Method and system for judging cancer treatment response based on radiomics features

Abstract

The present invention relates to the technical field of healthcare information processing, and specifically discloses a method and system for judging cancer treatment response based on radiomics features. The method includes: collecting a large number of diagnosis and treatment cases of cancer patients with a single and same type of cancer condition as a diagnosis and treatment case set for the corresponding type of cancer condition, and extracting a large number of lesion area image sequences covering the diagnosis and treatment process from the diagnosis and treatment case set. Each lesion area image in the lesion area image sequence is labeled with treatment response judgment information given by a professional doctor; using a large number of lesion area image sequences to build a treatment response judgment model for the corresponding type of cancer condition; obtaining a cancer treatment response judgment result of the current patient based on the treatment response judgment model and the lesion area image sequence of the current patient's currently traversed partial diagnosis and treatment process; thereby improving the accuracy of cancer treatment response judgment and helping doctors evaluate the treatment response of patients.

Description

Technical Field

[0001] The present invention relates to the technical field of healthcare information processing, and particularly to a method and system for judging cancer treatment response based on radiomics features. Background Art

[0002] At present, the judgment of cancer treatment response is an important part in the clinical treatment process. Traditional judgment methods mainly rely on doctors' observation of tumor size and morphology, which has certain subjectivity and errors. With the development of medical imaging technology, radiomics has become an emerging technical means. By deeply analyzing and mining medical images, a large amount of tumor-related feature information can be extracted, providing a more accurate and reliable basis for the judgment of treatment response.

[0003] However, there are still some problems in the existing radiomics technology for judging cancer treatment response, such as incomplete feature extraction and inconsistent judgment criteria. Therefore, it is of great significance to develop a method and system for judging cancer treatment response based on radiomics features.

[0004] Therefore, the present invention proposes a method and system for judging cancer treatment response based on radiomics features. Summary of the Invention

[0005] The present invention provides a method and system for judging cancer treatment response based on radiomics features, which combines the lesion area images in cancer treatment with deep learning algorithms to build a treatment response judgment model for a single type of cancer condition, avoiding problems such as incomplete feature extraction and inconsistent judgment criteria, improving the accuracy of cancer treatment response judgment, and helping doctors evaluate patients' treatment responses.

[0006] The present invention provides a method for judging cancer treatment response based on radiomics features, including:

[0007] S1: Collect a large number of medical treatment cases of cancer patients with a single same type of cancer condition as the medical treatment case set for the corresponding type of cancer condition, and extract a large number of lesion area image sequences covering the treatment process from the medical treatment case set. Each lesion area image in the lesion area image sequence is labeled with treatment response judgment information given by professional physicians;

[0008] S2: Use a large number of lesion area image sequences to build a treatment response judgment model for the corresponding type of cancer condition;

[0009] S3: Based on the treatment response judgment model and the lesion area image sequence of the current patient's current traversed partial treatment process, obtain the cancer treatment response judgment result of the current patient.

[0010] Preferably, for the method for judging cancer treatment response based on radiomics features, S1: Collect the diagnosis and treatment cases of a large number of cancer patients with a single and same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition, and extract the image sequences of the lesion areas covering the diagnosis and treatment process from the diagnosis and treatment case set, including:

[0011] S101: Collect the diagnosis and treatment cases of a large number of cancer patients with a single and same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition;

[0012] S102: In each diagnosis and treatment case in the diagnosis and treatment case set, extract the first image sequence of the lesion area obtained during the diagnosis process and the second image sequence of the lesion area obtained during the treatment process;

[0013] S103: Annotate the treatment response judgment information given by a professional doctor for each lesion area image in the first image sequence of the lesion area and the second image sequence of the lesion area to each lesion area image in the first image sequence of the lesion area and the second image sequence of the lesion area;

[0014] S104: Splice the first image sequence of the lesion area and the second image sequence of the lesion area in each diagnosis and treatment case to obtain a single image sequence of the lesion area for the corresponding type of cancer condition until all the image sequences of the lesion areas in the diagnosis and treatment case set are obtained.

[0015] Preferably, for the method for judging cancer treatment response based on radiomics features, S2: Use a large number of image sequences of lesion areas to build a treatment response judgment model for the corresponding type of cancer condition, including:

[0016] Based on the treatment response judgment information annotated for each lesion area image in all the image sequences of the lesion areas, divide all the image sequences of the lesion areas into a training image set of the lesion area and a test image set of the lesion area;

[0017] Based on the training image set of the lesion area and the test image set of the lesion area, build a treatment response judgment model for the corresponding type of cancer condition.

