Heterogeneity non-invasive typing method in hepatoma based on multiple omics and application

Abstract

The invention provides a non-invasive typing method for heterogeneity in hepatoma based on multiple omics and application of the non-invasive typing method. According to the method, by analyzing the multi-sequence magnetic resonance image data and combining radiomics, machine learning and multi-omics analysis technologies, the liver cancer patients are divided into different radiology types, the prognosis of the liver cancer patients is predicted, and a basis is provided for personalized precise treatment. The method is high in accuracy and can effectively guide clinical treatment decisions.

Description

Technical Field

[0001] The present invention relates to the fields of medical image analysis, computer-aided diagnosis and precision tumor treatment, and in particular to a multi-omics-based non-invasive typing method for intratumor heterogeneity of liver cancer and its application. Background Art

[0002] Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related deaths worldwide. Intratumor heterogeneity is an important factor in the failure of HCC treatment and disease recurrence. Traditional methods for detecting intratumor heterogeneity mainly rely on invasive biopsies, which have shortcomings such as large trauma, high cost, and difficulty in obtaining samples. In recent years, imaging genomics technology combined with artificial intelligence technology has shown great potential in tumor diagnosis and treatment. Non-invasive analysis of intratumor heterogeneity of HCC based on imaging genomics technology and tumor classification of HCC are expected to guide individualized precision treatment. Summary of the invention

[0003] The purpose of the present invention is to provide a multi-omics-based non-invasive typing method and application of liver cancer intratumor heterogeneity, in view of the deficiencies of the prior art. The core idea is to use multi-sequence magnetic resonance imaging imaging omics data, combined with transcriptomics and metabolomics data, to comprehensively characterize the intratumor heterogeneity of liver cancer, and ultimately achieve non-invasive typing of liver cancer patients and guide individualized precision treatment.

[0004] The technical solutions adopted to solve the above technical problems are as follows:

[0005] A non-invasive method for subtyping intratumor heterogeneity of liver cancer based on multi-omics imaging, including:

[0006] Collect multi-sequence magnetic resonance imaging data of several patients with liver cancer and extract imaging omics features respectively;

[0007] The SNF algorithm was used to fuse the radiomic features of several liver cancer patients to obtain a fusion network;

[0008] Based on the fusion network, spectral clustering algorithm was used to cluster liver cancer patients into different imaging types.

[0009] Furthermore, the multi-sequence magnetic resonance imaging data includes: T2WI, DWI, NCP, AP, PVP and DP.

[0010] Furthermore, the imaging genomics features include:

[0011] Several first-order features, several texture features, several logarithmic features, and several wavelet features.

[0012] Furthermore, after extracting the radiomics features, it also includes:

[0013] The Pearson correlation coefficient was used to calculate the correlation between features and remove highly correlated features.

[0014] Furthermore, the parameter settings of the SNF algorithm include: K value=20, sigma value=0.5, and T value=20.

[0015] Furthermore, the spectral clustering algorithm is a spectral clustering algorithm based on similarity network fusion.

[0016] An application of the typing results obtained by the non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics includes:

[0017] Study the intratumor heterogeneity of liver cancer;

[0018] Construct a liver cancer classification system;

[0019] Construct a liver cancer prognosis diagnosis system.

[0020] A liver cancer prognosis diagnosis system, comprising:

[0021] The data acquisition module is used to collect multi-sequence magnetic resonance imaging data of liver cancer patients and extract radiomics features;

[0022] A classification module is used to determine the imaging classification of liver cancer patients; it includes: extracting imaging omics features based on multi-sequence magnetic resonance imaging data of liver cancer patients; reducing the dimension of the obtained features by principal component analysis and projecting them into a coordinate system consistent with the latent space of the previous training samples to obtain a vector representation for the final classification task; using a distance-weighted k-nearest neighbor algorithm to find the k training samples with the closest Euclidean distance to the vector representation, and predicting the imaging classification of liver cancer patients based on weighted voting results of the categories to which the k training samples belong;

[0023] The prognostic module determines the patient's prognosis based on the typing results.

[0024] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the methods for non-invasive typing of intratumor heterogeneity of liver cancer based on multi-omics imaging omics is implemented.

[0025] A storage medium comprising computer executable instructions, which, when executed by a computer processor, implement the non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics.

