Colorectal cancer prognosis evaluation method and equipment based on multi-modal data and medium

Abstract

The invention relates to a colorectal cancer prognosis evaluation method and device based on multi-modal data and a medium, and belongs to the technical field of colorectal cancer prognosis evaluation, and the method comprises the following steps: obtaining a multi-modal to-be-detected data set of a target patient; inputting the test data into a preset colorectal cancer feature detection model to obtain a plurality of first key features and a plurality of second key features, and verifying the second key features according to the image imaging data to obtain a verification result; and when the verification result shows that the first key feature is correct, evaluating the colorectal cancer prognosis of the target patient based on the first key feature to obtain an evaluation result. According to the method, a more complete and accurate colorectal cancer prognosis judgment system can be jointly constructed through mutual association and mutual complementation of multi-modal data, and the accuracy of colorectal cancer development prediction of a patient receiving a radical operation is greatly improved.

Description

Technical Field

[0001] The present application relates to the technical field of colorectal cancer prognosis, and in particular to a method, device and medium for evaluating colorectal cancer prognosis based on multimodal data. Background Art

[0002] As one of the malignant tumors with the highest morbidity and mortality rates worldwide, colorectal cancer poses a serious threat to human health. According to relevant data from the World Health Organization, its incidence rate ranks among the highest among all types of cancers, and has been on the rise in recent years. In particular, in some developed countries and regions, the prevention and control of colorectal cancer is even more severe.

[0003] In the clinical diagnosis and treatment of colorectal cancer, prognostic evaluation has always been an extremely critical link. Accurate prognostic evaluation can provide an important basis for patients to develop personalized treatment plans, predict the risk of disease recurrence and metastasis, and rationally plan follow-up strategies, thereby significantly improving patients' survival rate and quality of life. However, traditional colorectal cancer prognostic evaluation methods mainly rely on clinical pathological indicators, such as the TNM stage of the tumor, histological grade, tumor marker levels, and patient age, gender and other factors. Although these indicators can reflect the severity of the patient's disease and prognosis to a certain extent, they can only provide relatively limited information, and it is difficult to comprehensively and accurately predict the individualized prognosis of colorectal cancer patients. Summary of the invention

[0004] In order to improve the problem that traditional colorectal cancer prognosis assessment methods are difficult to comprehensively and accurately predict the individualized prognosis of colorectal cancer patients, the present application provides colorectal cancer prognosis assessment methods, devices and media based on multimodal data.

[0005] In a first aspect, the present application provides a method for colorectal cancer prognosis assessment based on multimodal data, which adopts the following technical solution: A method for evaluating the prognosis of colorectal cancer based on multimodal data, wherein the multimodal data includes genomic data, proteomics data, microbiome data, metabolomics data, and imaging data, wherein the imaging data includes one of endoscopic images, PET-CT molecular images, and magnetic resonance images; the method for evaluating the prognosis of colorectal cancer includes: Acquire a multimodal dataset to be detected of a target patient, wherein the multimodal dataset to be detected includes test data and imaging data, wherein the test data includes at least two of genomic data, proteomics data, microbiome data, and metabolomics data collected from the patient after colorectal cancer surgery, and the imaging data is imaging data used to reflect the current surgical site of the target patient and the conditions of tissues or organs around the surgical site; Inputting the test data into a preset colorectal cancer feature detection model to obtain a plurality of first key features and a plurality of second key features, wherein the first key feature includes a plurality of sub-features corresponding to one or more omics data in the test data and indicating abnormalities, and the second key feature is an image manifestation of the first key feature; Verifying the second key feature according to the image data to obtain a verification result, wherein the verification result includes that the first key feature is correct and the first key feature is wrong; When the verification result shows that the first key feature is correct, the colorectal cancer prognosis of the target patient is evaluated based on the first key feature to obtain an evaluation result.

[0006] By adopting the above technical solutions, this application fully integrates the advantages of multimodal data. The test data covers a wide range of information from the genetic level to the microbial community and metabolic pathways, and can more comprehensively capture the complex molecular biological changes in colorectal cancer patients after surgery. The preset colorectal cancer feature detection model is used to analyze and mine these data, accurately extract the first key feature and its corresponding second key feature of the image manifestation, and effectively avoid the limitations and one-sidedness of a single data source; In addition, in the verification phase, the second key feature derived from other omics data was verified with the help of imaging data in multimodal data, further ensuring the accuracy and reliability of key feature extraction. When the verification results show that the first key feature is correct, the prognosis of colorectal cancer in patients can be evaluated based on these strictly screened and verified first key features, which can significantly improve the accuracy of prognosis evaluation.

[0007] In a specific implementation scheme, before inputting the test data into a preset colorectal cancer feature detection model, the method further includes: Obtain a multimodal reference data set of several specific patients in history, wherein the specific patients refer to all patients with stage II or stage III colorectal cancer who have undergone radical surgery, and each of the specific patients has a label, wherein the label is one of a first label indicating a good prognosis after surgery, a second label indicating the occurrence of colorectal cancer recurrence, and a third label indicating the occurrence of colon cancer metastasis, and the multimodal reference data set includes three sets of genomic data, proteomic data, microbiome data, and metabolomics data corresponding to each label; Establishing a colorectal cancer feature detection model, and training the colorectal cancer feature detection model using the test data; The training of the colorectal cancer feature detection model by using the test data includes: Inputting all the test data into the established colorectal cancer feature detection model; Several standard key features corresponding to the second label and the third label are obtained according to all the inspection data to complete the training.