[0018] Preferably, for the method for judging cancer treatment response based on radiomics features, based on the treatment response judgment information annotated for each lesion area image in all the image sequences of the lesion areas, divide all the image sequences of the lesion areas into a training image set of the lesion area and a test image set of the lesion area, including:

[0019] Regard the treatment response judgment information annotated for the last lesion area image in each image sequence of the lesion area as the decision treatment response judgment information for each image sequence of the lesion area;

[0020] Classifying all decision-making treatment response judgment information to obtain multiple types of decision-making treatment response judgment information;

[0021] determining whether the numbers of lesion region image sequences belonging to different decision-making treatment response judgment information types are the same; if so, dividing the lesion region image sequences belonging to each decision-making treatment response judgment information type into a sub-training lesion region image set and a sub-test lesion region image set of each decision-making treatment response judgment information type according to a preset ratio;

[0022] The sub-training lesion region image sets of all decision-making treatment response judgment information types are summarized to obtain a training lesion region image set, and at the same time, the sub-test lesion region image sets of all decision-making treatment response judgment information types are summarized to obtain a test lesion region image set;

[0023] Otherwise, the secondary decision-making treatment response judgment information of each lesion area image sequence is determined;

[0024] Based on the decision treatment response judgment information and sub-decision treatment response judgment information of all lesion area image sequences, all lesion area image sequences are divided into a training lesion area image set and a testing lesion area image set.

[0025] Preferably, the cancer treatment response judgment method based on imaging omics features classifies all decision-making treatment response judgment information to obtain multiple types of decision-making treatment response judgment information, including:

[0026] Perform synonymous association on all decision-making treatment response judgment information to obtain synonymous association results;

[0027] Based on the synonymous association results, multiple types of decision-making treatment response judgment information are obtained.

[0028] Preferably, the cancer treatment response judgment method based on imaging omics features determines the secondary decision treatment response judgment information of each lesion area image sequence, including:

[0029] In a reverse order search manner, the treatment response judgment information that appears for the first time and is different from the corresponding decision treatment response judgment information is retrieved in each lesion area image sequence as the secondary decision treatment response judgment information of the corresponding lesion area image sequence.

[0030] Preferably, the method for judging cancer treatment response based on imaging omics features, based on the decision treatment response judgment information and the sub-decision treatment response judgment information of all lesion area image sequences, divides all lesion area image sequences into a training lesion area image set and a test lesion area image set, including:

[0031] Among all the decision-making treatment response judgment information of the lesion area image sequences, the decision-making treatment response judgment information type containing the largest number of lesion area image sequences is determined as the reserved information type, and the decision-making treatment response judgment information types other than the reserved information types are regarded as non-reserved information types;

[0032] Based on the number of all decision-making treatment response judgment information of the retained information type and the number of all decision-making treatment response judgment information of each non-reserved information type in the decision-making treatment response judgment information of all lesion area image sequences, the amount of supplementary information of each non-reserved information type is determined;

[0033] Based on the amount of supplementary information, among all lesion region image sequences whose sub-decision treatment response judgment information is of the corresponding non-reserved information type, screen out the lesion region image sequences to be supplemented of the corresponding non-reserved information type, and based on the sub-decision treatment response judgment information of each lesion region image sequence to be supplemented, cut out the supplementary lesion region image sequences of the corresponding non-reserved information type from the corresponding lesion region image sequences to be supplemented;

[0034] The lesion region image sequences of non-reserved categories and the supplementary lesion region image sequences as well as all lesion region image sequences of retained categories are divided to obtain a training lesion region image set and a test lesion region image set.

[0035] Preferably, the method for judging the response to cancer treatment based on imaging omics features divides the lesion region image sequences of non-reserved categories and the supplementary lesion region image sequences and all lesion region image sequences of reserved categories to obtain a training lesion region image set and a test lesion region image set, including:

[0036] According to a preset ratio, the lesion region image sequence and the supplementary lesion region image sequence belonging to each non-reserved category are divided into a sub-training lesion region image set and a sub-testing lesion region image set of each non-reserved category, respectively;

[0037] According to a preset ratio, all lesion region image sequences belonging to the reserved category are divided into a sub-training lesion region image set and a sub-testing lesion region image set of the reserved category;

[0038] Summarizing the sub-training lesion region image sets of the reserved categories and the sub-training lesion region image sets of all non-reserved categories to obtain a training lesion region image set;

[0039] The sub-test lesion area image sets of the reserved categories and the sub-test lesion area image sets of all non-reserved categories are aggregated to obtain the test lesion area image set.

[0040] Preferably, for the method for judging cancer treatment response based on radiomics features, S3: based on the treatment response judgment model and the image sequence of the lesion area in the current partial diagnosis and treatment process of the current patient, obtain the cancer treatment response judgment result of the current patient, including:

[0041] Input the image sequence of the lesion area in the current partial diagnosis and treatment process of the current patient into the treatment response judgment model to obtain the cancer treatment response judgment result of the current patient.

[0042] The present invention provides a system for judging cancer treatment response based on radiomics features, which is used to execute the method for judging cancer treatment response based on radiomics features described in any one of Embodiments 1 to 9, including:

[0043] An image collection module, which is used to collect the diagnosis and treatment cases of a large number of cancer patients with a single same type of cancer condition as the diagnosis and treatment case set corresponding to the cancer condition of the same type, and extract a large number of image sequences of the lesion area covering the diagnosis and treatment process from the diagnosis and treatment case set. Among them, each lesion area image in the image sequence of the lesion area is marked with the treatment response judgment information given by a professional doctor;

[0044] A model building module, which is used to build a treatment response judgment model for the cancer condition of the corresponding type by using a large number of lesion area influence sequences;

[0045] A response judgment module, which is used to obtain the cancer treatment response judgment result of the current patient based on the treatment response judgment model and the image sequence of the lesion area in the current partial diagnosis and treatment process of the current patient.

