应放疗等重大临床决策提供循证支持和科学依据。
实施例二：
所述具备稳定性的影像组学特征，包括具有时间稳定性的影像组学特征；所述具有时间稳定性的影像组学特征指，在不同时间，基于相同条件相同位置采集的图像中的影像组学特征具备一致性的，为具备时间稳定性的影像组学特征。
同时避免重要原始信息的损失，并利用一致性相关系数(CCC)来评价影像组学特征的时间稳定性和跨模态等价性，当致性相关系数(CCC)大于某一个数值时，影像组学特征具备一致性。
为了评估和筛选稳定的CBCT影像组学特征，本发明优化预处理方法，实施例之一采用Pyradiomics软件实现影像组学特征的自动计算，采用不同时间基于相同条件，采集的前后两次CBCT图像，对相同位置的ROI提取影像组学特征。对于每一个特征，分别计算前后两次数值的一致性相关系数(Concordance Correlation Coefficient，CCC)，定义为：
Figure PCTCN2020102206-appb-000001
其中，S x或S y为X或Y变量的标准差，S yx为X与Y的协方差，
Figure PCTCN2020102206-appb-000002
或
Figure PCTCN2020102206-appb-000003
表示平均值。不同时间图像上同一影像组学特征的CCC大于0.90时，该影像组学特征被认为具有稳定的可重复性，具备一致性。所述具备稳定性的影像组学特征，包括体积及形状特征、一阶统计特征、纹理特征、小波等分析特征中的一种或几种的组合。
上述时间稳定性的影像组学特征筛选方法，还可以用于跨模态等价性的影像组学特征的筛选。以电子计算机X射线断层扫描技术(CT)、核磁共振(MR)、电子发射计算机断层显像(PET)等传统多模态影像为参照，从时序CBCT 图像中寻找具有跨模态等价性的影像组学特征，提取并验证最具稳定性和可重复性的特征。当不同模态图像上同一影像组学特征的CCC大于0.90时，该影像组学特征被认为具有稳定的可重复性，具备一致性。CCC的结果将部分用于支持预处理方法的选择和优化，从而尽可能多地获得具备稳定性的备选影像组学特征。
如果患者一直采用一种图像模态，不用考虑多模态一致性这个特征，只考虑时间稳定性即可。
分别以肿瘤靶区和正常肺组织为目标，从时序CBCT系列中进一步筛选出不同放疗分次间变化趋势明显的特征值，对应逐渐累积的辐射剂量。
在获得稳定的CBCT影像组学特征后，分别基于肿瘤靶区和正常肺组织，在时序CBCT系列上的相同ROI位置，计算影像组学特征随治疗分次和辐射累积剂量的变化趋势，并分别与肿瘤复发转移与否、以及放肺的发生与否等临床结局进行关联。
所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析肿瘤复发转移风险，当风险值高于阈值时，该患者在当前条件下有癌症高复发转移风险。分析肿瘤复发转移风险，确定肿瘤复发转移风险值，与肿瘤复发转移风险阈值比较。风险值及阈值的获取方式本发明不做限制。
一种实施方式为通过预测模型获得风险值和阈值，获得肿瘤是否复发转移。将具备稳定性的影像组学特征输入预测模型，得出肿瘤是否复发转移的结果。
所述预测模型通过发生和未发生肿瘤复发转移的患者数据为训练集，拟合影像组学特征变化趋势与肿瘤复发转移的关系，得出肿瘤复发转移风险的阈值；当影像组学特征输入预测模型后，获得一个风险值，风险值与阈值进行比较。输出为肿瘤是否复发转移的结果。
本发明实施例之一采用Scikit-Feature进行影像组学特征的选择，及Scikit-Learn软件包提供多种预测模型的选择，对患者影像组学特征及剂量分布数据进行预测模型建模，通过5折交叉验证，遍历多种特征选择算法与模型组合，通过AUC平均值和方差来度量模型的预测效果及鲁棒性，AUC平均值越高越好，方差越低越好。
