Digestive tract tumor HER2 expression level prediction method and device, electronic equipment and storage medium

Abstract

The invention discloses a digestive tract tumor HER2 expression level prediction method and device, electronic equipment and a storage medium, and relates to the technical field of medical detection. HER2 detection can be quickly and accurately realized through medical imaging information and a digestive tract tumor HER2 expression level prediction model. And the kit is especially suitable for preoperative HER2 detection or HER2 detection before administration, and is helpful for providing a more effective strategy for a treatment mode. Detection of HER2 positive patients may gain survival benefits. Therefore, the digestive tract tumor HER2 expression level prediction method provided by the invention has an important application value, and the digestive tract tumor HER2 expression level prediction method provided by the invention has universality and can be used for HER2 detection of people with various digestive tract tumors.

Description

Technical Field

[0001] The present invention relates to the field of medical detection technology, and in particular to a method, device, electronic device and storage medium for predicting HER2 expression levels in digestive tract tumors. Background Art

[0002] At present, the evolution of gastric cancer heterogeneity prediction technology has moved from single-omics to multimodal fusion, but it still faces challenges in data integration, standardization and clinical practicality. Traditional histopathological analysis uses HE staining, immunohistochemistry (IHC), etc. to observe tumor morphology, degree of differentiation and protein expression (such as HER2, Ki-67) to evaluate heterogeneity. This method relies on subjective interpretation and has poor reproducibility; it only reflects local regional characteristics and ignores spatial heterogeneity within the tumor. Methods using genomic technology mainly use gene sequencing (NGS) and fluorescence in situ hybridization (FISH) to detect driver mutations (such as TP53, CDH1) and copy number variations (such as HER2 amplification) to evaluate tumor heterogeneity. This method is costly and time-sensitive; it is difficult to analyze the spatiotemporal dynamics of clonal evolution. Based on traditional imaging feature analysis methods, there are problems such as manual features relying on expert experience and difficulty in cross-center data standardization.

[0003] Clinically, direct pathological testing of biopsy specimens can lead to inaccurate HER2 detection. Although postoperative pathological testing is often considered the gold standard for HER2 testing, preoperative or pre-treatment HER2 testing is still subject to inaccurate testing and a lack of standardized testing methods due to the heterogeneity of gastric cancer.

[0004] In view of this, the present invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, device, electronic device and storage medium for predicting the HER2 expression level of gastrointestinal tumors to solve the above technical problems.

[0006] The present invention is achieved in that:

[0007] In a first aspect, the present invention provides a method for predicting HER2 expression levels in digestive tract tumors, comprising:

[0008] Obtaining medical images of the patient to be tested, inputting the medical images of the patient to be tested into a digestive tract tumor HER2 expression level prediction model, and obtaining a digestive tract tumor HER2 expression level prediction result output by the digestive tract tumor HER2 expression level prediction model;

[0009] The prediction model for HER2 expression levels in digestive tract tumors is trained based on medical images of patients with digestive tract tumors and corresponding sample label information;

[0010] The HER2 expression level prediction model for digestive tract tumors includes: a feature extraction layer, a feature selection layer, and an output layer; the medical images of the patient to be tested are input into the feature extraction layer for feature extraction, and the feature extraction layer outputs radiomic features; the radiomic features are input into the feature selection layer, and the radiomic features that differ between HER2-positive and HER2-negative groups are screened using feature selection rules; the radiomic features that differ between HER2-positive and HER2-negative groups are input into the output layer for radiomic scoring, and the output layer outputs the predicted results of the HER2 expression level of the digestive tract tumor of the tested patient;

[0011] A prediction model for HER2 expression levels in gastrointestinal tumors was constructed using the following method: The radiomics features that identified differences between HER2-positive and HER2-negative gastric cancer patients were imported into a LASSO regression model with ten-fold cross-validation. The ten-fold cross-validation method was used to identify the optimal model parameter λ with the smallest deviation in the LASSO regression model, retaining independent variable features with non-zero regression coefficients. Based on the weight coefficients from the logistic regression analysis of the final screened radiomics features, a radiomics scoring formula was used to construct a model to obtain a specific score prediction model.

[0012] The radiomics scoring formula is as follows:

[0013] Rad-score = β0 + β1x1 + β2x2 + … + βnxn, where β0 is a constant term, βi is the logistic regression coefficient of the i-th radiomics feature xi, and xi is the i-th radiomics feature.