[0046] The beneficial effects of the present invention compared with the prior art are as follows: by combining the image of the lesion area in cancer treatment with the deep learning algorithm, a treatment response judgment model for a single type of cancer condition is built, avoiding problems such as incomplete feature extraction and inconsistent judgment criteria, improving the accuracy of cancer treatment response judgment, and helping doctors evaluate the treatment response of patients.

[0047] Other features and advantages of the present invention will be described in the following description, and some of them will become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the structure specifically pointed out in this application document.

[0048] The technical solution of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings

[0049] The drawings are used to provide a further understanding of the present invention, and constitute a part of the description. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0050] Figure 1 Flow chart of the method for judging cancer treatment response based on radiomics features in the embodiments of the present invention;

[0051] Figure 2 Flow chart of the specific implementation method of step S1 in the embodiments of the present invention;

[0052] Figure 3 Schematic diagram of the internal functional modules of the system for judging cancer treatment response based on radiomics features in the embodiments of the present invention. Specific embodiments

[0053] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not used to limit the present invention.

[0054] Embodiment 1:

[0055] The present invention provides a method for judging cancer treatment response based on radiomics features. Referring to Figure 1 , including:

[0056] S1: Collect a large number of diagnosis and treatment cases of cancer patients with a single and same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition, and extract a large number of lesion area image sequences covering the diagnosis and treatment process from the diagnosis and treatment case set. Among them, each lesion area image in the lesion area image sequence is labeled with the treatment response judgment information given by a professional doctor;

[0057] S2: Use a large number of lesion area influence sequences to build a treatment response judgment model for the corresponding type of cancer condition;

[0058] S3: Obtain the cancer treatment response judgment result of the current patient based on the treatment response judgment model and the lesion area image sequence of the current part of the diagnosis and treatment process of the current patient.

[0059] In this embodiment, the types of cancer conditions are, for example, the types obtained by dividing according to the early, middle, and late stages of different cancers, such as early-stage thyroid cancer, middle-stage thyroid cancer, late-stage thyroid cancer, early-stage liver cancer, middle-stage liver cancer, late-stage liver cancer, etc.

[0060] In this embodiment, the diagnosis and treatment case includes the lesion area image sequence and the doctor's diagnosis result obtained by the patient in the diagnosis stage, and the lesion area image sequence and the doctor's diagnosis and treatment advice obtained by the patient in the treatment stage.

[0061] In this embodiment, covering the diagnosis and treatment process includes the patient's diagnosis process and treatment process.

[0062] In this embodiment, the image sequence of the lesion area is a sequence obtained by sorting the images of the lesion area acquired during the diagnosis process (the time period from the patient's visit to the issuance of the final diagnosis result) and the treatment process (the time period from the start of the patient's treatment to the end of the treatment course) of the patient in chronological order.

[0063] In this embodiment, the image of the lesion area is an image obtained by using CT, MRI, PET, etc. and containing the part of the patient where cancer has occurred.

[0064] In this embodiment, the treatment response judgment information is the diagnosis information of the cancer condition or the diagnosis information of the treatment response given by a professional doctor for the image of the lesion area. For example, the location and speed of cancer cell spread, or conclusions such as whether the condition has completely remitted, partially remitted, or is stable.

[0065] In this embodiment, the treatment response judgment model is based on a machine learning algorithm and uses a large number of image sequences of the lesion area as training samples and test samples to build a model that can give the current cancer treatment response judgment result of the patient for the image sequence of the lesion area in the current traversed part of the diagnosis and treatment process of the current patient input into the model. It can try to use classic AlexNet, GoogLeNet, or deeper and more complex network structures such as ResNet, Inception, VGG, etc.

[0066] In this embodiment, the cancer treatment response judgment result is the conclusion information about the current cancer treatment response of the patient given by the treatment response judgment model based on the image sequence of the lesion area in the current traversed part of the diagnosis and treatment process of the current patient input into the model. For example, the location and speed of cancer cell spread, or conclusions such as whether the condition has completely remitted, partially remitted, or is stable.

[0067] The beneficial effects of the above technology are as follows: By combining the image of the lesion area in cancer treatment with a deep learning algorithm, a treatment response judgment model for a single type of cancer condition is built, avoiding problems such as incomplete feature extraction and inconsistent judgment criteria, improving the accuracy of cancer treatment response judgment, and helping doctors evaluate the treatment response of patients.

[0068] Embodiment 2:

[0069] On the basis of Embodiment 1, for the cancer treatment response judgment method based on radiomics features, S1: Collect a large number of diagnosis and treatment cases of cancer patients with a single and same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition, and extract the image sequence of the lesion area covering the diagnosis and treatment process from the diagnosis and treatment case set, referring to Figure 2 , including:

[0070] S101: Collect the diagnosis and treatment cases of a large number of cancer patients with a single type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition.

[0071] S102: In each diagnosis and treatment case in the diagnosis and treatment case set, extract the first lesion area image sequence obtained during the diagnosis process and the second lesion area image sequence obtained during the treatment process.