[0008] In a specific embodiment, the standard key features refer to the key features that show consistent or typical performance in the specific patient population corresponding to one of the second label or the third label, and each of the labels corresponds to one or more of the standard key features.

[0009] In a specific embodiment, the genomic data include multiple gene mutation indicators such as single nucleotide polymorphism, copy number variation and insertion / deletion mutation; the proteomics data include multiple protein expression level indicators of tumor marker proteins, cell cycle regulatory proteins and apoptosis-related proteins; the microbiome data include multiple microbial community composition indicators and microbial diversity indicators classified at the genus and phylum levels; and the metabolomics data include concentration indicators of small molecule metabolites.

[0010] In a specific implementation manner, obtaining the plurality of first key features and the plurality of second key features comprises: Calculate the matching degree of each item of the above test data according to all standard key features corresponding to the second label and the third label respectively, and obtain the recurrence matching degree corresponding to the first label and the transfer matching degree corresponding to the second label, wherein the recurrence matching degree is used to reflect the similarity between the test data of the target patient and the multimodal reference data set of the specific patient corresponding to the second label, and the transfer matching degree is used to reflect the similarity between the test data of the target patient and the multimodal reference data set of the specific patient corresponding to the third label; Obtaining a first key feature according to the recurrence matching degree and the metastasis matching degree; Obtaining a second key feature according to the first key feature; The step of calculating the matching degree of each item of data in the above test data set according to all standard key features corresponding to the second label and the third label to obtain the recurrence matching degree corresponding to the first label and the transfer matching degree corresponding to the second label includes: Determine the target sub-features corresponding to each label in the inspection data according to the standard key features corresponding to the second label and the third label respectively, wherein the number of the target sub-features is consistent with the number of the standard key features corresponding to the corresponding labels and corresponds one to one; A recurrence matching degree is obtained according to all the target sub-features corresponding to the second label, and a transfer matching degree is obtained according to all the target sub-features corresponding to the third label.

[0011] In a specific implementation manner, obtaining the first key feature according to the recurrence matching degree and the metastasis matching degree includes: Determine a recurrence matching threshold and a transfer matching threshold, wherein the recurrence matching threshold=W1*the number of target sub-features corresponding to the second label in the test data, and the transfer matching threshold=W2*the number of target sub-features corresponding to the third label in the test data; wherein W1 and W2 are adjustment coefficients, which are used to change the size of the recurrence / transfer matching threshold according to actual conditions; When the recurrence matching degree is greater than or equal to the recurrence matching threshold, all the target sub-features that meet the corresponding standard key feature among all the target sub-features corresponding to the second label are used as the first key feature; When the transfer matching degree is greater than or equal to the transfer matching threshold, all the target sub-features corresponding to the third tag that meet the corresponding standard key feature are used as the first key feature.

[0012] In a specific implementation scheme, the verification result includes that the first key feature is correct and the first key feature is wrong; the second key feature is verified according to the image imaging data, and the verification result obtained includes: If the actual image performance in the image data matches the second key feature, then determining that the verification result is that the first key feature is correct; Otherwise, it is determined that the verification result is a first key feature error and the process returns to the step of determining the recurrence matching threshold and the transfer matching threshold.

[0013] In a specific implementation scheme, when the verification result is that the first key feature is correct, the colorectal cancer prognosis of the target patient is evaluated according to the recurrence matching degree and the metastasis matching degree, and the evaluation results obtained include: respectively determining a first weight value of the recurrence matching degree and a second weight value of the transfer matching degree; Calculate the comprehensive risk score of the target patient according to the weights of the recurrence matching degree and the metastasis matching degree, wherein the comprehensive risk score=first weight value*(recurrence matching degree / the number of all target sub-features corresponding to the second label in the test data)+second weight value*(metastasis matching degree / the number of all target sub-features corresponding to the third label in the test data), first weight value+second weight value=1; The prognosis level is divided according to the comprehensive risk score to obtain an evaluation result, wherein the evaluation result includes a relatively optimistic prognosis, an evaluation result that the prognosis is at a medium level, and a not optimistic prognosis; The prognostic grade is divided according to the comprehensive risk score to obtain the evaluation result, which includes: determining a first threshold and a second threshold; When the comprehensive risk score is less than or equal to the first threshold, determining that the assessment result is that the prognosis is relatively optimistic; When the comprehensive risk score is greater than the first threshold and less than the second threshold, determining that the assessment result is that the prognosis is at a medium level; When the comprehensive risk score is greater than the second threshold, it is determined that the assessment result is that the prognosis is not optimistic.

[0014] In a second aspect, the present application provides a smart terminal, which adopts the following technical solution: A smart terminal, characterized in that it includes a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement a colorectal cancer prognosis assessment method based on multimodal data as described in the first aspect.

[0015] In a third aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium, wherein at least one instruction, at least one program, code set or instruction set is stored in the readable storage medium, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement a colorectal cancer prognosis assessment method based on multimodal data as described in the first aspect.

[0016] In summary, the present application includes at least one of the following beneficial technical effects: 1. Advantages of multi-dimensional data integration: By integrating genomics, proteomics, microbiome and metabolomics data, it can comprehensively reflect the complex molecular biological change network in the body of colorectal cancer patients after surgery. It is no longer limited to the one-sided perspective of a single data type, but comprehensively considers multiple aspects such as gene variation, protein expression regulation, microbial community ecology and metabolic pathway remodeling, so as to more accurately capture the key features related to colorectal cancer recurrence and metastasis; these multimodal data are interrelated and complementary, and together construct a more complete and accurate colorectal cancer prognosis judgment system, which greatly improves the accuracy of predicting the progression of patients' diseases.

