A method, device, medium, and procedure for predicting the survival of patients with CRLM.

Abstract

This invention provides a method, system, device, medium, and program product for predicting the survival of patients with chronic renal ulcer disease (CRLM), and also discloses a method, system, device, medium, and program product for predicting the risk of RAS gene mutations in CRLM patients, relating to the field of intelligent healthcare. By inputting venous and arterial phase images into a classifier to calculate the subject's DL score at different stages, integrating the DL scores from different stages, and using the integrated results to predict survival. Simultaneously, the classifier based on the calculated DL scores assesses the RAS gene status, predicts the subject's survival rate and survival time based on the RAS gene status, identifies suitable targeted drugs for treatment, and identifies patients most likely to benefit from cetuximab or bevacizumab treatment.

Description

Technical Field

[0001] This invention relates to the field of intelligent healthcare, and more specifically, to a method, apparatus, medium, and program product for predicting the survival of CRLM patients. Background Technology

[0002] Colorectal cancer (CRC) is the third most common cancer worldwide and the second leading cause of cancer-related deaths. Previous studies have shown that approximately 25-30% of patients diagnosed with colorectal cancer develop liver metastases (LM) during the course of their disease, and LM significantly impacts the prognosis of CRC.

[0003] Standard treatment for unresectable colorectal cancer liver metastases (CRLM) usually involves chemotherapy. Different patients require different drug treatments. For individual patients, finding appropriate early diagnosis and treatment, as well as prognostic biomarkers, is very important for improving the survival of CRLM patients. Predicting the survival of CRLM patients after treatment through reasonable methods is even more crucial. Summary of the Invention

[0004] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention provides a method, device, medium, and program product for predicting the survival of CRLM patients; the method of this invention calculates the DL score of the subject at different stages by inputting venous phase images and arterial phase images into a classifier, integrates the DL scores of different stages, and realizes the function of predicting survival based on the integrated results.

[0005] The first aspect of this application discloses a method for predicting the survival of CRLM patients, the method comprising:

[0006] S101: Acquire arterial and venous phase images of the subject containing the target region;

[0007] S102: Input the arterial phase image into the classifier to calculate the arterial phase DL score, input the venous phase image into the classifier to calculate the venous phase DL score, and integrate the arterial phase DL score and the venous phase DL score to obtain the total DL score;

[0008] S103: Predict the survival of the subject based on the total DL score.

[0009] In some embodiments, the method further includes: obtaining the subject's CEA value and / or DCR value; predicting the subject's survival based on any one or more of the DL score, CEA value, and DCR value;

[0010] Optionally, when predicting the survival of a subject based on the DL score, CEA value, and DCR value, the subject's survival is predicted by calculating the total score of the DL score, CEA value, and DCR value.

[0011] In some embodiments, when calculating the total score based on the DL score, CEA value, and DCR value, the calculation method for the total score includes: Y = 2.995 × 10⁻⁵ × CEA + (-1.581 × DCR) + (-6.483 × 10⁻¹ × DL), where Y is the total score.

[0012] In some embodiments, between S101 and S102, the method further includes: segmenting an arterial phase ROI region from the arterial phase image and segmenting a venous phase ROI region from the venous phase image; calculating an arterial phase DL score based on the arterial phase ROI region and calculating a venous phase DL score based on the venous phase ROI region; and integrating the arterial phase DL score and the venous phase DL score to obtain a total DL score.

[0013] A second aspect of this application discloses a method for predicting the risk of RAS gene mutations in CRLM patients, the method comprising:

[0014] S201: Acquire CECT images of the subject; the CECT images include venous phase and arterial phase images;

[0015] S202: Input the CECT image into the classifier to obtain the predicted RAS gene mutation risk.

[0016] The method for constructing the classifier includes:

[0017] S2021: Obtain training set samples including images of the arterial phase (AP) and venous phase (VP) and the corresponding RAS gene status labels; the RAS gene status labels include mutant and wild-type;

[0018] S2022: Input the arterial phase image into a machine learning model for training to obtain a predicted classification result, compare the predicted classification result with the classification label, optimize the model based on the comparison result, and obtain an arterial phase classifier based on the arterial phase image; input the venous phase image into a machine learning model for training to obtain a predicted classification result, compare the predicted classification result with the classification label, optimize the model based on the comparison result, and obtain a venous phase classifier based on the venous phase image;

[0019] S2023: The classifier is obtained by merging the arterial phase classifier and the venous phase classifier using an ensemble classifier.

[0020] In some embodiments, the method further includes: predicting the survival rate of the subject based on the RAS gene mutation risk; if the RAS gene mutation risk is low, a high survival rate is obtained; if the RAS gene mutation risk is high, a low survival rate is obtained.

[0021] Optionally, the survival of the subject can be predicted based on the RAS gene mutation risk; if the RAS gene mutation risk is low, the subject will have a long survival; if the RAS gene mutation risk is high, the subject will have a short survival.

[0022] Optionally, the survival rate or survival period includes any one or more of the following timeframes: 1 year, 3 years, 5 years.

[0023] In some embodiments, the method further includes: identifying the type of targeted drug suitable for the subject's treatment based on the RAS gene mutation risk; if the RAS gene mutation risk is low, determining that the targeted drug suitable for the subject's treatment is cetuximab; if the RAS gene mutation risk is high, determining that the targeted drug suitable for the subject's treatment is bevacizumab.