[0072] S103: Annotate the treatment response judgment information given by a professional physician for each lesion area image in the first lesion area image sequence and the second lesion area image sequence to each lesion area image in the first lesion area image sequence and the second lesion area image sequence.

[0073] S104: Concatenate the first lesion area image sequence and the second lesion area image sequence in each diagnosis and treatment case to obtain a single lesion area image sequence for the corresponding type of cancer condition until all the lesion area image sequences in the diagnosis and treatment case set are obtained.

[0074] In this embodiment, the first lesion area image sequence is a sequence obtained by sorting all the lesion area images obtained during the diagnosis process of a single case object (patient) in chronological order.

[0075] In this embodiment, the second lesion area image sequence is a sequence obtained by sorting all the lesion area images obtained during the treatment process of a single case object (patient) in chronological order.

[0076] The beneficial effects of the above technology are as follows: The first lesion area image sequence belonging to the diagnosis process and the second lesion area image sequence belonging to the treatment process in each diagnosis and treatment case are obtained, and through further concatenation and annotation of the treatment response judgment information, a lesion area image sequence covering the diagnosis and treatment process is obtained.

[0077] Embodiment 3:

[0078] Based on the cancer treatment response judgment method based on radiomics features in Embodiment 1, S2: Use a large number of lesion area influence sequences to build a treatment response judgment model for the corresponding type of cancer condition, including:

[0079] Based on the treatment response judgment information annotated for each lesion area image in all the lesion area image sequences, divide all the lesion area image sequences into a training lesion area image set and a test lesion area image set.

[0080] Based on the training lesion area image set and the test lesion area image set, build a treatment response judgment model for the corresponding type of cancer condition.

[0081] In this embodiment, the (partial) lesion area image sequences in the training lesion area image set that are used as training samples during the model training process for building the treatment response judgment model.

[0082] In this embodiment, the (partial) lesion area image sequences in the test lesion area image set that are used as training samples during the model testing process for building the treatment response judgment model.

[0083] In this embodiment, based on the training lesion area image set and the test lesion area image set, a treatment response judgment model for the corresponding type of cancer condition is built, including:

[0084] Regarding each (partial) lesion area image sequence in the training lesion area image set as the model input quantity, and regarding the treatment response judgment information labeled by the last lesion area image in its (partial) lesion area image sequence as the corresponding model output quantity, perform model training to obtain a trained model;

[0085] Use each (partial) lesion area image sequence in the test lesion area image set as the model input quantity of the trained model to obtain the model test output information of each (partial) lesion area image sequence;

[0086] Calculate the test result compliance between the model test output information of each (partial) lesion area image sequence and the treatment response judgment information labeled by the last lesion area image in its corresponding (partial) lesion area image sequence. This step is implemented using a preset compliance determination model. Input the model test output information of each (partial) lesion area image sequence and the treatment response judgment information labeled by the last lesion area image in its corresponding (partial) lesion area image sequence into the preset compliance determination model to obtain the corresponding test result compliance;

[0087] Among them, the preset compliance determination model is a model trained using a large number of two treatment response judgment information whose compliance with each other is manually marked as training samples;

[0088] And regard the average value of the test result compliances of all (partial) lesion area image sequences in the test lesion area image set as the evaluation value of the trained model;

[0089] When the evaluation value is not less than the preset evaluation threshold, then regard the trained model as the treatment response judgment model for the corresponding type of cancer condition;

[0090] Otherwise, based on the steps in the embodiment, a large number of new lesion area impact sequences are required to be re-acquired to retrain a new model, until the evaluation value obtained by the retrained model in the test phase is not less than the preset evaluation threshold, and the retrained model is used as a treatment response judgment model for the corresponding type of cancer.

[0091] The beneficial effect of the above technology is: using a large number of lesion area impact sequences to divide the training lesion area image set and the test lesion area image set to complete the model training and testing process, and then obtain a treatment response judgment model for a single type of cancer with guaranteed judgment accuracy.

[0092] Embodiment 4:

[0093] On the basis of Example 3, a method for judging cancer treatment response based on imaging omics features, based on the treatment response judgment information annotated on each lesion area image in all lesion area image sequences, divides all lesion area image sequences into a training lesion area image set and a test lesion area image set, including:

[0094] The treatment response judgment information of the last lesion region image annotation in each lesion region image sequence is regarded as the decision-making treatment response judgment information of each lesion region image sequence;

[0095] Classifying all decision-making treatment response judgment information to obtain multiple types of decision-making treatment response judgment information;

[0096] determining whether the numbers of lesion region image sequences belonging to different decision-making treatment response judgment information types are the same; if so, dividing the lesion region image sequences belonging to each decision-making treatment response judgment information type into a sub-training lesion region image set and a sub-test lesion region image set of each decision-making treatment response judgment information type according to a preset ratio;

[0097] The sub-training lesion region image sets of all decision-making treatment response judgment information types are summarized to obtain a training lesion region image set, and at the same time, the sub-test lesion region image sets of all decision-making treatment response judgment information types are summarized to obtain a test lesion region image set;

[0098] Otherwise, the secondary decision-making treatment response judgment information of each lesion area image sequence is determined;

[0099] Based on the decision treatment response judgment information and sub-decision treatment response judgment information of all lesion area image sequences, all lesion area image sequences are divided into a training lesion area image set and a testing lesion area image set.