搜集发生和未发生肿瘤复发转移的患者数据作为训练集，在训练集中利用LASSO回归，拟合影像组学特征变化趋势与肿瘤复发转移的关系。由于LASSO回归的L1正则化，预测性较差的变量会被剔除出模型，模型的最终输出为少数变量的线性组合，可作为肿瘤复发转移的判别函数。统计该判别函数在训练集中有和无复发转移两组病人的取值，比较其直方图，以两个直方图的交点对应的判别函数取值为阈值。当判别函数(风险值)高于该阈值时，认为患者在当前条件下有高复发转移风险，需进一步采取增加靶区处方剂量等自适应放疗策略进行及时干预。
预测模型包括不限于上述模型，还可以包括其它可以实现本发明目的的模型。预测模型的选择方式，计算各模型的ROC曲线，以ROC曲线下面积衡量模型的效力，比较得出最具预测效力的特征选择算法和分类器组合。基于最佳模型寻找预测肿瘤是否复发转移，放射性损伤比如放肺是否发生的影像组学临界阈值。
实施例三：
关于放射性损伤风险分析及模型建立的方法与实施例二相同，差别为训练集的组成不同。
所述预测模型通过利用发生和未发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集。一种具体实施方式为：利用等剂量线将患者各治疗分次的CBCT图像系列上的正常肺组织分割成亚区，基于各种特征选择算法和各种分类器的不同组合，利用放肺发生与否等随访数据关联影像组学特征建立机器学习预测模型。
该预测模型获得放射性损伤，比如放射性肺炎是否发生，可指导自适应放疗的时机和空间定位等临床决策，通过及时重新设计放疗计划，利用相对低风险的肺部亚区“分担”高风险亚区的部分剂量，从而在不牺牲靶区剂量和肿瘤控制率的同时，有效降低整体放肺风险。
实施例四：
一种通过影像组学特征判断癌症治疗反应的方法与一种通过影像组学特征判断癌症治疗反应的系统一一对应，参见上文系统实施例说明。
实施例五：
一种通过影像组学特征构建肿瘤复发转移风险预测模型的方法，收集发生和未发生肿瘤复发转移的患者数据为训练集；利用深度学习模型拟合影像组学特征变化趋势与肿瘤复发转移的关系，训练得出以影像组学特征为输入，以肿瘤是否复发转移为输出目标的预测模型。该方法可以应用于设备、或系统中。
实施例六：
一种通过影像组学特征构建放射性损伤风险预测模型的方法，收集发生和不发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集；利用深度学习模型拟合影像组学特征变化趋势与放射性损伤风险的关系，训练得出以影像组学特征为输入，以是否发生放射性损伤风险为输出目标的预测模型。该方法可以应用于设备、或系统中。
实施例五和六的方法一样，只是训练集的构成和模型输入、输出不同。以肺癌的转移和放射性损伤预测举例，其它癌症可以参照此方法：
一、影像组学特征稳定性与跨模态等价性验证方法
(1)对胸部仿真模体分别进行CT、CBCT等多模态成像，对图像进行刚性配准或形变配准或手工标注后，标注肺部区域。
(2)对图像进行预处理后提取肺部区域影像组学特征，比较不同模态的差异，计算一致性相关系数(CCC)。
(3)优化预处理手段，使尽可能多的影像组学特征满足不同模态间CCC>0.9。
(4)间隔一段时间对同一模体再次成像，重复(1-2)步。评估多模态影像组学特征的时间稳定性。优化预处理手段，使尽可能多的影像组学特征具备时间稳定性，即CCC>0.9。
(5)最终筛选出具备跨模态等价性(一致性)与时间稳定性(一致性)的时序CBCT系列影像组学特征。
上述步骤顺序可以改变。先时间稳定，后跨模态也是可以的。
二、肿瘤复发转移与放肺预测方法
(1)从临床数据库挑选并导出患者治疗前多模态图像、治疗中时序CBCT图像系列、放疗计划、剂量分布、靶区与危及器官标注、临床随访结果等数据，分为训练集和测试集。
(2)对治疗前多模态图像和治疗中时序CBCT图像系列进行形变配准，将定位CT上专家标注的结构映射到CBCT上，利用等剂量线将正常肺组织分割成相应亚区。
(3)计算各时间点多模态图像靶区、肺部ROI的影像组学特征，分析上一步骤筛选出的特征相对于治疗前基线值的变化趋势。