[0024] Optionally, between S2021 and S2022, the construction method further includes: segmenting the arterial phase ROI region from the arterial phase image, and segmenting the venous phase ROI region from the venous phase image; inputting the arterial phase ROI region into a machine learning model for training to obtain a predicted classification result, comparing the predicted classification result with the classification label, optimizing the model based on the comparison result, and obtaining an arterial phase classifier based on the arterial phase image; inputting the venous phase ROI region into a machine learning model for training to obtain a predicted classification result, comparing the predicted classification result with the classification label, optimizing the model based on the comparison result, and obtaining a venous phase classifier based on the venous phase image.

[0025] A third aspect of this application discloses a computer device, the device comprising: a memory and a processor; the memory being used to store a computer program; and the processor executing the computer program to implement the steps of the above-described method.

[0026] The fourth aspect of this application discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0027] The fifth aspect of this application discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0028] This application has the following beneficial effects:

[0029] 1. This application innovatively discloses a method for predicting the survival of patients with colorectal cancer liver metastases. This method calculates and integrates the DL score of the subject at different stages based on venous and arterial phase images. The DL score serves as an independent prognostic factor for patients with colorectal cancer liver metastases, used to predict the subject's survival. The calculation of the DL score is based on a predictive classifier (DL-AP+VP) constructed from venous and arterial phase images focusing on the liver metastasis region. Furthermore, this application identifies three independent predictive factors in constructing the predictive classifier. The prognosis of patients with colorectal cancer liver metastases can be predicted based on any one or more of these three factors, and a specific calculation formula is derived for achieving the predictive effect based on all three factors.

[0030] 2. This application innovatively uses CECT images of LM in CRLM patients to develop a DL model to predict RAS mutation status, realizing a non-invasive, accurate and reproducible method to assess RAS mutation status and improve the accuracy of targeted therapy decisions; it is based on the accurate and rapid quantification of RAS mutation status that the survival rate and survival time of subjects are predicted, the targeted drugs suitable for treatment are identified, and the patients most likely to benefit from cetuximab or bevacizumab treatment are identified.

[0031] This study aims to bridge the gap by developing and validating a DL radiomics (DLR) model based on contrast-enhanced computed tomography (CECT) images of the CRLM (Recurrent Leukemia-Liver-Lung Cancer) system to assess RAS genetic status. Our findings are expected to facilitate precise and non-invasive treatment adjustments in the management of CRLM. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of the method flow provided in the first aspect of the present invention;

[0034] Figure 2 This is a schematic diagram of a system for predicting the survival of CRLM patients provided in an embodiment of the present invention;

[0035] Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention;

[0036] Figure 4 This is a schematic diagram of the architecture of an exemplary computing device provided in an embodiment of the present invention;

[0037] Figure 5This is a schematic diagram of the storage medium provided in an embodiment of the present invention;

[0038] Figure 6 This is a schematic diagram of the method flow provided in the second aspect of the present invention;

[0039] Figure 7 This is a schematic diagram of a system for predicting the risk of RAS gene mutation in CRLM patients provided in an embodiment of the present invention;

[0040] Figure 8 This is a patient enrollment flowchart provided in an embodiment of the present invention;

[0041] Figure 9 This is a research flowchart provided in an embodiment of the present invention; Figure 9 A represents data preprocessing. Figure 9 B represents the DL model construction process. Figure 9 C represents risk stratification, and data preprocessing primarily relies on manual selection of ROIs and image preprocessing. Model construction consists of two different deep learning model structures (arterial phase and venous phase) and model ensemble methods.

[0042] Figure 10 This is a graph showing the ROC curve results of five models in the training queue, internal test queue, and external test queue provided in this embodiment of the invention. Figure 10 A is the ROC curve of the training queue. Figure 10 B is the ROC curve of the internal test queue. Figure 10 C is the ROC curve of the external test queue.

[0043] Figure 11 This is a nomogram of three independent predictors (CEA, DCR, and DL score) provided in this embodiment of the invention for predicting survival rate. Figure 11 A is a nomogram showing predictions from three independent predictors. Figure 11 BD displays TC, ITC, and ETC calibration curves to evaluate the performance of the DL nomogram. Figure 11 EG shows that the survival advantage of a model is assessed by calculating the net return at different risk threshold probabilities.

[0044] Figure 12 This is a schematic diagram comparing the survival rates of two RAS mutation states in the training queue, internal test queue, and external test queue provided in an embodiment of the present invention. Figure 12 A showed that the cumulative 1-year, 3-year, and 5-year overall survival rates in the low-risk RAS mutation status group were significantly higher than those in the high-risk group; Figure 12 B showed that the cumulative 1-year, 3-year, and 5-year OS rates in the low-risk RAS mutation status group of ITC were significantly higher than those in the RAS mutation status group. Figure 12C showed that the cumulative 1-year, 3-year, and 5-year OS rates in the low-risk RAS mutation status group of ETC were significantly higher than those in the RAS mutation status group.