[0100] In this embodiment, the types of decision-making treatment response judgment information are distinguished by the meanings of the decision-making treatment response judgment information. For example, if the meanings of two or more decision-making treatment response judgment information are the same, it means that these two or more decision-making treatment response judgment information belong to the same type of decision-making quality response judgment information.

[0101] In this embodiment, the preset ratio is, for example, the ratio of the number of lesion area image sequences in the training set and the test set is 7:3.

[0102] In this embodiment, the sub-training lesion area image set is a set composed of (partial) lesion area image region sequences that are part of the training lesion area image set.

[0103] In this embodiment, the sub-test lesion area image set is a set composed of (partial) lesion area image region sequences that are part of the test lesion area image set.

[0104] In this embodiment, the secondary decision-making treatment response judgment information is to retrieve, in a reverse search manner, the treatment response judgment information that first appears different from the corresponding decision-making treatment response judgment information type in each lesion area image sequence.

[0105] The beneficial effects of the above technologies are as follows: It realizes the determination of the decision-making treatment response judgment information for each lesion area image sequence, and by judging whether the number of lesion area image sequences belonging to different types of decision-making treatment response judgment information is the same, two methods of dividing the training lesion area image set and the test lesion area image set are given respectively, ensuring the generalization performance of the training samples and test samples used to build the treatment response judgment model.

[0106] Embodiment 5:

[0107] Based on the method for judging cancer treatment response based on radiomics features in Embodiment 4, all decision-making treatment response judgment information is classified to obtain multiple types of decision-making treatment response judgment information, including:

[0108] Perform synonymous association on all decision-making treatment response judgment information to obtain a synonymous association result;

[0109] Based on the synonymous association result, obtain multiple types of decision-making treatment response judgment information.

[0110] In this embodiment, performing synonymous association on all decision-making treatment response judgment information to obtain a synonymous association result includes:

[0111] Use an existing word segmentation model to segment all decision-making treatment response judgment information to obtain the word sequence of each decision-making treatment response judgment information;

[0112] Based on the preset synonym word bag of each word in the word sequence (the preset synonym word bag is collected in advance), determine the word group of words in the preset synonym word bag of each other in the two decision treatment response judgment information;

[0113] The quotient of the total number of word groups in the preset synonym word bags that belong to each other in the two decision-making treatment response judgment information and half of the total number of words in the word sequences of the two decision-making treatment response judgment information is regarded as the synonymy degree between the two decision-making treatment response judgment information;

[0114] All two decision-making treatment response judgment information whose synonymy degrees exceed a preset degree threshold are associated to obtain a synonymous association result.

[0115] In this embodiment, based on the synonymous association results, multiple types of decision-making treatment response judgment information are obtained, which are:

[0116] All pairwise decision-making treatment response judgment information with synonymous associations are summarized, and summarized using a preset summary model to obtain corresponding decision-making treatment response judgment information types;

[0117] In this way, multiple types of information for decision-making treatment response judgments are obtained;

[0118] The preset summarization model is a model that is obtained by training with information sets marked with semantic summary information as training samples and can perform semantic summarization on multiple input information.

[0119] The beneficial effect of the above technology is: through the synonymous association of all decision-making treatment response judgment information, it is possible to classify it according to semantic categories.

[0120] Embodiment 6:

[0121] On the basis of Example 4, a method for judging cancer treatment response based on imaging omics features is used to determine the secondary decision-making treatment response judgment information of each lesion area image sequence, including:

[0122] In a reverse order search manner, the treatment response judgment information that appears for the first time and is different from the corresponding decision treatment response judgment information is retrieved in each lesion area image sequence as the secondary decision treatment response judgment information of the corresponding lesion area image sequence.

[0123] In this embodiment, the reverse order retrieval method is to start from the last lesion area image in the lesion area image sequence, and then search the second to last lesion area image, the third to last lesion area image, and so on in this order.

[0124] The beneficial effects of the above technology are as follows: The determination method and practical significance of the sub - decision treatment response judgment information for each lesion area image sequence are given.

[0125] Example 7:

[0126] Based on the method for judging cancer treatment response based on radiomics features in Example 4, and based on the decision - making treatment response judgment information and sub - decision - making treatment response judgment information of all lesion area image sequences, all lesion area image sequences are divided into a training lesion area image set and a test lesion area image set, including:

[0127] Among the decision - making treatment response judgment information of all lesion area image sequences, determine the type of decision - making treatment response judgment information with the largest number of included lesion area image sequences as the retained information type, and regard the types of decision - making treatment response judgment information other than the retained information type as non - retained information types;

[0128] Based on the number of all decision - making treatment response judgment information of the retained information type and the number of all decision - making treatment response judgment information of each non - retained information type among the decision - making treatment response judgment information of all lesion area image sequences, determine the supplementary information amount for each non - retained information type;

[0129] Based on the supplementary information amount, among the lesion area image sequences where all sub - decision - making treatment response judgment information is of the corresponding non - retained information type, screen out the lesion area image sequences to be supplemented of the corresponding non - retained information type, and based on the sub - decision - making treatment response judgment information of each lesion area image sequence to be supplemented, intercept the supplementary lesion area image sequences of the corresponding non - retained information type in the corresponding lesion area image sequences to be supplemented;

[0130] Divide the lesion area image sequences of non - retained types, the supplementary lesion area image sequences, and all lesion area image sequences of the retained type to obtain a training lesion area image set and a test lesion area image set.