[0026] The present invention analyzes multi-sequence magnetic resonance imaging data, combines radiomics, machine learning and multi-omics analysis technology, divides liver cancer patients into different imaging types, predicts their prognosis, and formulates individualized treatment plans based on the prognosis and biological characteristics of different imaging types. The present invention uses a multi-center independent data set for external validation to ensure the robustness and scalability of the method of the present invention. The method has high accuracy and can effectively guide clinical treatment decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Visualization of the difference results of radiomics features of liver cancer with different imaging classifications;

[0028] Figure 2 This is a graph showing the survival differences among liver cancers with different imaging classifications (high heterogeneity, low heterogeneity, and moderate heterogeneity); Figure 2 A in the figure is the OS survival curve of liver cancer with different imaging classifications based on the training set; Figure 2 B in the figure is the RFS survival curve of liver cancer with different imaging classifications based on the training set; Figure 2 C in the figure is the OS survival curve of liver cancer with different imaging classifications based on validation set 1; Figure 2 D in the figure is the RFS survival curve of liver cancer with different imaging classifications based on validation set 1; Figure 2 E in the figure is the RFS survival curve of liver cancer with different imaging classifications based on validation set 2; Figure 2 F in the figure is the RFS survival curve of liver cancer with different imaging classifications based on the validation set 3;

[0029] Figure 3 This is a graph showing the analysis results of differential gene and pathway characteristics between tumors with high heterogeneity imaging phenotype and tumors with low heterogeneity imaging phenotype; Figure 3 A in the figure is a volcano plot, showing differentially expressed genes in high heterogeneity and low heterogeneity phenotypes; Figure 3 Figure B is a GSEA analysis confirming the upregulation of the ECM-receptor interaction pathway in the highly heterogeneous phenotype, indicating that the ECM-receptor interaction pathway is significantly enriched in the highly heterogeneous imaging phenotype; Figure 3 C in the figure is the NES ranking based on GSEA analysis; Figure 3 D in the figure is the ECM score of high, medium, and low heterogeneity in the training set, indicating the difference in ECM score among different heterogeneous phenotype groups; Figure 3 E in the figure is the expression statistics of ECM-related genes in different imaging types (high heterogeneity, low heterogeneity, and medium heterogeneity);

[0030] Figure 4 This is a graph showing the results of identifying therapeutic targets through transcriptomics. Figure 4A in the figure is a box plot of the abundance of immune cells in different imaging types (high heterogeneity, low heterogeneity, and moderate heterogeneity); Figure 4 Figure B is a t-SNE diagram of cells from 15 liver cancer patients, with 8 clusters; Figure 4 C in the figure is a bar graph of the content of various types of cells in different imaging classifications (high heterogeneity, low heterogeneity, and moderate heterogeneity), showing the proportion of cell types in different clusters; Figure 4 D in the figure is the result of GSVA analysis of tumor-associated fibroblasts (CAFs) based on heterogeneous phenotypes;

[0031] Figure 5 This is a diagram showing the differential cell communication results between high heterogeneity liver cancer and low heterogeneity liver cancer. Figure 5 Figure A is the result of analyzing cell communication between eight cell types using CellChat; Figure 5 Panel B is a heat map based on the analysis of the cell communication results, indicating the number of predicted ligand and receptor pairs between different cell types. DETAILED DESCRIPTION

[0032] The present invention provides a non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics, comprising the following steps:

[0033] Step 1: Collect multi-sequence magnetic resonance imaging data of several liver cancer patients and extract imaging features respectively;

[0034] Step 2: Use the SNF algorithm to fuse the radiomic features of several liver cancer patients to obtain a fusion network;

[0035] Step 3: Clustering is performed using the spectral clustering algorithm based on the fusion network to divide liver cancer patients into different imaging types.

[0036] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0037] In this example, a total of 851 patients with liver cancer MRI enhancement image data sets were collected from four centers as experimental data. The data from the first center was divided into a training set, and the data from the other three centers were divided into independent validation sets. The experimental environment was macOS system, and the platform was Pytorch3.6.