[0045] Figure 13 These are PFS data for patients with different RAS mutation risk levels treated with cetuximab or bevacizumab, provided in embodiments of the present invention. Figure 13 A showed that in patients with low-risk RAS mutations, those who received cetuximab had a longer PFS than those who did not receive cetuximab. Figure 13 B showed that in the RAS high-risk mutation group, the median PFS in the bevacizumab group was significantly longer than that in the non-bevacizumab group. Detailed Implementation

[0046] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0047] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Figure 1 This is a schematic flowchart of a method for predicting the survival of CRLM patients provided by an embodiment of the present invention. Specifically, the method includes the following steps:

[0050] S101: Acquire arterial and venous phase images of the subject containing the target region;

[0051] In some embodiments, the terms “subject” or “test subject” or “sample” as used herein refer to any animal (e.g., a mammal), including but not limited to humans, non-human primates, rodents, etc., which will become the recipient of a particular treatment. Generally, the terms “subject” and “patient” are used interchangeably herein when referring to human subjects. Preferably, the subject is a human.

[0052] S102: Input the arterial phase image into the classifier to calculate the arterial phase DL score, input the venous phase image into the classifier to calculate the venous phase DL score, and integrate the arterial phase DL score and the venous phase DL score to obtain the total DL score;

[0053] In some embodiments, the total DL score is obtained based on a pre-constructed classifier (DL-AP+VP).

[0054] S103: Predict the survival of the subject based on the total DL score;

[0055] In some embodiments, the method further includes: obtaining the subject's CEA value and / or DCR value; predicting the subject's survival based on any one or more of the DL score, CEA value, and DCR value;

[0056] Optionally, when predicting the survival of a subject based on the DL score, CEA value, and DCR value, the subject's survival is predicted by calculating the total score of the DL score, CEA value, and DCR value.

[0057] In some embodiments, when calculating the total score based on the DL score, CEA value, and DCR value, the calculation method for the total score includes: Y = 2.995 × 10⁻⁵ × CEA + (-1.581 × DCR) + (-6.483 × 10⁻¹ × DL), where Y is the total score, CEA is carcinoembryonic antigen, and DCR is the disease control rate.

[0058] In some embodiments, between S101 and S102, the method further includes: segmenting an arterial phase ROI region from the arterial phase image and segmenting a venous phase ROI region from the venous phase image; calculating an arterial phase DL score based on the arterial phase ROI region and calculating a venous phase DL score based on the venous phase ROI region; and integrating the arterial phase DL score and the venous phase DL score to obtain a total DL score.

[0059] A second aspect of this application discloses a method for predicting the risk of RAS gene mutations in CRLM patients, the method comprising:

[0060] S201: Acquire CECT images of the subject; the CECT images include venous phase and arterial phase images;

[0061] S202: Input the CECT image into the classifier to obtain the predicted RAS gene mutation risk.

[0062] The method for constructing the classifier includes:

[0063] S2021: Obtain training set samples including images of the arterial phase (AP) and venous phase (VP) and the corresponding RAS gene status labels; the RAS gene status labels include mutant and wild-type;

[0064] In some embodiments, between S2021 and S2022, the construction method further includes: segmenting an arterial phase ROI region from the arterial phase image and segmenting a venous phase ROI region from the venous phase image; inputting the arterial phase ROI region into a machine learning model for training to obtain a predicted classification result, comparing the predicted classification result with the classification label, optimizing the model based on the comparison result, and obtaining an arterial phase classifier based on the arterial phase image; inputting the venous phase ROI region into a machine learning model for training to obtain a predicted classification result, comparing the predicted classification result with the classification label, optimizing the model based on the comparison result, and obtaining a venous phase classifier based on the venous phase image.

[0065] S2022: Input the arterial phase image into a machine learning model for training to obtain a predicted classification result, compare the predicted classification result with the classification label, optimize the model based on the comparison result, and obtain an arterial phase classifier based on the arterial phase image; input the venous phase image into a machine learning model for training to obtain a predicted classification result, compare the predicted classification result with the classification label, optimize the model based on the comparison result, and obtain a venous phase classifier based on the venous phase image;

[0066] S2023: The classifier is obtained by merging the arterial phase classifier and the venous phase classifier using an ensemble classifier.

[0067] In some embodiments, the Voting Classifier, also known as an ensemble classifier, is a technique in machine learning that combines multiple classification models to improve predictive power. Specifically, this method involves the following steps: 1) Individual learner training: First, multiple different classification models are trained independently. These models can be of the same type (e.g., several logistic regression models) or of different types (e.g., decision trees, support vector machines, random forests, etc.); 2) Prediction aggregation: For each sample to be predicted, all individual learners produce a prediction result; 3) Voting decision: Majority voting: Selecting the class with the most predictions as the final prediction result; Weighted voting: Assigning different weights based on the performance (e.g., accuracy) of each learner, and summing the weighted results to determine the predicted class; Threshold voting: Setting a threshold, only when the predictions for a certain class exceed this threshold is it determined as the final class.