[0131] In this example, the supplementary information amount is the number of additional items required for the sub - decision - making treatment response judgment information of the corresponding non - retained information type.

[0132] In this example, based on the number of all decision - making treatment response judgment information of the retained information type and the number of all decision - making treatment response judgment information of each non - retained information type among the decision - making treatment response judgment information of all lesion area image sequences, the supplementary information amount for each non - retained information type is determined as:

[0133] In all the decision-making treatment response judgment information of the image sequences of the lesion regions, the difference between the quantity of all the decision-making treatment response judgment information of the retained information types and the quantity of all the decision-making treatment response judgment information of each non-retained information type is regarded as the supplementary information quantity corresponding to the non-retained information type.

[0134] In this embodiment, based on the supplementary information quantity, among the image sequences of the lesion regions where all the secondary decision-making treatment response judgment information is of the corresponding non-retained information type, the image sequences of the lesion regions to be supplemented for the corresponding non-retained information type are screened out as follows:

[0135] Assume that the supplementary information quantity corresponding to the non-retained information type is Y. Then, among the image sequences of the lesion regions where all the secondary decision-making treatment response judgment information is of the corresponding non-retained information type, Y image sequences of the lesion regions are screened out as the image sequences of the lesion regions to be supplemented for the corresponding non-retained information type;

[0136] Among them, the image sequence of the lesion region to be supplemented is the image sequence of the lesion region from which a part needs to be intercepted as the supplementary image sequence of the lesion region for the non-retained information type.

[0137] In this embodiment, based on the secondary decision-making treatment response judgment information of each image sequence of the lesion region to be supplemented, the supplementary image sequence of the lesion region for the corresponding non-retained information type is intercepted from the corresponding image sequence of the lesion region to be supplemented as follows:

[0138] The partial image sequence of the lesion region from the first image sequence of the lesion region to the last image sequence of the lesion region where the corresponding secondary decision-making treatment response judgment information is marked in the image sequence of the lesion region to be supplemented is used as the supplementary image sequence of the lesion region for the corresponding non-retained information type.

[0139] Among them, the supplementary image sequence of the lesion region is the partial image sequence of the lesion region used as a training sample or a test sample for the corresponding non-retained information type.

[0140] The beneficial effects of the above technology are as follows: It realizes the supplementation of the image sequences of the lesion regions with insufficient sample quantity for the decision-making treatment response judgment information types. By introducing the secondary decision-making treatment response judgment information, it can ensure to the greatest extent that the difference between the treatment response judgment information corresponding to the partial image sequence of the lesion region intercepted from the complete image sequence of the lesion region as the training sample or the test sample and its original final treatment response judgment information is small enough. Furthermore, in this way, it can also ensure a relatively high accuracy of the treatment response judgment model built using the image sequence samples screened in this way under the condition of limited sample quantity, can reduce the sample quantity requirement to a certain extent during the model building process, and can also reduce the number of times of model retraining.

[0141] Embodiment 8:

[0142] Based on Example 7, for the method of judging cancer treatment response based on radiomics features, the image sequences of lesion regions of non-retained types, the supplementary lesion region image sequences, and all lesion region image sequences of retained types are divided to obtain a training lesion region image set and a test lesion region image set, including:

[0143] According to a preset ratio (such as 7:3), the image sequences of lesion regions of each non-retained type and the supplementary lesion region image sequences are respectively divided into a sub-training lesion region image set and a sub-test lesion region image set of each non-retained type (assuming the total number of image sequences of lesion regions of a single non-retained type and the supplementary lesion region image sequences is N1, then the total number of image sequences of lesion regions in the sub-training lesion region image set corresponding to the non-retained type is 7 / 10N1, and the total number of image sequences of lesion regions in the sub-test lesion region image set corresponding to the non-retained type is 3 / 10N1);

[0144] According to a preset ratio (such as 7:3), all the image sequences of lesion regions of the retained type are divided into a sub-training lesion region image set and a sub-test lesion region image set of the retained type (assuming the total number of all image sequences of lesion regions of a single retained type is N2, then the total number of image sequences of lesion regions in the sub-training lesion region image set corresponding to the single retained type is 7 / 10N2, and the total number of image sequences of lesion regions in the sub-test lesion region image set corresponding to the single retained type is 3 / 10N2);

[0145] The sub-training lesion region image set of the retained type and the sub-training lesion region image sets of all non-retained types are aggregated to obtain a training lesion region image set;

[0146] The sub-test lesion region image set of the retained type and the sub-test lesion region image sets of all non-retained types are aggregated to obtain a test lesion region image set.

[0147] The partitioning process in this example is a random partitioning following a preset ratio.

[0148] The beneficial effects of the above technology are as follows: An implementation method for dividing the image sequences of lesion regions of non-retained types, the supplementary lesion region image sequences, and all lesion region image sequences of retained types to obtain a training lesion region image set and a test lesion region image set is given.