[0017] 2. The colorectal cancer feature detection model is trained using a large number of multimodal reference data sets of historically specific patients (stage II / III colorectal cancer patients who have undergone radical surgery), so that the model can learn the common characteristics and difference patterns of the patient population at the multimodal level under different prognostic labels (good prognosis after surgery, colorectal cancer recurrence, and colon cancer metastasis). For each patient's unique test data, this personalized model based on big data training can better adapt to the heterogeneity between different individual patients, avoiding the limitations of traditional general models in the face of complex individual differences, thereby providing each patient with a prognostic assessment that is more in line with their actual situation, further improving the accuracy and reliability of prognostic judgment.

[0018] 3. After extracting the first key feature related to recurrence and metastasis, the model further generated the corresponding second key feature of imaging manifestations based on the first key feature, and then verified the first key feature through the second key feature and imaging genomics features. This process formed a closed loop of two-way verification, confirming the key features from different angles. In this way, false feature extraction caused by data noise, model errors or other uncertain factors can be effectively eliminated, greatly improving the accuracy and reliability of the extracted first key feature. For example, if the first key feature is extracted unilaterally based on the model, some accidental data features that are not closely related to the actual condition may be misjudged as key features. However, after generating associations with imaging manifestations and subsequent reverse verification, the key features that are truly closely related to colorectal cancer recurrence and metastasis can be more accurately screened out.

[0019] 4. The first key feature is the abnormal features related to the development of the disease mined from the multimodal data level, while the second key feature is the intuitive presentation of these abnormal features in the image. By combining the second key feature with the actual imaging data for verification, a cross-verification from the perspective of molecular biology (multimodal data) to the perspective of anatomical imaging (imaging data) is achieved; data from different perspectives corroborate each other, making the determined key features more convincing. For example, from a genomics perspective, a specific mutation in a gene is found (the first key feature), and the corresponding imaging manifestation may be a specific morphological or density change in the intestine at the surgical site (the second key feature). When the imaging data does show such a change, it further confirms the effectiveness of the association between the gene mutation and the recurrence and metastasis of colorectal cancer, ensuring that the key features can truly reflect the development trend of the disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of a method for colorectal cancer prognosis assessment based on multimodal data according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0022] The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings.

[0023] An embodiment of the present application discloses a method for evaluating the prognosis of colorectal cancer based on multimodal data. The multimodal data in the present application includes genomic data, proteomics data, microbiome data, metabolomics data, and imaging data, wherein the imaging data includes one of endoscopic examination images, PET-CT molecular images, magnetic resonance imaging, and other imaging data.

[0024] Reference Figure 1 , a colorectal cancer prognosis assessment method based on multimodal data includes: S100, obtaining a multimodal dataset to be tested of a target patient; The target patient refers to a patient who currently needs to undergo prognosis assessment for colorectal cancer. The test data includes test data and imaging data. The test data includes at least two of genomic data, proteomics data, microbiome data, and metabolomics data collected from patients after colorectal cancer surgery. The imaging data is imaging data used to reflect the current surgical site of the target patient and the conditions of tissues or organs around the surgical site. Among them, genomic data include multiple gene mutation indicators such as single nucleotide polymorphism (SNP), copy number variation (CNV), insertion / deletion mutation (Indel), etc. These gene mutation indicators can be obtained through existing gene sequencing technology or gene chip technology (i.e., the corresponding gene test report); Proteomics data include multiple protein expression level indicators of common tumor marker proteins (embryonic antigen CEA, carbohydrate antigen 19-9, etc.), cell cycle regulatory proteins (cyclins, cyclin-dependent kinases CDKs and their inhibitors CKIs, etc.), and apoptosis-related proteins (apoptotic proteins Bax, Bad, anti-apoptotic proteins (Bcl-2, Bcl-xL, etc.), which can be obtained through existing mass spectrometry technology, enzyme-linked immunosorbent assay technology, etc. (i.e. blood test results); Microbiome data include multiple microbial community composition indicators classified at the genus level (Fusobacterium nucleatum, Bacteroides fragilis, Bifidobacterium, Lactobacillus, etc.) / classified at the phylum level (Firmicutes, Bacteroidetes, Proteobacteria, Actinobacteria, etc.) and microbial diversity indicators (species richness, Shannon diversity index, and Simpson diversity index), which can be obtained through metagenomic sequencing, analysis software and algorithms based on sequencing data (such as QIIME, Mothur) and other technologies (i.e., stool / intestinal mucosal tissue / blood sample test results); Metabolomics data include concentration indicators of small molecule metabolites (amino acid metabolites, carbohydrate metabolites, and fatty acid metabolites), which can be obtained through existing liquid chromatography-mass spectrometry, metabolic flux analysis and other technologies (i.e., blood / urine / tissue / feces sample test results).

[0025] S200, inputting the test data into a preset colorectal cancer feature detection model to obtain a plurality of first key features and a plurality of second key features; Among them, the imaging data of this embodiment takes several abdominal CT scan images of the target patient as an example; the first key feature includes several sub-features corresponding to one or more omics data in the test data that are abnormal, wherein genomic data corresponds to several gene sub-features, proteomics data corresponds to several protein sub-features, microbiomics data corresponds to several microbial sub-features, and metabolomics data corresponds to several metabolic sub-features; the second key feature is the imaging manifestation of the first key feature.