[0068] In some embodiments, a total of 185 eligible CRC patients were identified, with a mean age of 64.3 ± 11.2 years, including 132 males. They were randomly assigned in a 7:3 ratio to the training cohort (TC, n = 88), the internal testing cohort (ITC, n = 39), and the external testing cohort (ETC, n = 58). There was no significant difference in the proportion of RAS wild-type patients among the different cohorts (TC: 51.1%, ITC: 61.5%, ETC: 62.1%; P = 0.278). In the TC, RAS wild-type patients showed shallower primary tumor invasion and larger LM diameter compared to RAS mutant patients (P = 0.044 and 0.026, respectively). 35 RAS wild-type patients received cetuximab treatment, and 28 RAS mutant patients received bevacizumab treatment.

[0069] DLR Feature Extraction for Predicting RAS Mutations

[0070] A total of 1,126 radiomics features and 2,589 deep learning (DL) features were extracted from each ROI in the CECT phase. Fourteen significant radiomics features associated with RAS mutation status were selected. Compared with other classifiers, the XGBoost model demonstrated superior performance in developing R scores. Furthermore, DL signatures from AP and VP ROIs were input into the FCNN to indicate the probability of RAS mutation status. The research workflow for the DLR model used to predict RAS mutation status is as follows... Figure 9 As shown.

[0071] Performance comparison of RAS mutation prediction models

[0072] This study involved the development of five models: a radiomics model, an AP-DL model, a VP-DL model, an AP+VP-DL model, and a DL-R model. The performance of all models was then examined in three cohorts. Figure 10 The AP+VP-DL model based on biphasic DL features has the best discriminative power (AUC, 0.90), compared with the radiomics model (AUC, 0.93) and the VP-DL model (AUC, 0.87). The DL-R model (AUC, 0.97) was used to predict the RAS mutation status in ETC.

[0073] The area under the curve (AUC), sensitivity (SEN), specificity (SPE), positive predictive value (PPV), and negative predictive value (NPV) of the five models are shown in Table 1. In terms of quantitative indicators, such as AUC and PPV, the AP+VP-DL model generally outperformed the other predictive models. Multivariate analysis further confirmed that CEA, DCR, and AP+VP-DL scores were significantly associated with OS (Table 2). Therefore, we selected the DL model for risk stratification of pre-treatment RAS mutation status.

[0074] The method for constructing the DL-AP model includes: acquiring arterial phase CECT images of training set samples and corresponding RAS gene status labels; the RAS gene status labels include high risk and low risk; segmenting arterial phase ROI regions from the arterial phase images; inputting the arterial phase ROI regions into a machine learning model for training to obtain predicted classification results; comparing the predicted classification results with the classification labels; optimizing the model based on the comparison results to obtain a DL-AP model based on arterial phase images.

[0075] The method for constructing the DL-VP model includes: acquiring venous phase CECT images of training set samples and corresponding RAS gene status labels; the RAS gene status labels include high risk and low risk; segmenting the venous phase ROI region from the venous phase image; inputting the venous phase ROI region into a machine learning model for training to obtain the predicted classification result; comparing the predicted classification result with the classification label; optimizing the model based on the comparison result to obtain the DL-VP model based on the venous phase image.

[0076] The method for constructing a radiomics model includes: acquiring CECT images of training set samples; segmenting ROI regions from the CECT images and extracting radiomics features based on the ROI regions; inputting the radiomics features into a machine learning model for training to obtain predicted classification results; comparing the predicted classification results with the classification labels; optimizing the model based on the comparison results to obtain a radiomics model based on radiomics features; radiomics is the application of computer image processing techniques to transform image data of regions of interest (ROIs) into high-dimensional feature data that can be mined.

[0077] The method for constructing the DL-R model includes merging the radiomics model and the DL-AP+VP model to obtain the DL-R model.

[0078] DL nomogram for predicting CRLM survival

[0079] According to multivariate Cox regression analysis, CEA, DCR (disease control rate), and AP+VP-DL score (likely a combined arterial and venous phase score derived from imaging data) were significantly associated with OS in CRLM patients (Table 2). This study identified three independent predictors of OS in CRLM patients: CEA (HR: 1.185; 95% CI: 1.028–3.645; P = 0.004), DCR (HR: 0.177; 95% CI: 0.039–0.807; P = 0.025), and DL score (high-risk vs. low-risk RAS mutation status; HR: 5.252; 95% CI: 2.174–7.549; P < 0.001). These predictors were used to develop DL nomograms, such as... Figure 11 As shown in Figure A, independently identified hazard factors were incorporated into the nomogram using the following formula: Y = 2.995 × 10⁻⁵ × CEA + (-1.581 × DCR) + (-6.483 × 10⁻¹ × DL) score. Figure 11 The TC, ITC, and ETC calibration curves shown in BD were used to evaluate the performance of the DL nomogram. These calibration curves demonstrate a strong agreement between the predicted 5-year OS probability and the actual 5-year OS probability in CRLM patients, indicating that the DL nomogram accurately estimates survival outcomes.

[0080] Table 1. Performance comparison of all models for RAS state prediction

[0081]

[0082] The numbers represent 95% confidence intervals. Bold values ​​represent the optimal AUC values.

[0083] AUC, area under; ACC, accuracy; NPV, negative predictive value; PPV, positive predictive value;

[0084] TC, Training Queue; ITC, Internal Test Queue; ETC, External Test Queue. DL-AP, Arterial Phase Deep Learning; DL-VP, Venous Phase Deep Learning; DL: Deep Learning in both Arterial and Venous Phases; DL-R, Deep Learning Radiomics.