[0149] Example 9:

[0150] Based on Example 1, for the method of judging cancer treatment response based on radiomics features, S3: Based on the treatment response judgment model and the image sequences of lesion regions in the current part of the diagnosis and treatment process of the current patient, obtain the cancer treatment response judgment result of the current patient, including:

[0151] Input the image sequence of the lesion area in the current part of the diagnosis and treatment process of the current patient into the treatment response judgment model to obtain the cancer treatment response judgment result of the current patient.

[0152] In this embodiment, the current part of the traversed diagnosis and treatment process includes the diagnosis process that the patient has currently experienced and the treatment process up to the current moment.

[0153] The beneficial effect of the above technology is: It gives a specific implementation method of how to determine the cancer treatment response judgment result of the current patient by using the treatment response judgment model and the image sequence of the lesion area in the current part of the diagnosis and treatment process of the current patient.

[0154] Embodiment 10:

[0155] The present invention provides a cancer treatment response judgment system based on radiomics features for performing the cancer treatment response judgment method based on radiomics features described in any one of Embodiments 1 to 9, referring to Figure 3 , including:

[0156] An image collection module, which is used to collect a large number of diagnosis and treatment cases of cancer patients with a single and same type of cancer condition as a diagnosis and treatment case set for the corresponding type of cancer condition, and extract a large number of image sequences of lesion areas covering the diagnosis and treatment process from the diagnosis and treatment case set. Among them, each lesion area image in the image sequence of the lesion area is labeled with the treatment response judgment information given by a professional doctor;

[0157] A model building module, which is used to build a treatment response judgment model for the corresponding type of cancer condition by using a large number of image sequences of lesion areas;

[0158] A response judgment module, which is used to obtain the cancer treatment response judgment result of the current patient based on the treatment response judgment model and the image sequence of the lesion area in the current part of the diagnosis and treatment process of the current patient.

[0159] The beneficial effect of the above technology is: Combining the image of the lesion area in cancer treatment with the deep learning algorithm to build a treatment response judgment model for a single type of cancer condition, avoiding problems such as incomplete feature extraction and inconsistent judgment criteria, improving the accuracy of cancer treatment response judgment, and helping doctors evaluate the treatment response of patients.

[0160] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A method for judging cancer treatment response based on radiomics features, characterized in that, include: S1: Collect a large number of diagnosis and treatment cases of cancer patients with a single and identical type of cancer as a diagnosis and treatment case set of the corresponding type of cancer, and extract a large number of lesion area image sequences covering the diagnosis and treatment process from the diagnosis and treatment case set, wherein each lesion area image in the lesion area image sequence is annotated with treatment response judgment information given by professional physicians; S2: Using a large number of lesion area image sequences, a treatment response judgment model for the corresponding type of cancer is constructed; S3: Based on the treatment response judgment model and the image sequence of the lesion area of the current patient's current diagnosis and treatment process, the cancer treatment response judgment result of the current patient is obtained; Among them, step S2: using a large number of lesion area image sequences to build a treatment response judgment model for the corresponding type of cancer condition, including: Based on the treatment response judgment information annotated on each lesion region image in all lesion region image sequences, all lesion region image sequences are divided into a training lesion region image set and a testing lesion region image set; Based on the training lesion area image set and the test lesion area image set, a treatment response judgment model for the corresponding type of cancer is built; Among them, based on the treatment response judgment information annotated on each lesion area image in all lesion area image sequences, all lesion area image sequences are divided into a training lesion area image set and a test lesion area image set, including: The treatment response judgment information of the last lesion region image annotation in each lesion region image sequence is regarded as the decision-making treatment response judgment information of each lesion region image sequence; All decision-making treatment response judgment information is classified to obtain multiple types of decision-making treatment response judgment information, including: Using the existing word segmentation model to segment all decision-making treatment response judgment information, obtain the word sequence of each decision-making treatment response judgment information; Based on the preset synonym word bag of each word in the word sequence, determine the word group of words in the preset synonym word bag of each word in the two decision treatment response judgment information that belong to each other; The quotient of the total number of word groups in the preset synonym word bags that belong to each other in the two decision-making treatment response judgment information and half of the total number of words in the word sequences of the two decision-making treatment response judgment information is regarded as the synonymy degree between the two decision-making treatment response judgment information; Correlating all two decision-making treatment response judgment information whose synonymy degree exceeds a preset degree threshold to obtain a synonymous correlation result; All pairwise decision-making treatment response judgment information with synonymous association relationships in the synonymous association results are summarized, and summarized using a preset summary model to obtain corresponding decision-making treatment response judgment information types; determining whether the numbers of lesion region image sequences belonging to different decision-making treatment response judgment information types are the same; if so, dividing the lesion region image sequences belonging to each decision-making treatment response judgment information type into a sub-training lesion region image set and a sub-test lesion region image set of each decision-making treatment response judgment information type according to a preset ratio; Summarize the sub-training lesion region image sets of all decision-making treatment response judgment information types to obtain the training lesion region image set. At the same time, summarize the sub-test lesion region image sets of all decision-making treatment response judgment information types to obtain the test lesion region image set; Otherwise, determine the secondary decision-making treatment response judgment information for each lesion region image sequence; Based on the decision-making treatment response judgment information and secondary decision-making treatment response judgment information of all lesion region image sequences, divide all lesion region image sequences into a training lesion region image set and a test lesion region image set, including: Among the decision-making treatment response judgment information of all lesion region image sequences, determine the decision-making treatment response judgment information type with the largest number of included lesion region image sequences as the retained information type, and regard the decision-making treatment response judgment information types other than the retained information type as non-retained information types; Based on the number of all decision-making treatment response judgment information of the retained information type and the number of all decision-making treatment response judgment information of each non-retained information type in the decision-making treatment response judgment information of all lesion region image sequences, determine the supplementary information amount of each non-retained information type; Based on the supplementary information amount, in the lesion region image sequences where the secondary decision-making treatment response judgment information is the corresponding non-retained information type, screen out the lesion region image sequences to be supplemented of the corresponding non-retained information type, and based on the secondary decision-making treatment response judgment information of each lesion region image sequence to be supplemented, intercept the supplementary lesion region image sequences of the corresponding non-retained information type in the corresponding lesion region image sequences to be supplemented; Divide the lesion region image sequences of non-retained types, the supplementary lesion region image sequences, and all lesion region image sequences of the retained type to obtain the training lesion region image set and the test lesion region image set.