[0026] It should be noted that the colorectal cancer feature detection model preset in S200 is mainly used to obtain multiple first key features corresponding to the target patient based on the test data of the target patient, and the first key feature is an abnormal feature related to the development of colorectal cancer in the corresponding omics data. The meaning of the first key feature is briefly explained below: For genomic data, the corresponding gene sub-features are mutations of certain specific genes related to colorectal cancer, which can be expressed as mutations in the XX gene. For example, the KRAS gene mutation is more common in colorectal cancer. The protein encoded by this gene participates in the intracellular signal transduction pathway and is used to regulate cell proliferation, differentiation and other processes. However, the mutated KRAS protein is in a continuously activated state, which causes the intracellular growth-promoting signals to be continuously transmitted, thereby enhancing the proliferation, invasiveness and migration ability of tumor cells, which in turn makes the tumor more likely to recur and metastasize to distant sites. For proteomics data, the corresponding protein sub-features are abnormal expression levels of certain key proteins related to colorectal cancer, which can be expressed as abnormal expression of XX protein; excessive or low expression of some proteins may disrupt normal cell metabolism and signal transduction pathways, thereby promoting the progression of colorectal cancer, such as the classic colorectal cancer-related protein marker embryonic antigen (CEA). This protein often causes an increase in serum concentration in the patient's blood when the tumor recurs and metastasizes. This is because when tumor cells recur or metastasize to other sites, the tumor cells will continue to secrete CEA at the current site, causing the CEA level in the blood to continue to increase. Therefore, in clinical practice, dynamic monitoring of CEA changes can be used to assist in determining whether the tumor has recurred or metastasized.

[0027] For the microbiome data, the corresponding microbial sub-features are the imbalanced states of certain specific microbial communities related to colorectal cancer, which can be expressed as an imbalance in the state of XX microbial communities; specific microbial communities include harmful microbial communities such as Fusobacterium nucleatum and Bacteroides fragilis, and beneficial microbial communities such as Bifidobacterium and Lactobacillus. Among them, the excessive growth of harmful microorganisms or the continuous reduction of beneficial microorganisms may affect the intestinal microenvironment, immune regulation and other aspects, thereby creating conditions for the growth and deterioration of colorectal cancer; for example, the overgrowth of Fusobacterium nucleatum can promote the recurrence and metastasis of colorectal cancer through a variety of mechanisms. It can adhere to the surface of tumor cells and enhance the invasiveness of tumor cells by activating related signaling pathways in tumor cells. It can also regulate the immune response in the tumor microenvironment and inhibit the body's anti-tumor immunity, making it easier for tumor cells to escape immune surveillance and then spread in the body, leading to an increased risk of recurrence and metastasis.

[0028] For the metabolomics data, the corresponding metabolite feature is the abnormal change of the body's metabolites related to colorectal cancer, which can be expressed as abnormal changes in XX metabolites; in this embodiment, the abnormal changes in the body's metabolites include abnormal levels of energy / amino acid metabolites and changes in metabolic pathways, and when the metabolomics data of the target patient meets one of the above two (abnormal levels of energy / amino acid metabolites), the metabolite feature is abnormal changes in XX metabolites; for example, tumor cells in the colorectum will change the uptake and metabolism of amino acids, such as increased uptake of glutamine: glutamine can participate in the synthesis pathways of various organisms in the cell, provide nitrogen sources, carbon sources and other material bases for tumor cells, support the rapid proliferation of tumor cells and metabolic needs during metastasis, therefore, the increase in glutamine uptake is very likely to help tumor proliferation / metastasis, leading to an increased risk of recurrence and metastasis, therefore, the accumulation of the above metabolic intermediates can be regarded as one of the manifestations in the development of colorectal cancer.

[0029] In order to better illustrate the colorectal cancer prognosis method provided in this embodiment, before specifically explaining the step of S200, the following first illustrates the construction and training process of the colorectal cancer feature detection model preset in S200 through steps S110-S120: S110, obtaining a multimodal reference dataset of several specific patients in history; Among them, specific patients refer to all stage II / III colorectal cancer patients who have received radical surgery, and each specific patient has a label, which can be one of good postoperative prognosis (first label), colorectal cancer recurrence (second label), and colon cancer metastasis (third label). This embodiment takes the example that the number of specific patients under each label is 100; the multimodal reference data set includes three groups of genomics data, proteomics data, microbiomics data, and metabolomics data corresponding to each label.

[0030] S120, establishing a colorectal cancer feature detection model, and training the colorectal cancer feature detection model using a multimodal reference dataset; Specifically, the steps of training the colorectal cancer feature detection model in S120 include: S121, input all multimodal reference datasets into the established colorectal cancer feature detection model; Among them, the colorectal cancer feature detection model will classify all multimodal reference datasets according to the labels of specific patients corresponding to all multimodal reference datasets.