[0038] The method of this embodiment is as follows:

[0039] Step 1: Collect multi-sequence magnetic resonance imaging data of several liver cancer patients and extract imaging features respectively; including the following sub-steps:

[0040] Step 1.1: Collect multi-sequence MRI data of several liver cancer patients and perform image standardization preprocessing:

[0041] The MR imaging protocol of this embodiment follows the guidelines established by the American Association for the Study of Liver Diseases. Each magnetic resonance sequence was bias-corrected using the N4ITK algorithm and then aligned to the volume plane. The N4ITK magnetic resonance bias correction procedure was performed using the N4 algorithm integrated in the 3D Slicer software platform (version 4.11.0, www.Slicer.org). The volume planes were oriented using the classic module in the 3D Slicer software. Given the wide distribution of intensity values ​​in MR images, z-score normalization was used to standardize the image intensity to achieve a normal distribution with a mean (μ) of 1 and a standard deviation (σ) of 0. The normalized value of the image intensity (x), also called the z-score, is calculated as follows:

[0042] In addition, irrelevant information was removed (e.g., text information, using the Otsu thresholding method), and the images were resampled (e.g., all magnetic resonance images were resampled to 1 × 1 × 1 mm³ voxels using the B-spline interpolation method).

[0043] Step 1.2: Feature extraction: Tumor ROI (Region of Interest) was delineated for each sequence in MRI enhancement, and 546 quantitative imaging features were extracted from the ROI of the six sequences of T2WI, DWI, NCP, AP, PVP, and DP. The feature set extracted for each sequence includes (i) 9 first-order features, (ii) 33 texture features, (iii) 168 logarithmic features, and (iv) 336 wavelet features. The first-order features in group (i) describe the intensity distribution within the ROI, including indicators such as entropy, variance, and uniformity. Group (ii) features capture the pattern or spatial distribution of intensity and are extracted through methods such as gray-level co-occurrence matrix (GLCM), gray-level dependence matrix (GLDM), gray-level run length matrix (GLRLM), gray-level size zone matrix (GLSZM), and neighborhood gray-level difference matrix (NGTDM). Groups (iii) and (iv) projected the images into the frequency domain using logarithmic and wavelet transforms to characterize tumor texture, focusing on different frequency scales and directions within the tumor volume. A total of 3276 radiomic features were extracted for each patient.

[0044] In a preferred solution, after feature extraction, the process further includes: using the Pearson correlation coefficient to calculate the correlation between features, removing highly correlated features (with an absolute value of the correlation coefficient greater than 0.95), and finally screening out representative features for subsequent analysis.

[0045] Step 2: Use the SNF algorithm to fuse the radiomic features of several liver cancer patients to obtain a fusion network;

[0046] Considering the differences between first-order features and texture features, which can be regarded as different data dimensions, the present invention applies SNF to reveal unique imaging heterogeneity phenotypes. This analysis was performed using the R package “SNFtool” according to established protocols and guidelines. The median absolute deviation (MAD) is a measure of the variability of univariate quantitative data, similar to the standard deviation, but more robust to external factors. MAD is used to assess the variability of radiomic features between samples, and features with MAD>1 are classified as highly variable. The top 457 highly variable features were screened from the above 3276 radiomic features for subsequent cluster analysis. Subsequently, the square Euclidean distances of the normalized first-order and texture feature matrices were calculated using the “dist2” function in the R package “SNFtools”. Based on the pairwise square Euclidean distance matrix, a similarity fusion network was established, and the specific parameter settings included K=20, sigma=0.5, and T=20.

[0047] Step 3: Clustering is performed using the spectral clustering algorithm based on the fusion network to classify liver cancer patients into different imaging types; the details are as follows:

[0048] The “Spectrum” function in the R package Spectrum was used to identify the three clusters as the optimal clusters of the similarity network. Finally, tumor clustering was performed using the “spectralClustering” function in the R package SNFtools. The imaging genomics features of the three tumor types on magnetic resonance imaging are shown in Figure 2. Figure 1 As shown in the figure, the imaging features of tumors with high heterogeneity are the most mixed, while the imaging features of tumors with low heterogeneity are more uniform.

[0049] Furthermore, the prognostic results are statistically analyzed for the data after typing; Figure 2 As shown in Figure A, there were significant differences in the prognostic results of the three imaging classifications in the training set. Patients classified as high heterogeneity had a significantly lower overall survival (OS) rate compared with patients with low heterogeneity (P = 0.002) and moderate heterogeneity (P = 0.001). Figure 2 As shown in Figure B, patients with high heterogeneity had a lower recurrence-free survival (RFS) compared with those with low heterogeneity (P = 0.023) and moderate heterogeneity (P = 0.029). Multivariate Cox proportional hazards model identified high heterogeneity as an independent adverse prognostic factor for RFS [hazard ratio (HR) = 1.89, 95% CI: 1.07 to 3.34, P = 0.028; Table 1]. A similar association was also observed for OS (HR = 1.62, 95% CI: 1.09 to 2.42, P = 0.017; Table 1).