[0014] In a second aspect, the present invention further provides a device for predicting HER2 expression levels in digestive tract tumors, comprising: an acquisition module and a prediction module;

[0015] The acquisition module is used to obtain medical images of the patient to be tested;

[0016] A prediction module is used to input the medical image of the patient to be tested into the digestive tract tumor HER2 expression level prediction model to obtain the digestive tract tumor HER2 expression level prediction result output by the digestive tract tumor HER2 expression level prediction model;

[0017] The prediction model for HER2 expression levels in digestive tract tumors is trained based on medical images of patients with digestive tract tumors and corresponding sample label information;

[0018] The HER2 expression level prediction model for digestive tract tumors includes: a feature extraction layer, a feature selection layer, and an output layer; the medical images of the patient to be tested are input into the feature extraction layer for feature extraction, and the feature extraction layer outputs radiomic features; the radiomic features are input into the feature selection layer, and the radiomic features that differ between HER2-positive and HER2-negative groups are screened using feature selection rules; the radiomic features that differ between HER2-positive and HER2-negative groups are input into the output layer for radiomic scoring, and the output layer outputs the predicted results of the HER2 expression level of the digestive tract tumor of the tested patient;

[0019] A prediction model for HER2 expression levels in gastrointestinal tumors was constructed using the following method: The radiomics features that identified differences between HER2-positive and HER2-negative gastric cancer patients were imported into a LASSO regression model with ten-fold cross-validation. The ten-fold cross-validation method was used to identify the optimal model parameter λ with the smallest deviation in the LASSO regression model, retaining independent variable features with non-zero regression coefficients. Based on the weight coefficients from the logistic regression analysis of the final screened radiomics features, a radiomics scoring formula was used to construct a model to obtain a specific score prediction model.

[0020] The radiomics scoring formula is as follows:

[0021] Rad-score = β0 + β1x1 + β2x2 + … + βnxn, where β0 is a constant term, βi is the logistic regression coefficient of the i-th radiomics feature xi, and xi is the i-th radiomics feature.

[0022] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting the HER2 expression level of gastrointestinal tumors when executing the program.

[0023] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for predicting the HER2 expression level of gastrointestinal tumors.

[0024] The present invention has the following beneficial effects:

[0025] The present invention provides a method for predicting HER2 expression levels in gastrointestinal tumors. By leveraging medical imaging information and a prediction model for HER2 expression levels in gastrointestinal tumors, HER2 detection can be achieved quickly and accurately. This method is particularly suitable for preoperative HER2 detection or pre-drug administration, helping to provide more effective strategies for treatment. Patients who test HER2 positive may benefit from survival. HER2 status is a key basis for selecting targeted therapy. Therefore, the method for predicting HER2 expression levels in gastrointestinal tumors provided by the present invention has important application value and is universally applicable, and can be used for HER2 detection in a variety of gastrointestinal tumor populations. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 This is the ROC curve of the training set;

[0028] Figure 2 This is the ROC curve of the validation set;

[0029] Figure 3 LASSO dimensionality reduction result Figure 1 ;

[0030] Figure 4 LASSO dimensionality reduction result Figure 2 ;

[0031] Figure 5 Flowchart for image segmentation and feature extraction;

[0032] Figure 6 Flowchart for feature selection;

[0033] Figure 7 This is the ROC curve diagram of Example 2. DETAILED DESCRIPTION

[0034] Reference will now be made in detail to embodiments of the present invention, one or more examples of which are described below. Each example is provided to illustrate, not to limit, the present invention. Indeed, it will be apparent to those skilled in the art that various modifications and variations may be made to the present invention without departing from the scope or spirit of the invention. For example, features illustrated or described as part of one embodiment may be used in another embodiment to produce further embodiments.

[0035] Glossary:

[0036] wavelet: wavelet.

[0037] LLL: low-low-low mode.

[0038] LLH: Low-Low-High mode.

[0039] HLH: High-Low-High mode.

[0040] HHL: High-High-Low mode.

[0041] ZoneEntropy: Regional entropy, the entropy value calculated based on the GLSZM matrix, reflects the uncertainty or heterogeneity of the size distribution of connected regions.

[0042] DependenceNonUniformityNormalized: Normalized dependency non-uniformity, indicating a standardized non-uniform correlation index.

[0043] Rad-score: radiomics score.