2. The method for judging cancer treatment response based on radiomics features according to claim 1, wherein S1: Collect the diagnosis and treatment cases of a large number of cancer patients with a single and same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition, and extract the lesion region image sequences covering the diagnosis and treatment process from the diagnosis and treatment case set, including: S101: Collect the diagnosis and treatment cases of a large number of cancer patients with a single and same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition; S102: In each diagnosis and treatment case in the diagnosis and treatment case set, extract the first lesion region image sequence obtained during the diagnosis process and the second lesion region image sequence obtained during the treatment process; S103: Label the treatment response judgment information given by a professional physician for each lesion region image in the first lesion region image sequence and the second lesion region image sequence to each lesion region image in the first lesion region image sequence and the second lesion region image sequence; S104: Concatenate the first lesion region image sequence and the second lesion region image sequence in each diagnosis and treatment case to obtain a single lesion region image sequence for the corresponding type of cancer condition until all lesion region image sequences in the diagnosis and treatment case set are obtained.

3. The method for judging cancer treatment response based on radiomics features according to claim 1, wherein Classify all decision-making treatment response judgment information to obtain multiple decision-making treatment response judgment information types, including: Perform synonymous association on all decision-making treatment response judgment information to obtain the synonymous association result; Based on the synonymous association result, obtain multiple types of decision-making treatment response judgment information.

4. The method for judging cancer treatment response based on radiomics features according to claim 1, wherein, Determine the secondary decision-making treatment response judgment information for each lesion area image sequence, including: In a reverse search manner, retrieve the treatment response judgment information that first appears differently from the corresponding type of decision-making treatment response judgment information in each lesion area image sequence as the secondary decision-making treatment response judgment information for the corresponding lesion area image sequence.

5. The method for judging cancer treatment response based on radiomics features according to claim 1, wherein, Divide the lesion area image sequences of non-retained types, supplementary lesion area image sequences, and all lesion area image sequences of retained types to obtain a training lesion area image set and a test lesion area image set, including: According to a preset ratio, divide the lesion area image sequences belonging to each non-retained type and supplementary lesion area image sequences into sub-training lesion area image sets and sub-test lesion area image sets for each non-retained type; According to a preset ratio, divide all the lesion area image sequences belonging to the retained type into a sub-training lesion area image set and a sub-test lesion area image set of the retained type; Summarize the sub-training lesion area image set of the retained type and all the sub-training lesion area image sequences of non-retained types to obtain a training lesion area image set; Summarize the sub-test lesion area image set of the retained type and all the sub-test lesion area image sequences of non-retained types to obtain a test lesion area image set.

6. The method for judging cancer treatment response based on radiomics features according to claim 1, wherein S3: Based on the treatment response judgment model and the lesion area image sequence of the current patient's currently traversed partial diagnosis and treatment process, obtain the cancer treatment response judgment result of the current patient, including: Input the lesion area image sequence of the current patient's currently traversed partial diagnosis and treatment process into the treatment response judgment model to obtain the cancer treatment response judgment result of the current patient.

7. A cancer treatment response judgment system based on radiomics features, for performing the cancer treatment response judgment method based on radiomics features according to any one of claims 1 to 6, characterized in that, Including: An image collection module for collecting a large number of diagnosis and treatment cases of cancer patients with a single same type of cancer condition as the diagnosis and treatment case set for the corresponding type of cancer condition, and extracting a large number of lesion area image sequences covering the diagnosis and treatment process from the diagnosis and treatment case set, where each lesion area image in the lesion area image sequence is labeled with treatment response judgment information given by a professional physician; A model construction module for using a large number of lesion area image sequences to construct a treatment response judgment model for the corresponding type of cancer condition; A response judgment module for obtaining the cancer treatment response judgment result of the current patient based on the treatment response judgment model and the lesion area image sequence of the current patient's currently traversed partial diagnosis and treatment process.

Images