Method for predicting small cell lung cancer immunotherapy effect based on CT radiomics

Abstract

The invention discloses a method for predicting small cell lung cancer immunotherapy effect based on CT radiomics. The method comprises the following steps: firstly, carrying out lung enhancement CT scanning on a small cell lung cancer patient receiving immunotherapy before and after treatment, extracting radiomics characteristics and calculating delta-RF; in the aspect of model construction, 1037 radiomics features are extracted, and the model is constructed through a series of processing. The model display includes a variety of predictive indicators, such as an overall lifetime (OS), a progression-free lifetime (PFS), a complete remission (CR), a partial remission (PR), a column diagram, a consistency index, a subject working characteristic curve and its underlying area, a calibration curve, and a clinical decision curve. The model has a high C index and an AUC value for PFS and OS prediction, the AUC for curative effect prediction is 0.913, and a calibration curve is close to an ideal line. Compared with a prediction scheme based on NGS sequencing, the method has the advantages of being low in cost, non-invasive, efficient, dynamic, accurate and the like, and the image omics concept is used for predicting the curative effect of SCLC immunotherapy for the first time.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a method for predicting the immunotherapy effect of small cell lung cancer based on CT imaging genomics. Background Art

[0002] Lung cancer is the disease with the highest cancer-related mortality rate in the world. Small cell lung cancer (SCLC) is one of the most malignant types of lung malignancies due to its early metastasis and shorter survival period. Although immunotherapy has improved the survival rate of patients with advanced SCLC, some patients still have primary treatment resistance and cannot benefit from it. At present, there is an urgent need for accurate, convenient, fast and inexpensive prediction methods. Radiomics can provide patients with personalized prognosis assessment and treatment response prediction by extracting and analyzing medical image feature information with high throughput. CT imaging is a basic method for lung cancer diagnosis and treatment evaluation. CT-based radiomics has certain advantages in lung cancer management. The present invention aims to provide a CT-based radiomics method for predicting the efficacy of immunotherapy in patients with advanced SCLC. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a method for predicting the immunotherapy effect of small cell lung cancer based on CT imaging.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0005] The present invention provides a method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging, comprising the following steps:

[0006] 1) For patients with small cell lung cancer receiving immunotherapy, at least one lung enhanced CT scan was performed before treatment and 6-8 weeks after treatment. Radiomic features (RF) were extracted from the baseline and 6-8 weeks after treatment, and delta-RF was calculated. The delta-RF was defined as the relative net change of RF between the two time points, and the calculation method was relative net change = (post-treatment feature - pre-treatment feature) / pre-treatment feature;

[0007] 2) Model construction:

[0008] CT radiomics feature extraction: A third-party Python library pyRadiomics was used to extract 1037 radiomics features from each volume of interest, including direct features, wavelet transform features, logarithmic transform features, and gradient filter features. They were divided into 93 gradient features, 93 LoG features, 107 original features, and 744 wavelet features according to the image type. For the feature class, there were 198 first-order features, 264 GLCM features, 154 GLDM features, 176 GLRLM features, 176 GLSZM features, 55 NGTDM features, and 14 shape features. All features were extracted from 3D volumes. Thirty randomly selected CT images were used to analyze intra-observer and inter-observer differences. An intraclass correlation coefficient (ICC) greater than 0.8 was used to identify stable and repeatable features. The Radscore of each patient in the model was calculated. The score represents the weighted sum of the selected features, and the weights are the coefficients learned during model training.

[0009] 3) Construction, verification and statistical data analysis of Delta-radiomics: The hospital database was screened for patients with small cell lung cancer who had undergone immunotherapy. The clinical and imaging data information of eligible patients was finally used to participate in the model construction. The data was processed and analyzed using R language. The data was randomly divided into a training set and a validation set with a ratio of 7:3. The training set was used for model training, and the validation set was used for model verification. The parameters in each feature were normalized, and a rank sum test was performed on the outcome in the training set data to screen out parameters with P < 0.1. The least absolute shrinkage and selection operator logistic regression algorithm was used to select significant features from the screened parameters. A 10-fold cross-validation was used to determine the optimal parameter configuration, and the coefficients were determined based on the lambda value corresponding to the minimum deviation from the distance. Parameters with non-zero coefficients were screened out, and finally the non-zero parameters and their coefficients were used in the training set. The radiomics score was calculated in the data of the first and validation sets using the formula Radscore = b0 + b1X1 + b2X2 + b3*X3..., where b0 is a constant term, X1, X2... are the non-zero coefficient parameters selected, and b1, b2... are the values ​​corresponding to the parameters. The delta radiomics combination was used to construct a model, and a nomogram was drawn. The constructed nomogram was validated by 1000 bootstrap samplings, and a calibration curve was drawn to evaluate the calibration of the model. The receiver operating characteristic curve was further analyzed to calculate the AUC, sensitivity, specificity, accuracy and other indicators to evaluate its discrimination efficiency. A clinical decision curve was constructed to evaluate the clinical application value of the model, and the net benefit within the threshold probability range was quantified. Finally, the constructed model was validated in the validation set. All tests were two-sided tests, and differences were considered statistically significant when P < 0.05.