[0044] In a first aspect, the present invention provides a method for predicting HER2 expression levels in digestive tract tumors, comprising:

[0045] Obtaining medical images of the patient to be tested, inputting the medical images of the patient to be tested into a digestive tract tumor HER2 expression level prediction model, and obtaining a digestive tract tumor HER2 expression level prediction result output by the digestive tract tumor HER2 expression level prediction model;

[0046] The prediction model for HER2 expression levels in digestive tract tumors is trained based on medical images of patients with digestive tract tumors and corresponding sample label information;

[0047] The HER2 expression level prediction model for digestive tract tumors includes: a feature extraction layer, a feature selection layer, and an output layer; the medical images of the patient to be tested are input into the feature extraction layer for feature extraction, and the feature extraction layer outputs radiomic features; the radiomic features are input into the feature selection layer, and the radiomic features that differ between HER2-positive and HER2-negative groups are screened using feature selection rules; the radiomic features that differ between HER2-positive and HER2-negative groups are input into the output layer for radiomic scoring, and the output layer outputs the predicted results of the HER2 expression level of the digestive tract tumor of the tested patient;

[0048] A prediction model for HER2 expression levels in gastrointestinal tumors was constructed using the following method: The radiomics features that identified differences between HER2-positive and HER2-negative gastric cancer patients were imported into a LASSO regression model with ten-fold cross-validation. The ten-fold cross-validation method was used to identify the optimal model parameter λ with the smallest deviation in the LASSO regression model, retaining independent variable features with non-zero regression coefficients. Based on the weight coefficients from the logistic regression analysis of the final screened radiomics features, a radiomics scoring formula was used to construct a model to obtain a specific score prediction model.

[0049] The radiomics scoring formula is as follows:

[0050] Rad-score = β0 + β1x1 + β2x2 + … + βnxn, where β0 is a constant term, βi is the logistic regression coefficient of the i-th radiomics feature xi, and xi is the i-th radiomics feature.

[0051] By combining medical imaging information with a model predicting HER2 expression levels in gastrointestinal tumors, HER2 testing can be performed quickly and accurately. This method is particularly suitable for preoperative HER2 testing or pre-drug administration, thereby helping to provide effective strategies for treatment. Patients who test positive for HER2 may experience a survival benefit. HER2 status is a key factor in selecting targeted therapies. Therefore, the method for predicting HER2 expression levels in gastrointestinal tumors provided by the present invention has important application value and is universally applicable, allowing for HER2 testing in a variety of gastrointestinal tumor populations.

[0052] In a preferred embodiment of the present invention, the feature extraction layer is used to extract at least three of the following features:

[0053] Firstorder (first-order features), glcm (co-occurrence matrix), gldm (grayscale dependency matrix features), grllm (run length), glszm (region) and ngtdm (neighborhood difference).

[0054] Preferably, when extracting radiomics features, the CT values ​​of the ROI region are discretized based on a fixed bin width (bin width = 25 HU) to ensure the repeatability of feature extraction.

[0055] The digestive tract tumor is selected from gastric cancer, colon cancer, rectal cancer, colorectal cancer, gallbladder cancer, liver cancer, bile duct cancer, small intestine cancer or cardiac tumor.

[0056] In a preferred embodiment of the present invention, the radiomics scoring formula is as follows: Rad-score = -2.1348 + (-0.4068 × wavelet-LLL glszm ZoneEntropy) + (-0.3931 × wavelet-LLH glszmZoneEntropy) + (0.5955 × wavelet-HLH ngtdm Strength) + (-0.6793 × wavelet-HHL gldmDependenceNonUniformityNormalized);

[0057] In a preferred embodiment of the present invention, the preset threshold of Rad-score is 0.155;

[0058] In a preferred embodiment of the present invention, if the radiomics score (Rad-score) is greater than a threshold, the patient is predicted to have a HER2-positive digestive tract tumor; if it is lower than the threshold, the patient is predicted to have a HER2-negative digestive tract tumor.

[0059] In a preferred embodiment of the present invention, the feature selection rules in the feature selection layer are as follows:

[0060] Statistical analysis of imaging features was performed using statistical software. A repeatability test was performed on the same imaging image of the same patient. Features with a repeatability greater than 0.8 were selected for analysis and preprocessed with z-score standardization. Statistically significant imaging features between the HER2-positive and HER2-negative groups were selected.

[0061] In a preferred embodiment of the present invention, the statistical software is selected from R software, SAS, SPASS or Python;

[0062] In a preferred embodiment of the present invention, the feature selection rules in the feature selection layer screen out features with statistically significant differences between the HER2-positive and HER2-negative groups from the retained features through Mann-Whitney U test and t-test.

[0063] In a preferred embodiment of the present invention, obtaining medical images of the patient to be tested includes medical image acquisition and image segmentation; image segmentation includes:

[0064] Using ITK-SNAP version 3.8.0 software, the primary tumor was delineated in both the arterial and venous phases. The slice with the largest tumor area was selected, and a ROI was constructed along the tumor tissue contour to cover the entire lesion area. Image normalization was then performed.

[0065] In a preferred embodiment of the present invention, image normalization is to reduce the variability between different images using a global scale:

[0066] In a preferred embodiment of the present invention, the image normalization process is: harmonizing the signal intensity and reducing the unit variance by eliminating the mean; and normalizing each feature according to the mean value and standard deviation of each feature in the training set queue by z normalization.