[0031] S122, obtaining a number of standard key features corresponding to the second label and the third label respectively according to all multimodal reference data sets to complete the training; Among them, the standard key feature refers to the key feature that shows consistent or typical performance in a specific patient group corresponding to a certain label (the second label or the third label), and each label can correspond to one or more standard key features; specifically, the standard key feature is mainly determined based on the statistical analysis results of various indexes in the corresponding omics data. In this embodiment, the multimodal reference data set of a specific patient population labeled as having a good postoperative prognosis (the first label) is mainly used as a reference group, and then the multimodal reference data sets corresponding to the other two labels are compared with the reference group to obtain the corresponding standard key features. The following is an example of a specific patient group labeled as colorectal cancer recurrence (the second label) to illustrate how to obtain the standard key features (the principle of the method for obtaining the standard key features corresponding to the third label is the same as the principle of the following steps, and will not be repeated in this application): For genomic data, after collecting relevant data such as genome sequencing of specific patients under many second labels, various indicators of each gene (such as SNP site frequency, CNV status, etc.) will be statistically summarized; for example, for the SNP site of a specific gene, the frequency of specific mutations at this site in all characteristic patients is counted, and its average value is calculated. If this average value is significantly higher than that of a specific patient population with a good prognosis after surgery, and this increase shows a certain consistency in the recurrent patient population, then the corresponding frequency value is determined as the relevant value in the standard key feature. Similarly, for the copy number variation of oncogenes and tumor suppressor genes, the numerical range of their respective copy number variations in the recurrent patient population is also statistically analyzed, and the representative average value (such as the oncogene CNV is as high as a certain value and the tumor suppressor gene CNV is as low as a certain value) is taken to define it, so as to reflect that the specific patient population with recurrent colorectal cancer has common gene mutation characteristics at the genomic level; For proteomics data, statistical analysis is performed on the expression of each tumor marker protein, cell cycle regulatory protein, apoptosis-related protein, etc.; for example, for a specific tumor marker protein, the expression values ​​of the protein in all corresponding specific patients are collected, and the average expression level is calculated. If this level is significantly higher than that of the patient group with a good prognosis after surgery, and such high expression is common in the relapsed patient group, this higher expression value is set as the expression level value corresponding to the protein in the standard key feature (such as the expression level of XX protein is as high as a certain value). For key proteins that regulate the cell cycle, such as cyclin-dependent kinase inhibitors, their expression levels in the relapsed patient group are also counted. If they are generally low-expressed, the average value of their low expression (such as the expression level of cyclin-dependent kinase inhibitors is as low as a certain value) is taken as the standard key feature; For microbiome data, for specific microorganisms (such as harmful bacteria such as Fusobacterium nucleatum or beneficial bacteria such as Bifidobacterium), the number of specific microorganisms in the intestines of each specific patient is counted, and the average number is calculated. If it is found that the number of bacteria generally increases (such as the number of XX bacteria increases to a certain value) or decreases in the recurrent patient group, and this change is closely related to tumor recurrence, the numerical change value is determined as one of the standard key features. At the same time, for the relative proportion between different bacterial phyla (such as Firmicutes and Bacteroidetes), the above relative proportions of all corresponding specific patients are counted. If the relative proportion shows a significant and consistent increase or decrease compared with the specific patient population with a good prognosis after surgery (such as the relative proportion of XX bacterial phyla to XX bacterial phyla increases / decreases to a certain value), it will be included in the category of standard key features; For metabolomics data, for small molecule metabolites such as glutamine, glucose, and lactate, their synthesis amounts or concentrations in the corresponding specific patient groups are statistically analyzed. If an increase or decrease in synthesis amount is common (such as the synthesis amount of XX metabolite increases / decreases to a certain value), and the change is consistent with the common metabolic characteristics of the group, the corresponding change value will be used as part of the standard key features.

[0032] The following are standard key features corresponding to the specific patient population with the label of colorectal cancer recurrence (secondary label): The standard key features corresponding to genomic data are: the SNP site frequency of the XX gene increases to a certain value, the oncogene CNV is as high as a certain value, and the tumor suppressor gene CNV is as low as a certain value; the standard key features corresponding to proteomics data are: the expression level of the XX protein is as high as a certain value, and the expression level of the cyclin-dependent kinase inhibitor is as low as a certain value; the standard key features corresponding to microbiomics data are: the number of XX bacteria increases to a certain value, and the relative proportion of the XX phylum to the XX phylum increases to a certain value; the standard key features corresponding to metabolomics data are: the synthesis amount of the XX metabolite decreases to a certain value; wherein, the above-mentioned "certain value" in this embodiment takes the average value of all corresponding parameters (one of the above-mentioned SNP site frequency, oncogene CNV, expression level of XX protein, etc.) in the corresponding specific patient population.

[0033] Specifically, the steps of obtaining the first key feature and the second key feature by the colorectal cancer feature detection model in S200 include: S210, calculating the matching degree of each item of the test data according to all standard key features corresponding to the second label and the third label, and obtaining the recurrence matching degree corresponding to the first label and the transfer matching degree corresponding to the second label; Among them, the recurrence matching degree is used to reflect the similarity between the test data of the target patient and the multimodal reference data set of a specific patient with a label of colorectal cancer recurrence, and the metastasis matching degree is used to reflect the similarity between the test data of the target patient and the multimodal reference data set of a specific patient with a label of colon cancer metastasis; specifically, in this embodiment, the calculation of the recurrence matching degree and the metastasis matching degree are both quantitatively scored: S211, determining the target sub-feature corresponding to each label in the inspection data according to the standard key features corresponding to the second label and the third label respectively; The number of target sub-features corresponding to each tag is consistent with the number of standard key features corresponding to the tag and corresponds one to one, such as "the frequency of SNP sites of gene a increases to X 1 The standard key feature of "corresponds to "the frequency of SNP sites in gene a is X 2"The target sub-features of ", the relative proportion of phylum b and phylum c is reduced to Y 1 The standard key feature of "corresponds to the relative proportion of b bacteria to c bacteria is Y 2 "The target sub-features.

[0034] S212, obtaining a recurrence matching degree according to all target sub-features corresponding to the second label, and obtaining a transfer matching degree according to all target sub-features corresponding to the third label; Specifically, the calculation method of recurrence matching is as follows (the calculation of transfer matching is similar and will not be repeated here): S2121, if the target sub-feature meets the corresponding standard key feature, the recurrence matching score is increased by one; otherwise, the recurrence matching score remains unchanged; Among them, the target sub-feature conforms to the corresponding standard key feature means that the difference between the feature value in the target sub-feature and the feature value in the standard key feature is within a preset error range. The error range in this embodiment takes 2% of the feature value in the standard key feature as an example.