[0050] Table 1. Multivariate Cox proportional hazards model for RFS and OS in the training set

[0051]

[0052] The present invention conducted multi-omics analysis on the three typing results and conducted in-depth research combining genetic differences with tumor microenvironment characteristics. Figure 3 As shown in Figure A, in the gene difference analysis between high heterogeneity tumors and low heterogeneity tumors, it was found that there were significant differences in gene expression in different imaging phenotypes. Figure 3 B and Figure 3 As shown in C, GSEA analysis showed that multiple extracellular matrix (ECM)-related pathways were significantly enriched in highly heterogeneous tumors, including ECM-receptor interactions, focal adhesion, and cell adhesion molecules. Further analysis of the ECM-related gene set, such as Figure 3 D and Figure 3 As shown in Figure E, the ECM score of highly heterogeneous tumors was significantly increased, and the expression of ECM-related genes was also significantly upregulated, which further verified the above results. Figure 4 Further analysis revealed significant differences in cellular interactions and immune microenvironments among heterogeneous tumors. Figure 4 As shown in Figure A, the cancer-associated fibroblasts (CAFs) scores of highly heterogeneous tumors were significantly increased, reflecting that their tumor microenvironment had more complex immune and matrix interaction characteristics. Figure 4 B and Figure 4 In the tSNE diagram shown in Figure C, high heterogeneity tumors showed a higher abundance of CAFs. In addition, based on pathway enrichment analysis, such as Figure 4 As shown in Figure D, classical tumor signaling pathways including MAPK signaling pathway, Notch signaling pathway and P53 signaling pathway were significantly activated in highly heterogeneous tumors, accompanied by enrichment of cell cycle, apoptosis and focal adhesion pathways, indicating stronger growth and invasion capabilities. Further analysis of differential gene expression and pathway interaction characteristics showed that, Figure 5 A and Figure 5 As shown in Figure 2B, the intensity of immune cell interactions in high-heterogeneity tumors is significantly higher than that in low-heterogeneity tumors, which is consistent with the highly active immune function in their microenvironment.

[0053] In summary, Figure 3 , Figure 4 and Figure 5 Together, they revealed significant features of highly heterogeneous tumors in terms of genetic differences, extracellular matrix function, and immune microenvironment, further supporting the molecular basis of their invasiveness and complexity.

[0054] Method validation and clinical application

[0055] The prognostic value of the heterogeneous phenotype of this method was externally validated by other multicenter cohorts (validation set 1, validation set 2, and validation set 3) using the same method as in the previous example. Figure 2 C in Figure 2 D in Figure 2 E and Figure 2 As shown in F, patients with high heterogeneity phenotype showed worse OS and RFS compared with patients with low heterogeneity. These multicenter results indicate that imaging classification effectively reflects the prognosis of patients and can serve as a valuable prognostic stratification tool in liver cancer.

[0056] In addition, the present invention also proposes the clinical application of imaging phenotypes: for newly diagnosed liver cancer patients, after obtaining multi-sequence magnetic resonance images, the image standardization preprocessing in step 1 of the above embodiment method is first performed, and the 457 highly variable features in step 2 are extracted. Then, the obtained features are projected into a coordinate system consistent with the latent space of the previous training samples through principal component analysis (PCA) to obtain the final vector representation for the classification task. Then, based on the distance-weighted k-nearest neighbor algorithm (kNN), we find the k training samples with the closest Euclidean distance to the sample. , and based on the weighted voting results of the categories to which these k samples belong, the patient's liver cancer imaging classification results are predicted Specifically, the process of predicting liver cancer imaging typing by distance-weighted kNN can be described as:

[0057]

[0058] in, Represents a category among M categories; Indicates the i-th sample to distance; , is a signal function used to assist in the calculation The probability of belonging to each category.