[0067] Image standardization is used to eliminate signal differences between different CT and other imaging images due to factors such as different medical testing centers, equipment, and magnification.

[0068] In a preferred embodiment of the present invention, after image segmentation, image conversion is also included; the image conversion includes: obtaining an image based on Laplace-Gaussian (LoG) transform and a wavelet decomposition image based on 8 filter channels, and then inputting the image based on Laplace-Gaussian (LoG) transform and the wavelet decomposition image into a feature extraction layer for feature extraction.

[0069] In a preferred embodiment of the present invention, basic feature extraction is used to obtain basic radiomics features of patient imaging, including 18 firstorder, 24 glcm, 14 gldm, 16 grllm, 16 glszm, and 5 ngtdm features. Furthermore, image transformation can be used to obtain more radiomics features.

[0070] The sigma parameters of the Laplace-Gaussian (LoG) transform are selected as 1.0 mm, 2.0 mm and 3.0 mm.

[0071] In a preferred embodiment of the present invention, the following features of the image after image segmentation are extracted at the feature extraction layer: Firstorder, glcm, gldm, grllm, glszm and ngtdm; the following features of the image after Laplace-Gaussian (LoG) transformation are extracted at the feature extraction layer: Firstorder, glcm (co-occurrence matrix), gldm (grayscale dependence matrix feature), grllm (run length), glszm (region) and ngtdm (neighborhood difference); the following features of the wavelet decomposition image are extracted at the feature extraction layer: Firstorder, glcm (co-occurrence matrix), gldm (grayscale dependence matrix feature), grllm (run length), glszm (region) and ngtdm (neighborhood difference);

[0072] In a preferred embodiment of the present invention, the following number of basic features are extracted from the image after image segmentation at the feature extraction layer: 18 Firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm and 5 ngtdm; the following number of features are extracted from the image after Laplace-of-Gaussian (LoG) transformation at the feature extraction layer: 18 Firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm and 5 ngtdm; the following number of features are extracted from the wavelet decomposition image at the feature extraction layer: 18 Firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm and 5 ngtdm.

[0073] In a preferred embodiment of the present invention, 279 features of the image after Laplace-of-Gaussian (LoG) transformation are extracted at the feature extraction layer, and 744 features of the wavelet decomposition image are extracted at the feature extraction layer.

[0074] In other embodiments, those skilled in the art may also use the Pyradiomics computing language included in the 3D Slicer software (version 5.6.1) to select features that need to be automatically extracted as needed.

[0075] In a second aspect, the present invention further provides a device for predicting HER2 expression levels in digestive tract tumors, comprising: an acquisition module and a prediction module;

[0076] The acquisition module is used to obtain medical images of the patient to be tested;

[0077] A prediction module is used to input the medical image of the patient to be tested into the digestive tract tumor HER2 expression level prediction model to obtain the digestive tract tumor HER2 expression level prediction result output by the digestive tract tumor HER2 expression level prediction model;

[0078] The prediction model for HER2 expression levels in digestive tract tumors is trained based on medical images of patients with digestive tract tumors and corresponding sample label information;

[0079] The HER2 expression level prediction model for digestive tract tumors includes: a feature extraction layer, a feature selection layer, and an output layer; the medical images of the patient to be tested are input into the feature extraction layer for feature extraction, and the feature extraction layer outputs radiomic features; the radiomic features are input into the feature selection layer, and the radiomic features that differ between HER2-positive and HER2-negative groups are screened using feature selection rules; the radiomic features that differ between HER2-positive and HER2-negative groups are input into the output layer for radiomic scoring, and the output layer outputs the predicted results of the HER2 expression level of the digestive tract tumor of the tested patient;

[0080] A prediction model for HER2 expression levels in gastrointestinal tumors was constructed using the following method: The radiomics features that identified differences between HER2-positive and HER2-negative gastric cancer patients were imported into a LASSO regression model with ten-fold cross-validation. The ten-fold cross-validation method was used to identify the optimal model parameter λ with the smallest deviation in the LASSO regression model, retaining independent variable features with non-zero regression coefficients. Based on the weight coefficients from the logistic regression analysis of the final screened radiomics features, a radiomics scoring formula was used to construct a model to obtain a specific score prediction model.

[0081] The radiomics scoring formula is as follows:

[0082] Rad-score = β0 + β1x1 + β2x2 + … + βnxn, where β0 is a constant term, βi is the logistic regression coefficient of the i-th radiomics feature xi, and xi is the i-th radiomics feature.