[0010] As a preferred technical solution of the present invention, the predictive efficacy of the method for survival time is: for the prediction of PFS, the training set C index is 0.824, and the AUCs of 6-month, 12-month, and 18-month PFS in the ROC curve are 0.888, 0.881, and 0.859, respectively; for the prediction of OS, the training set C index is 0.862, and the AUCs of 6-month, 12-month, and 18-month OS in the ROC curve are 0.975, 0.890, and 0.930, respectively.

[0011] As a preferred technical solution of the present invention, the predictive efficacy of the method for the efficacy of immunotherapy is: the AUC of the ROC curve of the training set is 0.913, the calibration curve is close to the ideal line, and the net benefit within the threshold probability range of the decision curve is higher than the all line and the none line.

[0012] As a preferred technical solution of the present invention, the model prediction formula is:

[0013] PFS prediction: The total score is calculated by Radscore PFS, corresponding to the predicted probability of PFS at different time points;

[0014] OS prediction: The total score is calculated by Radscore_OS, corresponding to the OS prediction probability at different time points;

[0015] Efficacy prediction: The total score is calculated by Radscore, which corresponds to the predicted probability of immunotherapy efficacy.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] The present invention is a method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging genomics, providing a method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging genomics, so as to improve the accuracy and effectiveness of treatment for SCLC patients, realize individualized precision treatment, improve survival rate, improve quality of life and reduce ineffective medical expenses. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0019] Figure 1 : Schematic diagram of the process structure of the method of the present invention;

[0020] Figure 2 : PFS prediction model prediction formula diagram in the present invention;

[0021] Figure 3 : The prediction formula diagram of the OS prediction model in the present invention;

[0022] Figure 4 : The prediction formula diagram of the efficacy prediction model in the present invention; DETAILED DESCRIPTION

[0023] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0024] Example 1

[0025] like Figure 1-4 As shown, the present invention provides a method for predicting the immunotherapy effect of small cell lung cancer based on CT imaging, comprising the following steps:

[0026] 1. Data Collection

[0027] Methods: For patients with small cell lung cancer receiving immunotherapy (ICI), at least one lung enhanced CT scan was performed before treatment and 6-8 weeks after treatment. Radiomic features (RF) were extracted from the baseline and 6-8 weeks after treatment, and delta-RF was calculated (delta-RF = (features after treatment - features before treatment) / features before treatment).

[0028] 2. Model Construction

[0029] CT radiomics feature extraction

[0030] 1037 radiomic features were extracted from each volume of interest (VOI) using the third-party Python library pyRadiomics, including various types of features (such as gradient features, LoG features, etc.), and the features were extracted from the 3D volume. By analyzing the intra-observer and inter-observer differences of 30 randomly selected CT images, stable and repeatable features were identified with an intraclass correlation coefficient (ICC) greater than 0.8, and the Radscore (representing the weighted sum of the selected features, with the weights being the coefficients learned by model training) was calculated for each patient.

[0031] 3. Construction, verification and statistical data analysis of Delta-radiomics

[0032] Patient data from hospitals within a specific time period were screened, and finally 83 eligible patients were involved in model construction, which was processed and analyzed using R language (version 4.3.3). The data were randomly divided into training set and validation set with a ratio of 7:3. After normalizing the characteristic parameters, the rank sum test was performed in the training set to screen parameters with P < 0.1. The least absolute shrinkage and selection operator (LASSO) logistic regression algorithm was used to select significant features. The optimal parameter configuration was determined by 10-fold cross validation, and the coefficient was determined based on the lambda value. After screening the non-zero coefficient parameters, the radiomics score (Radscore formula) was calculated. The model was constructed and a nomogram was drawn. Bootstrap sampling validation, calibration curve, ROC curve analysis (calculation of AUC and other indicators) were performed, and a clinical decision curve (DCA) was constructed. Finally, it was validated in the validation set. The test was a two-sided test, and P < 0.05 was statistically significant.

[0033] Model presentation and prediction performance

[0034] Clinical and model predictors

[0035] (1) Overall survival (OS): The time from the start of treatment to death due to any cause. OS is considered the best efficacy endpoint in cancer clinical trials and is the preferred endpoint when the patient's survival can be fully evaluated.

[0036] (2) Progression-free survival (PFS): The time from the start of treatment to tumor progression or death from any cause. Compared with OS, PFS adds the node of disease progression, and “disease progression often precedes death”. PFS is often shorter than OS and can be evaluated before OS.