(4)采用不同的特征选择算法与分类器组合，输入上一步中影像组学特征治疗前后的变化值，对肿瘤复发转移和放肺进行预测。采用5-折交叉验证方法来评估不同模型的ROC平均值及标准差，优选鲁棒性最佳的模型，并结合临床结局寻找影像组学特征变化的阈值，当变化超过阈值时提示具有肿瘤复发转移或放肺风险。
(5)监测治疗过程中各影像组学特征相对于基线值的变化趋势，通过前述模型评估剂量累积过程中各治疗分次对应的肿瘤复发转移或放肺的风险值，利用对应的个体化阈值指导自适应放疗等临床决策。
上述步骤顺序可以根据实际情况增减，或者改变顺序。
三、模型验证方法及改进手段
(1)采上一步骤获得的模型，基于测试集多模态图像分别进行肿瘤复发转移和放肺风险评估，按照评估出的风险与阈值的大小关系将患者分为高危组和低危组。
(2)回顾性分析高危组和低危组的肿瘤复发转移和放肺情况，采用对数秩检验分析两条Kaplan-Meier曲线的差别是否具有统计学意义以验证模型的准确性。
(3)在定位CT与各治疗分次时序CBCT形变配准的图像上，分别重建并累积患者实际接受的治疗剂量。
(4)将实际剂量与计划剂量的偏差与影像组学特征的变化进行关联分析，为影像组学模型提供剂量学解释，并将剂量偏差作为模型输入，进一步提高模型预测的准确率。
分别采用肿瘤复发转移和放肺预测模型计算验证集中的判别函数值，根据是否大于临界阈值将验证集分为两组，分别画出以无肿瘤复发或无放肺生存率为纵轴的Kaplan-Meier曲线，用对数秩检验检查各个模型的两条曲线的差异是否有统计学意义。如有，则证明模型表现良好，进一步可做Cox回归分析，预测肿瘤复发率或放肺发生率随时间的变化。
预测模型迭代改进：由于患者在放疗疗程中可能出现体重减轻，靶 区缩小，腔体器官体积、形状、内容物变化等解剖结构与定位CT的差异，患者实际剂量与放疗计划的剂量分布可能存在不同。在代表患者真实情况的CBCT图像上，通过与定位CT的形变配准等技术进行剂量重建，分析患者实际剂量与计划剂量的差异和位置分布，与影像组学特征的变化趋势和结局进行关联，可能为影像组学模型的预测结果提供剂量学解释。同时，该数据也可作为输入，进一步迭代和改进模型的预测精度和效力。
本发明实施例可能带来的有益效果包括但不限于：指导临床及时发现肿瘤靶区和/或正常肺组织接近阈值警示范围的病例和准确位置。对于发现具有复发和转移倾向的高危患者和靶区，可以通过增加治疗次数和处方剂量等方法改善肿瘤控制概率，同时严格监控放射损伤特征值动态变化，对于高危正常组织，可通过自适应放疗等技术，及时利用风险相对较低的其他正常组织分担部分剂量，从而以整体更低的辐射风险获得更好的靶区剂量和疗效。该方法也可以被应用到肺癌等各种癌种，为个体化精确放疗提供新的科学指引和技术手段的同时，形成可向基层辐射和推广的国人循证医学证据，促进更广泛的医疗公平和优质医疗资源可及性。有利于临床推广和产业应用。
需要说明的是，不同实施例可能产生的有益效果不同，在不同的实施例里，可能产生的有益效果可以是以上任意一种或几种的组合，也可以是其他任何可能获得的有益效果。
上文已对基本概念做了描述，显然，对于本领域技术人员来说，上述详细披露仅仅作为示例，而并不构成对本发明的限定。
此外，除非权利要求中明确说明，本发明所述处理元素和序列的顺序、数字字母的使用、或其他名称的使用，并非用于限定本发明流程和方法的顺序。尽管上述披露中通过各种示例讨论了一些目前认为有用的发明实施例，但应当理解的是，该类细节仅起到说明的目的，附加的权利要求并不仅限于披露的实施例，相反，权利要求旨在覆盖所有符合本发明实施例实质和范围的修正和等价组合。
同理，应当注意的是，为了简化本发明披露的表述，从而帮助对一个或多个发明实施例的理解，前文对本发明实施例的描述中，有时会将多种特征归并至一个实施例、附图或对其的描述中。但是，这种披露方法并不意味着本发明对象所需要的特征比权利要求中提及的特征多。
最后，应当理解的是，本发明中所述实施例仅用以说明本发明实施例的原则。其他的变形也可能属于本发明的范围。因此，作为示例而非限制，本 发明实施例的替代配置可视为与本发明的教导一致。相应地，本发明的实施例不仅限于本发明明确介绍和描述的实施例。