[0085] Table 2. Factors related to OS in univariate and multivariate Cox regression analyses.

[0086]

[0087]

[0088] Abbreviations: Ci, confidence interval; kps, Karnofsky functional status; DCR, disease control rate; CEA, carcinoembryonic antigen; WBC, white blood cells; LY, lymphocytes; LDH, lactate dehydrogenase; OS, overall survival.

[0089] Decision curve analysis (DCA) was performed to assess the model's survival advantage by calculating the net benefit at different risk threshold probabilities. Considering the proportional harm of necessary treatment relative to the negative impact of unnecessary treatment, the net benefit was determined by subtracting the false positive rate from the true positive rate. DCA showed that the nomogram provided a greater advantage compared to the alternative model with the most reasonable threshold probability among the three cohorts. Figure 11 EG). The Cox model consistency index (CI) (Table 3) shows that the DL model outperforms the DL-C model in ETC (CI: 0.5543; 95% CI: 0.4546-0.6540; P = 0.0016).

[0090] Table 3. Consistency Index (CI) of the Cox Model

[0091]

[0092] Deep learning models for identifying survival benefits

[0093] The cumulative 1-year, 3-year, and 5-year overall survival rates in the low-risk RAS mutation status group were 99%, 36%, and 13%, respectively, significantly higher than those in the high-risk group (100%, 29%, and 5%). (High-risk RAS mutation status group in TC (log-rank 153 test, P = 0.025)) Figure 12A). Similarly, the cumulative 1-year, 3-year, and 5-year overall survival (OS) rates in the low-risk RAS mutation status group were 91%, 43%, and 13%, respectively, significantly higher than the 67%, 22%, and 0% in the RAS mutation status group. (Log-rank test, P = 0.0021) Figure 12 B). Furthermore, the cumulative 1-year, 3-year, and 5-year overall survival (OS) rates in the low-risk RAS mutation status group were 96%, 35%, and 15%, respectively, significantly higher than the 88%, 12%, and 0% in the RAS mutation status group. (The high-risk RAS mutation status group predicted in the ETC is shown in the log-rank test; P = 0.03). Figure 12 C).

[0094] Prediction of targeted therapy response

[0095] exist Figure 13 This paper presents PFS data for patients with different RAS mutation risk levels treated with cetuximab or bevacizumab. In patients with low-risk RAS mutations, those treated with cetuximab had a longer PFS than those not treated with cetuximab (19 vs 9 months, 95% CI: 13-8-12; P < 0.001). Figure 13 In the high-risk RAS mutation group, the median PFS in the bevacizumab group was 11 months (95% CI: 8–13 months), significantly longer than the 6 months in the non-bevacizumab group (95% CI: 5–13 months) (P = 0.042). Figure 13 B).

[0096] In some embodiments, the method further includes: predicting the survival rate of the subject based on the RAS gene mutation risk; if the RAS gene mutation risk is low, a high survival rate is obtained; if the RAS gene mutation risk is high, a low survival rate is obtained.

[0097] Optionally, the survival of the subject can be predicted based on the RAS gene mutation risk; if the RAS gene mutation risk is low, the subject will have a long survival; if the RAS gene mutation risk is high, the subject will have a short survival.

[0098] Optionally, the survival rate or survival period includes any one or more of the following timeframes: 1 year, 3 years, 5 years.

[0099] In some embodiments, the method further includes: identifying the type of targeted drug suitable for the subject's treatment based on the RAS gene mutation risk; if the RAS gene mutation risk is low, determining that the targeted drug suitable for the subject's treatment is cetuximab; if the RAS gene mutation risk is high, determining that the targeted drug suitable for the subject's treatment is bevacizumab.

[0100] In colorectal cancer patients, RAS gene mutation status has a crucial impact on treatment options for advanced colorectal cancer. Determining RAS gene mutation status is essential for subsequent treatment in these patients. Currently, RAS gene mutation status detection typically requires analysis of tissue specimens obtained through surgery or biopsy, a method that is not only invasive but also expensive. This is particularly important for patients with old tissue specimens who cannot undergo biopsy; therefore, non-invasive detection of RAS gene status via radiomics is crucial. Furthermore, RAS mutations are associated with poor prognosis and resistance to anti-EGFR monoclonal antibodies. Recent CRICKET and CAVE trials have highlighted the efficacy of cetuximab-based rechallenge strategies in third-line treatment of RAS wild-type metastatic CRC. Circulating tumor DNA (ctDNA) analysis offers hope for guiding treatment decisions, especially for RAS wild-type metastatic CRC. However, the sensitivity of ctDNA NGS detection remains challenging, raising concerns about undetected mutations. Therefore, a non-invasive and accurate method for assessing RAS genetic status is crucial for personalized CRLM treatment and optimizing outcomes.

[0101] In this invention, the prognostic indicators include objective response rate, overall survival (OS), progression-free survival (PFS), objective response rate (ORR), time to progress (TTP), disease-free survival (DFS), time to treatment failure (TTF), response rate (RR), complete response (CR), and partial response (PR). The term "prognosis" is recognized in the art and includes predictions of the likelihood of disease course or development, particularly of disease remission, disease recurrence, tumor recurrence, metastasis, and death. "Good prognosis" means the likelihood that a patient with cancer, particularly pancreatic cancer, will remain disease-free (i.e., cancer-free). "Poor prognosis" means the likelihood of recurrence or relapse, metastasis, or death of the underlying cancer or tumor.