[0083] In a preferred embodiment of the present invention, the radiomics scoring formula is as follows: Rad-score = -2.1348 + (-0.4068 × wavelet-LLL glszm ZoneEntropy) + (-0.3931 × wavelet-LLH glszmZoneEntropy) + (0.5955 × wavelet-HLH ngtdm Strength) + (-0.6793 × wavelet-HHL gldmDependenceNonUniformityNormalized);

[0084] In a preferred embodiment of the present invention, the preset threshold of Rad-score is 0.155;

[0085] In a preferred embodiment of the present invention, if the radiomics score (Rad-score) is greater than a threshold, the patient is predicted to have a HER2-positive digestive tract tumor; if it is lower than the threshold, the patient is predicted to have a HER2-negative digestive tract tumor.

[0086] When predicting the HER2 expression level in gastrointestinal tumors, it is only necessary to extract four significantly different imaging features from the imaging images of the patient to be tested, so as to accurately and quickly obtain HER2 negative or positive results, facilitating the accurate prediction of HER2 before surgery or treatment.

[0087] The digestive tract tumor HER2 expression level prediction model provided by the present invention has an AUC value of up to 0.76 in the validation set, showing good predictive performance.

[0088] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting the HER2 expression level of gastrointestinal tumors when executing the program.

[0089] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for predicting the HER2 expression level of gastrointestinal tumors.

[0090] Specifically, the electronic device may include a memory, a processor, a bus, and a communication interface, wherein the memory, processor, and communication interface are electrically connected to each other directly or indirectly to enable data transmission or interaction. For example, these components may be electrically connected to each other via one or more buses or signal lines. The processor may process information and / or data related to target identification to perform one or more functions described in this application.

[0091] The memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0092] A processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU) or a network processor (NP). It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0093] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely below. Where specific conditions are not specified in the embodiments, conventional conditions or conditions recommended by the manufacturer are used. Where the manufacturer of the reagents or instruments is not specified, all are conventional products that can be purchased commercially.

[0094] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0095] Example 1

[0096] This embodiment provides a method for predicting HER2 expression levels in gastric cancer, which includes image acquisition (i.e., obtaining medical images of the patient to be tested), image segmentation, radiomics feature extraction, radiomics feature selection, and construction and verification of a prediction model for HER2 expression levels in gastric cancer.

[0097] (1) Image acquisition: A total of 656 gastric cancer patients who underwent radical gastrectomy and HER2 testing at the Fourth Hospital of Hebei Medical University between 2016 and 2023 were included, including 96 HER2-positive patients and 560 HER2-negative patients. Inclusion criteria: 1. Pathologically confirmed gastric adenocarcinoma, 2. Postoperative HER2 testing, 3. High-quality abdominal enhanced CT images. Exclusion criteria: 1. Preoperative neoadjuvant therapy, 2. Concomitant other malignancies.

[0098] (2) Image segmentation: Senior radiologists (6 years of experience in imaging diagnosis) used ITK-SNAP version 3.8.0 software to delineate the primary tumor in the arterial and venous phases, select the slice with the largest tumor area, and construct the ROI along the contour of the tumor tissue to cover the entire lesion area, avoiding some unnecessary structures (such as blood vessels, bile ducts, and healthy parenchyma) to avoid changing the heterogeneity analysis.

[0099] A post-reconstruction batch coordination method (i.e., image normalization) is used to reduce the variability between centers using a global scale, that is, to coordinate the signal intensity and reduce the unit variance by eliminating the mean; z-normalization normalizes each feature according to the mean and standard deviation of each feature in the training set cohort by providing some comparable results.

[0100] The segmented image was subjected to Laplace-of-Gaussian (LoG) transform, and a wavelet decomposition image based on 8 filtering channels was obtained.

[0101] (3) Feature extraction: 3D Slicer software (version 5.6.1) equipped with the Pyradiomics computing language was used to automatically extract a total of 1116 radiomic features in the venous phase. During feature extraction, the CT values ​​of the ROI region were discretized based on a fixed bin width (bin width = 25HU), including 18 basic features: firstorder, 24 glcm, 14 gldm, 16 grllm, 16 glszm, and 5 ngtdm. A total of 744 (93*8) features were extracted from the wavelet decomposition image based on 8 filter channels, and 279 (93*3) features were extracted from the Laplace-Gaussian (LoG) transform image (sigma parameters were selected as 1.0mm, 2.0mm, and 3.0mm), for a total of 1116 radiomic features. Flowchart: Figure 5 shown.

[0102] (4) Feature selection: Statistical analysis was performed using R software (version 4.3.1, http: / / www.r-project.org). The ICC test was performed to select features with intergroup reproducibility greater than 0.8 for inclusion in the analysis, and z-score standardization preprocessing was performed. The Mann-Whitney U test and t-test were used to screen out the features with statistically significant differences (P < 0.05) between the HER2-positive and HER2-negative groups from the retained features. The flowchart is shown in Figure 2. Figure 6 shown.