Claims (20)

  1. 一种通过影像组学特征判断癌症治疗反应的系统，其特征在于，包括：合成单元，用于将患者定位图像和疗程中的时序图像进行匹配，将定位图像上的数据映射到疗程中的时序图像上，将时序图像上选定的部位分割成相应多个感兴趣区；获取单元，用于提取患者影像中具备稳定性的影像组学特征；分析单元，用于随治疗进度，根据所述具备稳定性的影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。
  2. 根据权利要求1所述的系统，其特征在于，所述具备稳定性的影像组学特征，包括具有时间稳定性的影像组学特征；所述具有时间稳定性的影像组学特征指，在不同时间，基于相同条件相同位置采集的图像中的影像组学特征具备一致性的，为具备时间稳定性的影像组学特征。
  3. 根据权利要求2所述的系统，其特征在于，所述具备稳定性的影像组学特征，还包括具有跨模态等价性的影像组学特征；所述具有跨模态等价性的影像组学特征指，同一对象相同位置不同图像模态的同一影像组学特征具备一致性的，为具有跨模态等价性的影像组学特征。
  4. 根据权利要求1所述的系统，其特征在于，具备稳定性的影像组学特征具有变化趋势与逐渐累积的辐射剂量相关联的特点。
  5. 根据权利要求1所述的系统，其特征在于，将患者定位图像和疗程中的时序图像进行匹配，包括形变配准方法匹配、或手工标注方法匹配、或刚性配准方法匹配。
  6. 根据权利要求1所述的系统，其特征在于，所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析肿瘤复发转移风险，当风险值高于阈值时，该患者在当前条件下有癌症高复发转移风险。
  7. 根据权利要求6所述的系统，其特征在于，所述分析肿瘤复发转移风险包括，将具备稳定性的影像组学特征输入预测模型，得出肿瘤是否复发转移的结果；所述预测模型通过发生和未发生肿瘤复发转移的患者数据为训练集，拟合影像组学特征变化趋势与肿瘤复发转移的关系，得出肿瘤复发转移风险的阈值；预测模型以影像组学特征为输入，以肿瘤是否复发转移为输出目标，训练得到。
  8. 根据权利要求1所述的系统，其特征在于，所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析放射性损伤风险，当风险值高于阈值时，该患者在当前条件下放射性损伤风险较高。
  9. 根据权利要求8所述的系统，其特征在于，所述分析放射性损伤风险包括，将具备稳定性的影像组学特征输入预测模型，得出是否发生放射性损伤风险的结果；所述预测模型通过利用发生和未发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集，拟合影像组学特征变化趋势与放射性损伤风险的关系，得出放射性损伤风险的阈值；预测模型以影像组学特征为输入，以是否发生放射性损伤风险为输出目标，训练得到。
  10. 一种通过影像组学特征判断癌症治疗反应的方法，其特征在于，包括：将患者定位图像和疗程中的时序图像进行匹配，将定位图像上的数据映射到疗程中的时序图像上，将时序图像上选定的部位分割成相应多个感兴趣区；提取患者影像中具备稳定性的影像组学特征；随治疗进度，根据所述具备稳定性的影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应。
  11. 根据权利要求10所述的方法，其特征在于，所述具备稳定性的影像组学特征，包括具有时间稳定性的影像组学特征；所述具有时间稳定性的影像组学特征指，在不同时间，基于相同条件相同位置采集的图像中的影像组学特征具备一致性的，为具备时间稳定性的影像组学特征。