[0035] It should be noted that, in fact, the error range in step S2121 needs to be determined separately according to the specificity of each standard key feature in reflecting the development of colorectal cancer; at the same time, for the same group feature data, if there are multiple corresponding standard key features, the matching values ​​corresponding to each standard key feature need to be weighted and summed according to a certain weight, and the weight can be determined in advance according to factors such as the importance of each standard key feature in the disease recurrence mechanism, for example, the weight of gene mutation is set to 0.6, and the weight related to gene expression regulation is set to 0.4; in addition, in actual situations, the tests completed by patients may be uneven due to physical / economic / time reasons, so in this embodiment, if the test data lacks a target sub-feature corresponding to a certain standard key feature, the standard key feature is directly ignored and deemed to be not in compliance with the error range condition (that is, the original recurrence / metastasis matching score is kept unchanged); next, combined with the above weight distribution method, the calculation of the recurrence matching score is described through an example: Assume that in the current colorectal cancer feature detection model, the standard key features corresponding to the second label are: ① The SNP site frequency of gene a increases to X 1 ②Oncogene CNV up to Z 1 ③The relative proportion of phylum b and phylum c decreased to Y 1 ④d protein expression level was as high as S 1 ⑤e metabolite synthesis is reduced to T 1 ; The target sub-features corresponding to the second label in the test data of the target patient are: (1) The SNP site frequency of gene a is X 2 ; (2) Oncogene CNV is Z 2; (3) The relative proportion of phylum b to phylum c is Y 2 ; (5) The amount of metabolite synthesis is T 2 ; If only ① and (1), ⑤ and (5) meet the error range conditions, then the recurrence matching degree corresponding to the target patient = 0.6*1+0.4*0+0+0+1=1.6.

[0036] S220, obtaining a first key feature according to the recurrence matching degree and the metastasis matching degree; Specifically, S220 includes: S221, determining a recurrence matching threshold and a metastasis matching threshold; Where, recurrence matching threshold = W 1 *The number of target sub-features in the test data that correspond to the second label, transfer matching threshold = W 2 *The number of target sub-features corresponding to the third label in the test data; where W 1 , W 2 are adjustment coefficients, which are used to change the recurrence / metastasis matching threshold according to the actual situation. 1 With W 2 Take 0.5 as an example.

[0037] S222, when the recurrence matching degree is greater than or equal to the recurrence matching threshold, all target sub-features that meet the corresponding standard key feature among all target sub-features corresponding to the second label are used as first key features; Among them, a recurrence matching degree greater than or equal to the recurrence matching threshold represents that the target patient has a greater probability of recurrence of colorectal cancer, and all target sub-features that meet the corresponding standard key features represent that the target sub-feature is abnormal and similar to the characteristics manifested by colorectal cancer recurrence. The first key feature can be understood as an abnormal conclusion related to the corresponding key sub-feature.

[0038] S223, when the transfer matching degree is greater than or equal to the transfer matching threshold, all target sub-features that meet the corresponding standard key features among all target sub-features corresponding to the third tag are used as first key features; Among them, a metastasis matching degree greater than or equal to the metastasis matching threshold indicates that the target patient has a greater probability of colorectal cancer metastasis, and all target sub-features that meet the corresponding standard key features indicate that the target sub-feature is abnormal and similar to the characteristics exhibited by colorectal cancer metastasis.

[0039] S230, obtaining a second key feature according to the first key feature; Among them, the second key feature is the feature of the first key feature that appears in some form on the image of the patient's surgical site, including the morphological features and density features of the intestine at the surgical site in the image data, as well as the relative position relationship features of the surrounding tissues and organs and the surgical site, etc. For example, when the first key feature is a mutation in the KRAS gene, the corresponding second key feature is unclear image boundaries, lobed changes, surrounding tissue infiltration, uneven density, etc. When the first key feature is an imbalance in the state of the Fusobacterium nucleatum community, the corresponding second key feature is intestinal mucosal congestion, edema, rough surface, erosion, ulcer formation, etc.; specifically, the association relationships between different target sub-features and the second key features are stored in the intelligent terminal that implements the method of the present application so that they can be associated and called at any time. These association relationships are common knowledge in the field and can be known through medical research papers and reviews, image differential diagnosis manuals, etc. in this field, and will not be repeated here.

[0040] S300, verifying the second key feature according to the image data to obtain a verification result; In this embodiment, step S300 is mainly performed by a professional doctor to verify the second key feature by comparing several image data of the target patient; specifically, S300 includes: S310, if the actual image performance in the image data matches the second key feature, then determining that the verification result is that the first key feature is correct; S320, otherwise, determine that the verification result is a first key feature error and return to S221 to re-determine a new recurrence matching threshold and a transfer matching threshold; The new recurrence matching threshold or the transfer matching threshold is mainly determined by modifying the adjustment coefficient W1 or W2.