[0103] (5) Model construction

[0104] It was imported into the LASSO regression model with ten-fold cross-validation, and the ten-fold cross-validation method was used to screen the optimal model parameter λ with the smallest deviation of the LASSO regression model, and the independent variable features with non-zero regression coefficients were retained; according to the weight coefficients of the logistic regression analysis of the final screened imaging omics features, the imaging omics scoring formula (Rad-score = β0 + β1x1 + β2x2 + ... + βnxn, where β0 is the constant term and βi is the logistic regression coefficient of the i-th imaging omics feature xi) was used to construct a model to obtain a specific scoring prediction model.

[0105] The specific formula is Rad-score=-2.1348+(-0.4068×wavelet-LLL glszm ZoneEntropy)+(-0.3931×wavelet-LLH glszm ZoneEntropy)+(0.5955×wavelet-HLH ngtdmStrength)+(-0.6793×wavelet-HHL gldm DependenceNonUniformityNormalized). The preset threshold of Rad-score in the present invention is 0.155.

[0106] If the radiomics score (Rad-score) is greater than the threshold, the patient is predicted to have HER2-positive gastric cancer; if it is lower than the threshold, the patient is predicted to have HER2-negative gastric cancer.

[0107] ROC curve reference of training set Figure 1 As shown, the ROC curve of the validation set is referenced Figure 2 shown.

[0108] Training set model performance

[0109]

[0110] Validation set model performance

[0111]

[0112] The above results show that the gastric cancer HER2 expression level prediction model provided by the present invention has high performance and is reliable.

[0113] Figure 3 The results of LASSO dimensionality reduction analysis show that the optimal penalty coefficient lambda (λ) in the LASSO model for pathological omics feature screening was determined using ten-fold cross-validation.

[0114] Figure 4 To avoid overfitting, the present invention uses a 1 standard error criterion to determine the number of selected features.

[0115] Example 2

[0116] This embodiment provides a method for predicting HER2 expression levels in colorectal cancer, which includes image acquisition (i.e., obtaining medical images of the patient to be tested), image segmentation, radiomics feature extraction, and construction and verification of a prediction model for HER2 expression levels in colorectal cancer.

[0117] (1) Image acquisition: A total of 91 patients with colorectal cancer who underwent radical surgery for colorectal cancer and underwent HER2 testing at the Fourth Hospital of Hebei Medical University between 2016 and 2023 were included, including 79 HER2-positive patients and 12 HER2-negative patients. Inclusion criteria: 1. Pathologically confirmed colorectal adenocarcinoma, 2. Postoperative HER2 testing, 3. High-quality abdominal enhanced CT images. Exclusion criteria: 1. Preoperative neoadjuvant therapy, 2. Concomitant other malignancies.

[0118] (2) Image segmentation: Senior radiologists (6 years of imaging diagnostic experience) used ITK-SNAP version 3.8.0 software to delineate the primary tumor in the venous phase and select the slice with the largest tumor area. ROIs were constructed along the contour of the tumor tissue to cover the entire lesion area, avoiding some unnecessary structures (such as blood vessels, bile ducts, and healthy parenchyma) to avoid changing the heterogeneity analysis.

[0119] A post-reconstruction batch coordination method (i.e., image normalization) is used to reduce the variability between centers using a global scale, that is, to coordinate the signal intensity and reduce the unit variance by eliminating the mean; z-normalization normalizes each feature according to the mean and standard deviation of each feature in the training set cohort by providing some comparable results.

[0120] The segmented image was subjected to Laplace-of-Gaussian (LoG) transform, and a wavelet decomposition image based on 8 filtering channels was obtained.

[0121] (3) Feature extraction: 3D Slicer software (version 5.6.1) equipped with the Pyradiomics computing language was used to automatically extract a total of 1116 radiomic features in the venous phase. During feature extraction, the CT values ​​of the ROI region were discretized based on a fixed bin width (bin width = 25HU), including 18 basic features: firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm, and 5 ngtdm. A total of 744 (93*8) features were extracted from the wavelet decomposition image based on 8 filter channels, and a total of 279 (93*3) features were extracted based on the Laplace-Gaussian (LoG) transform image (sigma parameters were selected as 1.0mm, 2.0mm, and 3.0mm), totaling 1116 radiomic features.

[0122] (4) Model verification: The same prediction model as the scoring prediction model in Example 1 was used for calculation. If the radiomics score (Rad-score) was greater than the threshold, the patient was predicted to have HER2-positive colorectal cancer; if it was lower than the threshold, the patient was predicted to have HER2-negative colorectal cancer.