  12. 根据权利要求11所述的方法，其特征在于，所述具备稳定性的影像组学特征，还包括具有跨模态等价性的影像组学特征；所述具有跨模态等价性的影像组学特征指，同一对象相同位置不同图像模态的同一影像组学特征具备一致性的，为具有跨模态等价性的影像组学特征。
  13. 根据权利要求10所述的方法，其特征在于，具备稳定性的影像组学特征具有变化趋势与逐渐累积的辐射剂量相关联的特点。
  14. 根据权利要求10所述的方法，其特征在于，将患者定位图像和疗程中的时序图像进行匹配，包括形变配准方法匹配、或手工标注方法匹配、或刚性配准方法匹配。
  15. 根据权利要求10所述的方法，其特征在于，随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析肿瘤复发转移风险，当风险值高于阈值时，该患者在当前条件下有癌症高复发转移风险。
  16. 根据权利要求15所述的方法，其特征在于，所述分析肿瘤复发转移风险包括，将具备稳定性的影像组学特征输入预测模型，得出肿瘤是否复发转移的结果；所述预测模型通过发生和未发生肿瘤复发转移的患者数据为训练集，拟合影像组学特征变化趋势与肿瘤复发转移的关系，得出肿瘤复发转移风险的阈值；预测模型以影像组学特征为输入，以肿瘤是否复发转移为输出目标，训练得到。
  17. 根据权利要求10所述的方法，其特征在于，所述随治疗进度，根据影像组学特征参数在感兴趣区的变化趋势评估患者的治疗反应，包括分析放射性损 伤风险，当风险值高于阈值时，该患者在当前条件下放射性损伤风险较高。
  18. 根据权利要求17所述的方法，其特征在于，所述分析放射性损伤风险包括，将具备稳定性的影像组学特征输入预测模型，得出是否发生放射性损伤风险的结果；所述预测模型通过利用发生和未发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集，拟合影像组学特征变化趋势与放射性损伤风险的关系，得出放射性损伤风险的阈值；预测模型以影像组学特征为输入，以是否发生放射性损伤风险为输出目标，训练得到。
  19. 一种通过影像组学特征构建肿瘤复发转移风险预测模型的方法，其特征在于，收集发生和未发生肿瘤复发转移的患者数据为训练集；利用深度学习模型拟合影像组学特征变化趋势与肿瘤复发转移的关系，训练得出以影像组学特征为输入，以肿瘤是否复发转移为输出目标的预测模型。
  20. 一种通过影像组学特征构建放射性损伤风险预测模型的方法，其特征在于，收集发生和不发生放射损伤患者的不同感兴趣区影像组学特征时序数据为训练集；利用深度学习模型拟合影像组学特征变化趋势与放射性损伤风险的关系，训练得出以影像组学特征为输入，以是否发生放射性损伤风险为输出目标的预测模型。
PCT/CN2020/102206 2020-07-15 2020-07-15 一种通过影像组学特征判断癌症治疗反应的方法及系统 WO2022011616A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2020/102206 WO2022011616A1 (zh) 2020-07-15 2020-07-15 一种通过影像组学特征判断癌症治疗反应的方法及系统
CN202080001312.8A CN112262440A (zh) 2020-07-15 2020-07-15 一种通过影像组学特征判断癌症治疗反应的方法及系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/102206 WO2022011616A1 (zh) 2020-07-15 2020-07-15 一种通过影像组学特征判断癌症治疗反应的方法及系统