[0041] S400, when the verification result is that the first key feature is correct, the colorectal cancer prognosis of the target patient is evaluated according to the recurrence matching degree and the metastasis matching degree to obtain an evaluation result; Specifically, S400 includes: S410, respectively determining a first weight value of the recurrence matching degree and a second weight value of the transfer matching degree; Among them, the respective weights of the recurrence matching degree and the metastasis matching degree are mainly determined based on clinical research and past experience. Since the recurrence and metastasis of colorectal cancer have different degrees of influence on the final course of the patient's condition in the prognosis judgment of colorectal cancer, the allocation weights of the two will also be different. In this embodiment, the weight of the impact of colorectal cancer recurrence on the patient's long-term survival and quality of life is 0.6, and the weight of metastasis is 0.4 as an example (the weight values ​​here are only examples. In actual application, they need to be adjusted based on in-depth and accurate analysis and research on a large number of cases. Generally, it is necessary to comprehensively consider multiple factors, such as the difficulty of treatment after colorectal cancer recurrence and metastasis, the degree of damage to the functions of various organs in the body, and the relative importance of the two in increasing the risk of death in patients).

[0042] S420, calculating a comprehensive risk score of the target patient according to the weights of the recurrence matching degree and the metastasis matching degree; Among them, the comprehensive risk score = the first weight value * (recurrence matching degree / the number of all target sub-features corresponding to the second label in the test data) + the second weight value * (transfer matching degree / the number of all target sub-features corresponding to the third label in the test data), the first weight value + the second weight value = 1.

[0043] S430, dividing the prognosis level according to the comprehensive risk score to obtain the evaluation result; The evaluation results include: a relatively optimistic prognosis, a moderate prognosis, and a not optimistic prognosis. Specifically, S330 includes: S431, determining a first threshold and a second threshold; The first threshold is smaller than the second threshold. In this embodiment, the first threshold is 0.3 and the second threshold is 0.7.

[0044] S432, when the comprehensive risk score is less than or equal to the first threshold, determining that the assessment result is that the prognosis is relatively optimistic; Among them, a comprehensive risk score less than or equal to the first threshold indicates that the target patient has a relatively low possibility of colorectal cancer recurrence and metastasis, and the prognosis is relatively optimistic; in terms of follow-up treatment, routine consolidation treatment and regular review may be the main treatment. It is expected that the patient's condition will remain stable for a long period of time, and the quality of life will be relatively less affected by the disease. The 5-year survival rate (only a reference indicator, which can be adjusted according to actual conditions) may reach 80% or above.

[0045] S433, when the comprehensive risk score is greater than the first threshold and less than the second threshold, determining that the assessment result is that the prognosis is at a medium level; Among them, a comprehensive risk score greater than the first threshold and less than the second threshold means that the target patient has a certain degree of risk of colorectal cancer recurrence and metastasis, and the prognosis is at a medium level; for such patients, it is necessary to closely monitor changes in the disease, and auxiliary treatment measures may need to be strengthened, such as appropriate adjustment of chemotherapy regimens, increase of targeted therapy, etc., and the review interval should be shortened to timely detect possible recurrence or metastasis. The 5-year survival rate may be between 30% and 80%.

[0046] S434, when the comprehensive risk score is greater than the second threshold, determining that the assessment result is that the prognosis is not optimistic; Among them, a comprehensive risk score greater than the second threshold indicates that the target patient has a high risk of colorectal cancer recurrence and metastasis, and the prognosis is not optimistic; clinicians need to actively adjust treatment strategies and consider more radical comprehensive treatment methods, such as combining multiple chemotherapy drugs and cooperating with immunotherapy, etc. At the same time, they should provide patients with psychological counseling and life care and other support. The 5-year survival rate of such patients is often less than 30%.

[0047] Based on the same inventive concept mentioned above, an embodiment of the present application also discloses an intelligent terminal, which includes a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement a colorectal cancer prognosis assessment method based on multimodal data as provided in the above method embodiment.

[0048] Based on the same inventive concept mentioned above, an embodiment of the present application also discloses a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set can be loaded and executed by a processor to implement a colorectal cancer prognosis assessment method based on multimodal data provided in the above method embodiment.

[0049] It should be understood that the "plurality" mentioned in this article refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0050] A person skilled in the art can understand that all or part of the steps of implementing the above-mentioned embodiments can be completed by hardware, or can be completed by a program to instruct the relevant hardware. The program can be stored in a computer-readable storage medium. The above-mentioned storage medium includes, for example: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0051] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for colorectal cancer prognosis assessment based on multimodal data, characterized in that: The multimodal data includes genomic data, proteomics data, microbiome data, metabolomics data and imaging data, and the imaging data includes one of endoscopic images, PET-CT molecular images and magnetic resonance images; the colorectal cancer prognosis assessment method includes: Acquire a multimodal dataset to be detected of a target patient, wherein the multimodal dataset to be detected includes test data and imaging data, wherein the test data includes at least two of genomic data, proteomics data, microbiome data, and metabolomics data collected from the patient after colorectal cancer surgery, and the imaging data is imaging data used to reflect the current surgical site of the target patient and the conditions of tissues or organs around the surgical site; Inputting the test data into a preset colorectal cancer feature detection model to obtain a plurality of first key features and a plurality of second key features, wherein the first key feature includes a plurality of sub-features corresponding to one or more omics data in the test data and indicating abnormalities, and the second key feature is an image manifestation of the first key feature; Verifying the second key feature according to the image data to obtain a verification result, wherein the verification result includes that the first key feature is correct and the first key feature is wrong; When the verification result shows that the first key feature is correct, the colorectal cancer prognosis of the target patient is evaluated based on the first key feature to obtain an evaluation result.