[0059] After obtaining the imaging classification of newly diagnosed liver cancer patients, personalized precision diagnosis and treatment can be performed. According to the above studies, high heterogeneity tumors are accompanied by worse prognosis, more aggressive tumor biological behavior and ECM remodeling, while low heterogeneity tumors have a better prognosis and show a trend of tumor metabolic reprogramming. Therefore, after identifying tumors with high imaging heterogeneity classification, follow-up monitoring of these patients can be strengthened, and targeted anti-ECM drugs can be added to conventional treatment options, which is expected to improve the prognosis of patients with high heterogeneity liver cancer.

[0060] It can be seen that the method of the present invention can classify liver cancer patients into different imaging types. Different imaging types can be used to study the intratumor heterogeneity of liver cancer; construct a liver cancer classification system and a liver cancer prognosis diagnosis system to further guide individualized precision treatment.

[0061] Specifically, a liver cancer prognosis diagnosis system of the present invention comprises:

[0062] The data acquisition module is used to collect multi-sequence magnetic resonance imaging data of liver cancer patients and extract radiomics features;

[0063] The classification module is used to determine the imaging classification of liver cancer patients;

[0064] The prognostic module determines the patient's prognosis based on the typing results.

[0065] Corresponding to the aforementioned embodiment of a method for non-invasive classification of liver cancer intratumor heterogeneity based on multi-omics imaging omics, the present invention also provides an embodiment of an electronic device.

[0066] An electronic device provided by an embodiment of the present invention includes one or more processors, which are used to implement a non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics in the above embodiment.

[0067] The electronic device of the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer.

[0068] The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capability in which it is located to read the corresponding computer program instructions in the non-volatile memory into the memory for execution. From the hardware level, the electronic device of the present invention includes a processor, a memory, a network interface, and a non-volatile memory. In addition, any device with data processing capability in which the device in the embodiment is located can also include other hardware according to the actual function of the device with data processing capability, which will not be described in detail.

[0069] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0070] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0071] An embodiment of the present invention also provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics in the above embodiment is implemented.

[0072] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be any device with data processing capability, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A non-invasive classification method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics, characterized in that: include: Collect multi-sequence magnetic resonance imaging data of several patients with liver cancer and extract imaging omics features respectively; The SNF algorithm was used to fuse the radiomic features of several liver cancer patients to obtain a fusion network; Based on the fusion network, spectral clustering algorithm was used to cluster liver cancer patients into different imaging types.

2. The method according to claim 1, characterized in that Multi-sequence MRI data include: T2WI, DWI, NCP, AP, PVP and DP.

3. The method according to claim 1, characterized in that The radiomics features include: Several first-order features, several texture features, several logarithmic features, and several wavelet features.

4. The method according to claim 1, characterized in that: After extracting radiomics features, it also includes: The Pearson correlation coefficient was used to calculate the correlation between features and remove highly correlated features.

5. The method according to claim 1, characterized in that The parameter settings of the SNF algorithm include: K value = 20, sigma value = 0.5, and T value = 20.

6. The method according to claim 1, characterized in that The spectral clustering algorithm is a spectral clustering algorithm based on similarity network fusion.

7. An application of the typing results obtained by the non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics according to any one of claims 1 to 6, characterized in that: include: Study the intratumor heterogeneity of liver cancer; Construct a liver cancer classification system; Construct a liver cancer prognosis diagnosis system.

8. A liver cancer prognosis diagnosis system, characterized in that: include: The data acquisition module is used to collect multi-sequence magnetic resonance imaging data of liver cancer patients and extract radiomics features; A classification module is used to determine the imaging classification of liver cancer patients; it includes: extracting imaging omics features based on multi-sequence magnetic resonance imaging data of liver cancer patients; reducing the dimension of the obtained features by principal component analysis and projecting them into a coordinate system consistent with the latent space of the previous training samples to obtain a vector representation for the final classification task; using a distance-weighted k-nearest neighbor algorithm to find the k training samples with the closest Euclidean distance to the vector representation, and predicting the imaging classification of liver cancer patients based on weighted voting results of the categories to which the k training samples belong; The prognostic module determines the patient's prognosis based on the typing results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics as described in any one of claims 1 to 6 is implemented.

10. A storage medium comprising computer executable instructions, which, when executed by a computer processor, implement the non-invasive typing method for intratumor heterogeneity of liver cancer based on multi-omics imaging omics as described in any one of claims 1 to 6.

Images