Publications (1)

Publication Number Publication Date
WO2022011616A1 true WO2022011616A1 (zh) 2022-01-20

Family

ID=74225415

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/102206 WO2022011616A1 (zh) 2020-07-15 2020-07-15 一种通过影像组学特征判断癌症治疗反应的方法及系统

Country Status (2)

Country Link
CN (1) CN112262440A (zh)
WO (1) WO2022011616A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115440383A (zh) * 2022-09-30 2022-12-06 中国医学科学院北京协和医院 用于预测晚期癌症患者pd-1/pd-l1单抗治疗疗效的系统
CN115458158A (zh) * 2022-09-23 2022-12-09 深圳大学 一种针对脓毒症患者的急性肾损伤预测系统
CN117934480A (zh) * 2024-03-25 2024-04-26 天津市第五中心医院 一种用于肺损伤预测的肺部ct图像处理方法及系统

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113257413B (zh) * 2021-06-22 2021-10-19 安翰科技(武汉)股份有限公司 基于深度学习的癌症预后生存预测方法、设备及存储介质
TWI783907B (zh) * 2022-05-24 2022-11-11 華碩電腦股份有限公司 輔助診斷系統及其方法
CN116687353B (zh) * 2023-08-01 2023-12-19 宁波杜比医疗科技有限公司 新辅助化疗疗效评估系统、设备及介质
CN117476246B (zh) * 2023-12-25 2024-04-19 福建大数据一级开发有限公司 基于多类型复发事件的患者生存分析方法、介质及装置
CN117877721A (zh) * 2024-03-12 2024-04-12 天津市肿瘤医院（天津医科大学肿瘤医院） 一种个性化甲状腺癌放疗后内分泌治疗管理系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102525534A (zh) * 2010-12-22 2012-07-04 株式会社东芝 医用图像处理装置、医用图像处理方法
CN108766563A (zh) * 2018-05-25 2018-11-06 戴建荣 基于剂量组学的放射治疗结果预测方法和系统
WO2018225063A1 (en) * 2017-06-04 2018-12-13 Rappaport Family Institute For Research In The Medical Sciences Method of predicting personalized response to cancer treatment with immune checkpoint inhibitors and kits therefor
CN109583447A (zh) * 2018-12-11 2019-04-05 山东省肿瘤防治研究院（山东省肿瘤医院） 一种影像组学分析模型构建方法及分析方法
CN109829488A (zh) * 2019-01-17 2019-05-31 桂林电子科技大学 一种头颈部癌症局部复发预测装置
WO2019152412A1 (en) * 2018-01-31 2019-08-08 General Electric Company System and method for detecting anatomical regions

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7418123B2 (en) * 2002-07-12 2008-08-26 University Of Chicago Automated method and system for computerized image analysis for prognosis
US10282588B2 (en) * 2016-06-09 2019-05-07 Siemens Healthcare Gmbh Image-based tumor phenotyping with machine learning from synthetic data