[0123] The ROC curve of Example 2 is referenced Figure 7 shown.

[0124] The model prediction performance results are as follows:

[0125]

[0126]

[0127] The above results show that the colorectal cancer HER2 expression level prediction model provided by the present invention has high performance and is reliable.

[0128] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for predicting HER2 expression levels in digestive tract tumors, characterized in that: It includes: Obtaining a medical image of a patient to be tested, inputting the medical image of the patient to be tested into a digestive tract tumor HER2 expression level prediction model, and obtaining a digestive tract tumor HER2 expression level prediction result output by the digestive tract tumor HER2 expression level prediction model; The digestive tract tumor HER2 expression level prediction model is trained based on medical images of gastric cancer patients and corresponding sample label information; The digestive tract tumor HER2 expression level prediction model includes: feature extraction layer , feature selection layer and output layer; inputting the medical image of the patient to be tested into the feature extraction layer for feature extraction, and obtaining the radiomics features output by the feature extraction layer; inputting the radiomics features into the feature selection layer, and screening the radiomics features of the differences between the HER2-positive and HER2-negative groups through feature selection rules; inputting the radiomics features of the differences between the HER2-positive and HER2-negative groups into the output layer for radiomics scoring, and obtaining the prediction result of the HER2 expression level of the digestive tract tumor of the tested patient output by the output layer; The HER2 expression level prediction model for digestive tract tumors is constructed by the following method: the radiomics features of the differences between the screened HER2-positive gastric cancer patients and the HER2-negative gastric cancer patients are introduced into a LASSO regression model with ten-fold cross-validation, the ten-fold cross-validation method is used to screen the optimal model parameter λ with the smallest deviation of the LASSO regression model, and the independent variable features with regression coefficients not equal to 0 are retained; based on the weight coefficients of the logistic regression analysis of the final screened radiomics features, a model is constructed using the radiomics scoring formula to obtain a scoring prediction model; The radiomics scoring formula is as follows: Rad-score = β0 + β1x1 + β2x2 + … + βnxn, where β0 is a constant term, βi is the logistic regression coefficient of the i-th radiomics feature xi, and xi is the i-th radiomics feature.

2. The method for predicting HER2 expression level in digestive tract tumors according to claim 1, characterized in that: The feature extraction layer is used to extract at least three of the following features: Firstorder (first-order features), glcm (co-occurrence matrix), gldm (grayscale dependency matrix features), grllm (run length), glszm (region), and ngtdm (neighborhood difference); Preferably, the CT values ​​of the ROI region are discretized based on a fixed bin width (bin width = 25 HU) during radiomics feature extraction; Preferably, the digestive tract tumor is selected from gastric cancer, colon cancer, rectal cancer, colorectal cancer, gallbladder cancer, liver cancer, bile duct cancer, small intestine cancer or cardiac tumor.

3. The method for predicting HER2 expression level in digestive tract tumors according to claim 2, characterized in that: The radiomics scoring formula is as follows: Rad-score = -2.1348 + (-0.4068 × wavelet-LLL glszm ZoneEntropy)+(-0.3931×wavelet-LLH glszm ZoneEntropy)+(0.5955×wavelet-HLH ngtdm Strength)+(-0.6793×wavelet-HHL gldm DependenceNonUniformityNormalized); Preferably, the preset threshold of Rad-score is 0.155; Preferably, if the radiomics score (Rad-score) is greater than a threshold, the patient is predicted to have a HER2-positive digestive tract tumor; if it is lower than the threshold, the patient is predicted to have a HER2-negative digestive tract tumor.

4. The method for predicting HER2 expression level in digestive tract tumors according to claim 1, characterized in that: The feature selection rules in the feature selection layer are as follows: Statistical analysis of the imaging features was performed using statistical software. A repeatability test was performed on the same imaging image of the same patient. Features with a repeatability greater than 0.8 were selected for analysis and preprocessed with z-score standardization. The imaging features with statistically significant differences between the HER2-positive and HER2-negative groups were selected. Preferably, the statistical software is selected from R software, SAS, SPASS or Python; Preferably, the feature selection rules in the feature selection layer screen out features with statistically significant differences between HER2-positive and HER2-negative groups from the retained features through Mann-Whitney U test and t-test.

5. The method for predicting HER2 expression level in digestive tract tumors according to claim 1, characterized in that: The acquisition of medical images of the patient to be tested includes medical image acquisition and image segmentation; the image segmentation includes: Using ITK-SNAP version 3.8.0 software, the primary tumor was delineated in both the arterial and venous phases. The slice with the largest tumor area was selected, and a ROI was constructed along the tumor tissue contour to cover the entire lesion area. Image normalization was then performed. Preferably, the image normalization process uses a global scale to reduce the variability between different images: Preferably, the image normalization process is: harmonizing the signal intensity and reducing the unit variance by eliminating the mean; and z-normalizing each feature according to the mean value and standard deviation of each feature in the training set.