2. The method for colorectal cancer prognosis according to claim 1, characterized in that: Before inputting the test data into a preset colorectal cancer feature detection model, the method further includes: Obtain a multimodal reference data set of several specific patients in history, wherein the specific patients refer to all patients with stage II or stage III colorectal cancer who have undergone radical surgery, and each of the specific patients has a label, wherein the label is one of a first label indicating a good prognosis after surgery, a second label indicating the occurrence of colorectal cancer recurrence, and a third label indicating the occurrence of colon cancer metastasis, and the multimodal reference data set includes three sets of genomic data, proteomic data, microbiome data, and metabolomics data corresponding to each label; Establishing a colorectal cancer feature detection model, and training the colorectal cancer feature detection model using the test data; The training of the colorectal cancer feature detection model by using the test data includes: Inputting all the test data into the established colorectal cancer feature detection model; Several standard key features corresponding to the second label and the third label are obtained according to all the inspection data to complete the training.

3. The method for colorectal cancer prognosis according to claim 2, characterized in that: The standard key features refer to the key features that show consistent or typical performance in the specific patient group corresponding to one of the second label or the third label, and each of the labels corresponds to one or more of the standard key features.

4. The method for colorectal cancer prognosis according to claim 3, characterized in that: The genomics data include multiple gene mutation indicators such as single nucleotide polymorphism, copy number variation and insertion / deletion mutation; the proteomics data include multiple protein expression level indicators of tumor marker proteins, cell cycle regulatory proteins and apoptosis-related proteins; the microbiome data include multiple microbial community composition indicators and microbial diversity indicators classified at the genus and phylum levels; the metabolomics data include concentration indicators of small molecule metabolites.

5. The method for colorectal cancer prognosis according to claim 2, characterized in that: The obtaining of the plurality of first key features and the plurality of second key features comprises: Calculate the matching degree of each item of the above test data according to all standard key features corresponding to the second label and the third label respectively, and obtain the recurrence matching degree corresponding to the first label and the transfer matching degree corresponding to the second label, wherein the recurrence matching degree is used to reflect the similarity between the test data of the target patient and the multimodal reference data set of the specific patient corresponding to the second label, and the transfer matching degree is used to reflect the similarity between the test data of the target patient and the multimodal reference data set of the specific patient corresponding to the third label; Obtaining a first key feature according to the recurrence matching degree and the metastasis matching degree; Obtaining a second key feature according to the first key feature; The step of calculating the matching degree of each item of the above test data according to all standard key features corresponding to the second label and the third label to obtain the recurrence matching degree corresponding to the first label and the transfer matching degree corresponding to the second label includes: Determine the target sub-features corresponding to each label in the inspection data according to the standard key features corresponding to the second label and the third label respectively, wherein the number of the target sub-features is consistent with the number of the standard key features corresponding to the corresponding labels and corresponds one to one; A recurrence matching degree is obtained according to all the target sub-features corresponding to the second label, and a transfer matching degree is obtained according to all the target sub-features corresponding to the third label.

6. The method for colorectal cancer prognosis according to claim 5, characterized in that: The obtaining of the first key feature according to the recurrence matching degree and the metastasis matching degree comprises: Determine a recurrence matching threshold and a transfer matching threshold, wherein the recurrence matching threshold=W1*the number of target sub-features corresponding to the second label in the test data, and the transfer matching threshold=W2*the number of target sub-features corresponding to the third label in the test data; wherein W1 and W2 are adjustment coefficients, which are used to change the size of the recurrence / transfer matching threshold according to actual conditions; When the recurrence matching degree is greater than or equal to the recurrence matching threshold, all the target sub-features that meet the corresponding standard key feature among all the target sub-features corresponding to the second label are used as the first key feature; When the transfer matching degree is greater than or equal to the transfer matching threshold, all the target sub-features corresponding to the third tag that meet the corresponding standard key feature are used as the first key feature.

7. The method for colorectal cancer prognosis according to claim 6, characterized in that: The verification result includes whether the first key feature is correct or wrong; the second key feature is verified according to the image data, and the verification result obtained includes: If the actual image performance in the image data matches the second key feature, then determining that the verification result is that the first key feature is correct; Otherwise, it is determined that the verification result is a first key feature error and the process returns to the step of determining the recurrence matching threshold and the transfer matching threshold.

8. The method for colorectal cancer prognosis according to claim 7, characterized in that: When the verification result shows that the first key feature is correct, the colorectal cancer prognosis of the target patient is evaluated according to the recurrence matching degree and the metastasis matching degree, and the evaluation results include: respectively determining a first weight value of the recurrence matching degree and a second weight value of the transfer matching degree; Calculate the comprehensive risk score of the target patient according to the weights of the recurrence matching degree and the metastasis matching degree, wherein the comprehensive risk score=first weight value*(recurrence matching degree / the number of all target sub-features corresponding to the second label in the test data)+second weight value*(metastasis matching degree / the number of all target sub-features corresponding to the third label in the test data), first weight value+second weight value=1; The prognosis level is divided according to the comprehensive risk score to obtain an evaluation result, wherein the evaluation result includes a relatively optimistic prognosis, an evaluation result that the prognosis is at a medium level, and a not optimistic prognosis; The prognostic grade is divided according to the comprehensive risk score to obtain the evaluation result, which includes: determining a first threshold and a second threshold; When the comprehensive risk score is less than or equal to the first threshold, determining that the assessment result is that the prognosis is relatively optimistic; When the comprehensive risk score is greater than the first threshold and less than the second threshold, determining that the assessment result is that the prognosis is at a medium level; When the comprehensive risk score is greater than the second threshold, it is determined that the assessment result is that the prognosis is not optimistic.

9. An intelligent terminal, characterized in that: It comprises a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement a colorectal cancer prognosis assessment method based on multimodal data as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by a processor to implement a colorectal cancer prognosis assessment method based on multimodal data as described in any one of claims 1 to 8.

Images