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102525534A (zh) * 2010-12-22 2012-07-04 株式会社东芝 医用图像处理装置、医用图像处理方法
WO2018225063A1 (en) * 2017-06-04 2018-12-13 Rappaport Family Institute For Research In The Medical Sciences Method of predicting personalized response to cancer treatment with immune checkpoint inhibitors and kits therefor
WO2019152412A1 (en) * 2018-01-31 2019-08-08 General Electric Company System and method for detecting anatomical regions
CN108766563A (zh) * 2018-05-25 2018-11-06 戴建荣 基于剂量组学的放射治疗结果预测方法和系统
CN109583447A (zh) * 2018-12-11 2019-04-05 山东省肿瘤防治研究院（山东省肿瘤医院） 一种影像组学分析模型构建方法及分析方法
CN109829488A (zh) * 2019-01-17 2019-05-31 桂林电子科技大学 一种头颈部癌症局部复发预测装置

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115458158A (zh) * 2022-09-23 2022-12-09 深圳大学 一种针对脓毒症患者的急性肾损伤预测系统
CN115458158B (zh) * 2022-09-23 2023-09-15 深圳大学 一种针对脓毒症患者的急性肾损伤预测系统
CN115440383A (zh) * 2022-09-30 2022-12-06 中国医学科学院北京协和医院 用于预测晚期癌症患者pd-1/pd-l1单抗治疗疗效的系统
CN117934480A (zh) * 2024-03-25 2024-04-26 天津市第五中心医院 一种用于肺损伤预测的肺部ct图像处理方法及系统
CN117934480B (zh) * 2024-03-25 2024-05-28 天津市第五中心医院 一种用于肺损伤预测的肺部ct图像处理方法及系统

Also Published As

Publication number Publication date
CN112262440A (zh) 2021-01-22

Similar Documents

Publication Publication Date Title
WO2022011616A1 (zh) 一种通过影像组学特征判断癌症治疗反应的方法及系统
JP7039153B2 (ja) 敵対的生成ネットワークを使用した画像強調
Mall et al. A comprehensive review of deep neural networks for medical image processing: Recent developments and future opportunities
Kakar et al. Automatic segmentation and recognition of lungs and lesion from CT scans of thorax
CN112951406A (zh) 一种基于ct影像组学的肺癌预后辅助评估方法及系统
El-Baz et al. Appearance analysis for diagnosing malignant lung nodules
CN108778416A (zh) 使用组织参数估计根据mr数据的伪ct生成
CN108770373A (zh) 使用特征回归模型根据mr数据的伪ct生成
CN113610845B (zh) 肿瘤局部控制预测模型的构建方法、预测方法及电子设备
Peng et al. Segmentation of lung in chest radiographs using hull and closed polygonal line method
JP2010515557A (ja) 画像処理システム、及び方法。
CN113208640B (zh) 一种基于乳腺专用pet影像组学预测腋窝淋巴结转移的方法
WO2021209887A1 (en) Rapid, accurate and machine-agnostic segmentation and quantification method and device for coronavirus ct-based diagnosis
Li et al. PSENet: Psoriasis severity evaluation network
Zhao et al. Cascade and fusion of multitask convolutional neural networks for detection of thyroid nodules in contrast-enhanced CT
Camps et al. Automatic quality assessment of transperineal ultrasound images of the male pelvic region, using deep learning
Gunasekaran Leveraging object detection for the identification of lung cancer
Shan et al. Lung cancer diagnosis based on an ann optimized by improved teo algorithm
Irene et al. Segmentation and approximation of blood volume in intracranial hemorrhage patients based on computed tomography scan images using deep learning method
Zhang et al. Detection and quantification of intracerebral and intraventricular hemorrhage from computed tomography images with adaptive thresholding and case-based reasoning
Wang et al. Deep transfer learning-based multi-modal digital twins for enhancement and diagnostic analysis of brain mri image
Xu et al. Improved cascade R-CNN for medical images of pulmonary nodules detection combining dilated HRNet
Landgren et al. An automated system for the detection and diagnosis of kidney lesions in children from scintigraphy images
CN115100155A (zh) 放射性肺炎预测模型的建立方法与系统
Diao et al. Lung Cancer Detection with 3D Ensemble Convolution Neural Network

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20945086

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20945086

Country of ref document: EP

Kind code of ref document: A1