6. The method for predicting HER2 expression level in digestive tract tumors according to claim 5, characterized in that: After image segmentation, image conversion is also included; the image conversion includes: obtaining an image based on Laplace-Gaussian (LoG) transformation and a wavelet decomposition image based on 8 filter channels, and then inputting the image based on Laplace-Gaussian (LoG) transformation and the wavelet decomposition image into the feature extraction layer for feature extraction; Preferably, the following features of the image after image segmentation are extracted at the feature extraction layer: Firstorder, glcm, gldm, grllm, glszm and ngtdm; the following features of the image after Laplace-Gaussian (LoG) transformation are extracted at the feature extraction layer: Firstorder, glcm, gldm, grllm, glszm and ngtdm; the following features of the wavelet decomposition image are extracted at the feature extraction layer: glcm, gldm, grllm, glszm and ngtdm; Preferably, the following number of basic features of the image after image segmentation is extracted at the feature extraction layer: 18 Firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm and 5 ngtdm; the following number of features of the image after Laplace-Gaussian (LoG) transformation is extracted at the feature extraction layer: 18 Firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm and 5 ngtdm; the following number of features of the wavelet decomposition image is extracted at the feature extraction layer: 18 Firstorder, 24 glcm, 14 gldm, 16 glrlm, 16 glszm and 5 ngtdm; Preferably, a total of 279 features of the image after Laplace-of-Gaussian (LoG) transformation are extracted at the feature extraction layer, and a total of 744 features of the wavelet decomposition image are extracted at the feature extraction layer.

7. A device for predicting HER2 expression levels in digestive tract tumors, characterized in that: It includes: Acquisition module and prediction module; The acquisition module is used to obtain medical images of the patient to be tested; The prediction module is used to input the medical image of the patient to be tested into the digestive tract tumor HER2 expression level prediction model to obtain the digestive tract tumor HER2 expression level prediction result output by the digestive tract tumor HER2 expression level prediction model; The digestive tract tumor HER2 expression level prediction model is trained based on medical images of gastric cancer patients and corresponding sample label information; The digestive tract tumor HER2 expression level prediction model includes: a feature extraction layer, a feature selection layer, and an output layer; the medical image of the patient to be tested is input into the feature extraction layer for feature extraction, and the radiomics features output by the feature extraction layer are obtained; the radiomics features are input into the feature selection layer, and the radiomics features of the differences between HER2-positive and HER2-negative groups are screened out through feature selection rules; the radiomics features of the differences between HER2-positive and HER2-negative groups are input into the output layer for radiomics scoring, and the output layer outputs the prediction result of the HER2 expression level of the digestive tract tumor of the tested patient; The HER2 expression level prediction model for digestive tract tumors is constructed by the following method: the radiomics features of the differences between the screened HER2-positive gastric cancer patients and the HER2-negative gastric cancer patients are introduced into a LASSO regression model with ten-fold cross-validation, the ten-fold cross-validation method is used to screen the optimal model parameter λ with the smallest deviation of the LASSO regression model, and the independent variable features with regression coefficients not equal to 0 are retained; based on the weight coefficients of the logistic regression analysis of the final screened radiomics features, a model is constructed using the radiomics scoring formula to obtain a specific scoring prediction model; The radiomics scoring formula is as follows: Rad-score = β0 + β1x1 + β2x2 + … + βnxn, where β0 is a constant term, βi is the logistic regression coefficient of the i-th radiomics feature xi, and xi is the i-th radiomics feature.

8. The device for predicting HER2 expression level in gastric cancer according to claim 7, characterized in that: The radiomics scoring formula is as follows: Rad-score = -2.1348 + (-0.4068 × wavelet-LLL glszm ZoneEntropy)+(-0.3931×wavelet-LLH glszm ZoneEntropy)+(0.5955×wavelet-HLH ngtdm Strength)+(-0.6793×wavelet-HHL gldm DependenceNonUniformityNormalized); Preferably, the preset threshold of Rad-score is 0.155; Preferably, if the radiomics score (Rad-score) is greater than a threshold, the patient is predicted to have a HER2-positive digestive tract tumor; if it is lower than the threshold, the patient is predicted to have a HER2-negative digestive tract tumor.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for predicting the HER2 expression level of digestive tract tumors according to any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the HER2 expression level of a digestive tract tumor according to any one of claims 1 to 6 is implemented.

Images