[0102] Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention, such as... Figure 3 As shown, the device may include: one or more processors and one or more memories; wherein the memories store computer-readable code that, when run by the one or more processors, can perform the methods described above.

[0103] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, operations, and logic block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 or ARM architecture.

[0104] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0105] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 4 The architecture of the computing device 3000 shown is used for implementation. For example... Figure 4 As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage devices in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the methods provided in this disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 4 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 4 One or more components in the computing device shown.

[0106] This invention also includes a computer-readable storage medium, such as... Figure 5 The diagram illustrates a storage medium provided in an embodiment of the present invention. The computer storage medium 4020 stores computer-readable instructions 4010. When the computer-readable instructions 4010 are executed by a processor, the method described above according to embodiments of the present disclosure can be performed. The computer-readable storage medium in the embodiments of the present disclosure can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Synchronous Link Dynamic Random Access Memory (SLDRAM), and Direct Memory Bus Random Access Memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0107] This disclosure also provides a computer program product or system, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0108] In some embodiments, this embodiment also discloses a system for predicting the survival of CRLM patients, such as Figure 2 As shown, the system includes:

[0109] The first acquisition module is used to acquire arterial and venous phase images of the subject containing the target region;

[0110] The scoring calculation module is used to input the arterial phase image into the classifier to calculate the arterial phase DL score, input the venous phase image into the classifier to calculate the venous phase DL score, and integrate the arterial phase DL score and the venous phase DL score to obtain the total DL score;

[0111] The first prediction module is used to predict the survival of the subject based on the total DL score.

[0112] In some embodiments, this embodiment also discloses a system for predicting the risk of RAS gene mutations in CRLM patients, such as Figure 7 As shown, the system includes:

[0113] The second acquisition module is used to acquire CECT images of the subject; the CECT images include venous phase and arterial phase images;

[0114] The second prediction module is used to input the CECT image into the classifier to obtain the predicted RAS gene mutation risk.

[0115] Materials and methods:

[0116] From August 2016 to October 2023, a retrospective study was conducted on 514 patients with histopathologically confirmed CRLM at Beijing Luhe Hospital and Peking University Cancer Hospital. Tumor staging and RAS gene status were assessed according to the American Joint Committee on Cancer (AJCC) TNM staging and pathological examination criteria.

[0117] Inclusion criteria included: (1) histopathological diagnosis of CRC; (2) availability of RAS mutation status; (3) measurable metastatic disease according to RECIST version 1.1; (4) abdominal enhanced computed tomography (CECT) scan prior to systemic therapy for all CRLM patients; and (5) at least one measurable liver metastasis. Exclusion criteria included: (1) presence of other malignancies; (2) concomitant heart or kidney failure or severe central nervous system injury; (3) CECT image quality insufficient for reliable assessment; and (4) incomplete follow-up records. A similar protocol was implemented at Peking University Cancer Hospital during the same period.

[0118] Assessment and follow-up

[0119] CRC patients were monitored regularly until death or the date of their last follow-up (October 31, 2023). Follow-up evaluations, including CECT scans and laboratory tests, were performed every 8–12 weeks. These evaluations were conducted before each treatment or during scheduled follow-up visits, at least every 3–4 weeks. Two experienced radiologists specializing in abdominal imaging with over 5 years of expertise independently reviewed and interpreted the imaging data according to RECIST guidelines. When evaluations were inconsistent, consensus was reached through discussion.

[0120] Overall survival (OS) refers to the time from the start of initial systemic therapy at the onset of liver metastases (LM) to all-cause death or the last follow-up. Progression-free survival (PFS) is defined as the time from the start of initial systemic therapy to the occurrence of progressive disease (PD) (characterized by anomalous nodular, disseminated, and / or atypical peripheral enhancement patterns in the liver parenchyma) or subsequent action.

[0121] Clinical and pathological information

[0122] Key patient demographic and clinical variables included age, sex, Karnofsky Performance Status (KPS), alcohol and smoking history, histological grade, primary tumor location, maximum diameter and number of tumors (LMs), time to metastasis, metastatic location, carcinoembryonic antigen (CEA) level, aspartate aminotransferase (AST) level, alanine aminotransferase (ALT) level, albumin level, and TNM stage. Additionally, cetuximab use during treatment was recorded.

[0123] Construction of deep learning radiomics model

[0124] Regions of interest (ROIs) for LM nodules were manually drawn by a radiologist with 10 years of experience in abdominal imaging. ROIs were extracted using ITK-Snap 3.8.0. The deep learning (DL) model was constructed based on the two largest tumors from the CECT arterial (AP) and venous (VP) phases. The ROI size for each CECT image was resized to 64x64 pixels. DL signatures were extracted from the AP-DL model using a voting classifier with ShuffleNet V2 and from the VP-DL model using VGG19. During training, the parameters of the DL model were iteratively updated via backpropagation. The model input was the image ROI, and the output was a prediction of the RAS mutation status. The binary cross-entropy between the model output and the label was used as the loss function. To mitigate overfitting, data augmentation strategies were employed, including random sampling from video frames, random flipping, and input image rotation. Several methods were used for dimensionality reduction of radiomics features: SelectKBest, correlation comparison, minimum absolute shrinkage, and the LASSO selection operator. These methods were used to identify the most important radiomics features associated with the RAS mutation status. The selected radiomics features are then fed into a machine learning (ML) classifier, and the output of the best-performing model is converted into a probabilistic risk score, known as the radiomics score (R score). Furthermore, the deep learning score is generated using a fully connected neural network (FCNN) based on previously extracted key deep learning features. The DL model is trained using PyTorch 2.2.0 within the Python 3.8.2 framework. Detailed mathematical definitions, descriptions, and training procedures are provided in the supplementary methods.

[0125] Construction and performance of prediction models

[0126] In addition to the DL (AP / VP / AP+VP) model, radiomics and DL-Radiomic (DL-R) models were constructed and evaluated. Radiomics features were extracted from tumor ROIs using PyRadiomics, and cross-validation (CV) and the Least Absolute Shrinkage and Selection Operator (LASSO) were applied to feature selection. Then, a radiomics model was constructed using logistic regression with selected key features. The DL and radiomics models were merged to create the DL-R model. The predictive effectiveness of all models was validated and compared using ITC and ETC.

[0127] Statistical analysis

[0128] Quantitative variables were expressed as mean ± standard deviation (SD) or interquartile range median (IQR) and compared using appropriate statistical methods. Categorical variables were expressed as counts with matching percentages and compared using the chi-square test or Fisher's exact test. The area under the receiver operating characteristic (AUC) curve was used to demonstrate predictive power. The DeLong test was used to compare the AUC curves of several models. The optimal threshold for the DL model was determined using X-tile software. Statistical analysis was performed using SPSS version 23.0 and R software version 3.5.1, with significance set at P < 0.05.

[0129] Accurate and rapid quantification of RAS mutation status is crucial for identifying CRLM patients suitable for cetuximab or bevacizumab treatment. However, compared to imaging-based detection methods, the practicality of RAS detection is limited by the dynamic changes in the proportion of cells expressing RAS mutations and the invasiveness of tissue sampling. Therefore, a non-invasive, accurate, and reproducible method is needed to assess RAS mutation status. In our study, we developed a deep learning model to predict RAS mutation status using CECT images of LM from CRLM patients. The model's AUC values ​​for TC, ITC, and ETC were 0.86, 1740.82, and 0.98, respectively. The practicality of the model was further validated by generating a DL score, which helps identify patients most likely to benefit from cetuximab or bevacizumab treatment. Furthermore, the DL score is an independent prognostic factor for patients with colorectal liver metastases. This deep learning model represents a promising non-invasive tool for assessing RAS mutation status in CRLM patients, potentially improving the accuracy of targeted therapy decisions.

[0130] Grad-CAM provides a "visual interpretation" for decision-making in convolutional neural networks, enabling visualization of key network regions. It classifies the stress level (LM) region in CECT based on a coarse topographic map. In this study, we observed heatmaps of high-weight regions surrounding the LM region in AP and VP, which are consistent with the predicted LM region. Furthermore, the weights of regions surrounding the LM play a more significant role in prediction than the weights of the tumor region itself. These findings demonstrate the discriminative power and robustness of DL models in identifying and analyzing LM.

[0131] Radiomics is a complex quantitative technique that extracts imaging features invisible to the naked eye. Previous studies on radiomic prediction of CRC RAS ​​gene status have primarily focused on positron emission tomography (PET), magnetic resonance imaging (MRI), and CECT images of the primary tumor site. However, a gap remains in multimodal studies combining radiomics with LM CECT images. For example, Zhao et al. used pretreated 18F-fluorodeoxyglucose (FDG) PET in CRC patients to build a predictive model, with an area under the curve (AUC) of 0.759 (95% CI: 0.585–0.909) in the training phase and 0.701 (95% CI: 0.468–0.916) in the cohort. Despite its high accuracy, PET scans are expensive compared to CT and MRI and are less commonly used in clinical practice. Ming et al. found that a radiomics model based on multiparameter MRI had a diagnostic accuracy of 0.651 in predicting KRAS mutations in rectal cancer. Similarly, Cui et al. reported a diagnostic accuracy of 0.722 for predicting KRAS mutations using an MRI-based radiomics model, validated in both internal and external cohorts. While these studies demonstrated high diagnostic accuracy, they suffered from limitations such as small sample sizes (196) and a lack of extensive external validation. Furthermore, all these studies used tumor images from the primary colon and rectum. The presence of intestinal contents and gas can lead to inaccurate delineation of regions of interest (ROIs). Given the high incidence of advanced CRC liver metastases and the minimal difference in genetic status between primary and metastatic lesions, our study utilized CECT images of the CRLM. By delineating the ROI of the LM, we minimized the inaccuracies in ROI delineation, resulting in more reliable results.

[0132] Our results demonstrate that the DL model constructed using AP and VP ROI exhibits superior performance in both ITC and ETC, with AUC values ​​of 0.82 and 0.98, respectively. These values ​​surpass those achieved by the DL-R model, radiomics model, and clinical model. This superior performance can be attributed to the deep learning algorithm's ability to extract extensive image feature information from biphasic CECT, thereby capturing crucial details in the images. Nevertheless, the addition of radiomics features did not provide additional value in our study. However, we recognize that radiomics features can help identify genotypes of colorectal cancer. Therefore, combining these two sets of features could be complementary and potentially improve the model's predictive accuracy. Unfortunately, in this study, we did not observe higher AUC values ​​when combining features compared to using the DL model alone.

[0133] We also developed a visual nomogram based on a deep learning (DL) model, integrating traditional tumor biomarkers such as CEA, DCR, and RAS mutation risk. This nomogram is designed for personalized and non-invasive prediction of cumulative overall survival (OS) rates in CRLM patients before treatment. We observed significant differences in cumulative OS curves when CRLM patients were categorized as high-risk or low-risk for RAS mutations using our proposed DL model. Our results showed that patients with high-risk RAS mutations had significantly lower 1-, 3-, and 5-year cumulative OS rates than those with low-risk RAS mutations. These differences are consistent with actual outcomes in the RAS-mutant and RAS-wildtype groups. Furthermore, comparisons of the benefits of targeted therapy between the high-risk and low-risk groups showed that bevacizumab demonstrated benefit in the high-risk group, while cetuximab provided significant benefit in the low-risk group. These findings are consistent with actual clinical outcomes. This indirectly reaffirms the value of DL models in predicting RAS mutation status and demonstrates their potential in predicting the efficacy of targeted therapies.

[0134] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0135] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0136] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0137] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0138] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0140] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations can be made to these embodiments or their features without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure.

Claims

1. A method for predicting the survival of patients with CRLM, characterized in that, The method includes: S101: Acquire arterial and venous phase images of the subject containing the target region; acquire the subject's CEA and DCR values; S102: Segment the arterial phase ROI region from the arterial phase image and segment the venous phase ROI region from the venous phase image; input the arterial phase ROI region into the classifier to calculate the arterial phase DL score, input the venous phase ROI region into the classifier to calculate the venous phase DL score, and integrate the arterial phase DL score and the venous phase DL score to obtain the total DL score. S103: The survival of the subject is predicted by calculating the total score of the total DL score, CEA value, and DCR value; the total score is calculated as follows: Y = 2.995 × 10 -5 ×CEA+(-1.581×DCR)+(-6.483×10 -1 ×DL), where Y is the total score.

2. A method for predicting the risk of RAS gene mutation in CRLM patients, characterized in that, The method includes: S201: Acquire CECT images of the subject; the CECT images include venous phase and arterial phase images; S202: Input the CECT image into the classifier to obtain the predicted RAS gene mutation risk; The method further includes: obtaining the subject's CEA value and DCR value; calculating the total score using the total score calculation method described in claim 1; and predicting the subject's survival time based on the total score. The method for constructing the classifier includes: S2021: Obtain training set samples including images of the arterial and venous phases and corresponding RAS gene status labels for the samples; the RAS gene status labels include mutant and wild-type; S2022: Segment the arterial phase ROI region from the arterial phase image and the venous phase ROI region from the venous phase image; input the arterial phase ROI region into a machine learning model for training to obtain a predicted classification result, compare the predicted classification result with the classification label, optimize the model based on the comparison result, and obtain an arterial phase classifier based on the arterial phase image; input the venous phase ROI region into a machine learning model for training to obtain a predicted classification result, compare the predicted classification result with the classification label, optimize the model based on the comparison result, and obtain a venous phase classifier based on the venous phase image; S2023: The classifier is obtained by merging the arterial phase classifier and the venous phase classifier using an ensemble classifier.

3. The method for predicting the risk of RAS gene mutation in CRLM patients according to claim 2, characterized in that, The method further includes: predicting the survival rate of the subject based on the RAS gene mutation risk; if the RAS gene mutation risk is low, the subject has a high survival rate; if the RAS gene mutation risk is high, the subject has a low survival rate.

4. The method for predicting the risk of RAS gene mutation in CRLM patients according to claim 3, characterized in that, The survival rate includes one or more of the following timeframes: 1 year, 3 years, 5 years.

5. The method for predicting the risk of RAS gene mutation in CRLM patients according to claim 3, characterized in that, Predicting subject survival based on the RAS gene mutation risk; If the risk of RAS gene mutation is low, the subjects will have a long survival time; if the risk of RAS gene mutation is high, the subjects will have a short survival time.

6. The method for predicting the risk of RAS gene mutation in CRLM patients according to claim 5, characterized in that, The lifespan includes one or more of the following durations: 1 year, 3 years, 5 years.

7. The method for predicting the risk of RAS gene mutation in CRLM patients according to claim 2, characterized in that, The method further includes: identifying the type of targeted drug suitable for the subject's treatment based on the RAS gene mutation risk; if the RAS gene mutation risk is low, the subject is determined to be suitable for treatment with cetuximab; if the RAS gene mutation risk is high, the subject is determined to be suitable for treatment with bevacizumab.

8. A computer device, characterized in that, The device includes: a memory and a processor; the memory is used to store a computer program; the processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-7.

Images