[0037] (3) Complete remission (CR): All tumor target lesions completely disappear. All target nodules must be reduced to normal size (short axis <10 mm) compared with before tumor treatment. When determining the efficacy, all target lesions must be evaluated. Partial remission (PR): Compared with before tumor treatment, the sum of the diameters of all measurable tumor target lesions is reduced by ≥30%.

[0038] (4) Nomogram: A statistical chart that predicts the probability of a specific event (such as disease occurrence, treatment effect, etc.) by converting the effects of multiple predictive variables into points on a line and finally summarizing them into a total score. The nomogram is based on the results of multivariate regression analysis (Logistic regression model, Cox regression model), integrates multiple influencing factors, and then draws scaled segments on the same plane at a certain ratio to express the relationship between the variables in the prediction model. The nomogram can convert complex regression equations into simple visual graphics, making the prediction results easier to read and more convenient to evaluate the patient's condition. It can be used in clinical practice for multi-indicator joint diagnosis or prediction of disease risk or prognosis. The single score (Point) represents the single score corresponding to each variable under different values; the total score (TotalPoint) represents the total score of the single scores corresponding to all variables after the values ​​are taken. The total score corresponds to the probability of the model prediction result.

[0039] (5) Consistency index (C index): used to calculate the degree of differentiation between the Cox model prediction value and the actual value in survival analysis, and is often used to evaluate the prediction accuracy of the patient prognosis model. The value range is between 0.5 and 1. 0.5-0.6 means basically no predictive ability; 0.6-0.7 means low accuracy; 0.7-0.9 means good accuracy and acceptable; and above 0.9 means high accuracy.

[0040] (6) Receiver operating characteristic curve (ROC curve): It is a curve obtained by plotting the true positive rate and the false positive rate, which can reflect the relationship between sensitivity and specificity. The horizontal axis represents the false positive rate (1-specificity), and the vertical axis represents the true positive rate (sensitivity). Any point on the curve represents the sensitivity and specificity corresponding to a specific positive standard value of a screening test. It is used to judge the diagnostic efficacy of the disease, the discriminating ability of the diagnostic test or the prediction model. The larger the area under the ROC curve, the better the discriminating ability of the model. The closer the curve is to the upper left corner, the higher the prediction accuracy.

[0041] (7) AUC: represents the area under the ROC curve. It is used to quantify the performance of an indicator / model and can be used to compare the performance of two indicators / models. The AUC value ranges from 0.5 to 1. The closer it is to 1, the better the diagnostic performance of the indicator / model and the higher the accuracy. AUC = 0.5 to 0.7, low accuracy; AUC = 0.7 to 0.9, good accuracy; AUC> 0.9, high accuracy.

[0042] (8) Calibration curve: A graphical tool that evaluates the accuracy of predicted probabilities by comparing the predicted probabilities with the observed probabilities at different quantiles. A calibrated model should make the calibration curve as close to the ideal line as possible, indicating that the model's predicted probabilities match the actual probabilities.

[0043] (9) Clinical decision curve (DCA): to evaluate the clinical application value of the model and quantify the net benefit within the threshold probability range.

[0044] Prediction performance

[0045] Prediction of survival time (progression-free survival (PFS) and overall survival (OS):

[0046]

[0047] PFS prediction: The C index of the training set was 0.824, and the AUCs of 6-month, 12-month, and 18-month PFS in the ROC curve were 0.888, 0.881, and 0.859, respectively.

[0048] OS prediction: The C index of the training set was 0.862, and the AUCs of 6-month, 12-month, and 18-month OS in the ROC curve were 0.975, 0.890, and 0.930, respectively.

[0049] Prediction of immunotherapy efficacy: The AUC of the ROC curve of the training set was 0.913, the calibration curve was close to the ideal line, and the net benefit within the threshold probability range of the decision curve was higher than the comparison line, indicating that the model had practical value.

[0050] Model prediction formula (nomogram)

[0051] PFS prediction: The total score calculated by Radscore PFS corresponds to the predicted probability of PFS at different time points.

[0052] OS prediction: The total score is calculated by Radscore_OS to predict the OS probability at different time points.

[0053] Efficacy prediction: The total score corresponding to the predicted probability of immunotherapy efficacy (CR+PR) is calculated by Radscore.

[0054] Compared with the only SCLC prediction scheme based on NGS sequencing, the present invention has the advantages of low cost, non-invasive, high efficiency, dynamic and accurate. Moreover, this study is the first to apply the concept of imaging omics to predict the efficacy of SCLC immunotherapy, and there is no other alternative.

[0055] According to the above data collection requirements, patients underwent lung enhanced CT scans to ensure accurate extraction of radiomic features at two time points and calculation of delta-RF.

[0056] During the model building phase, the steps of feature extraction, data processing, model training and verification are strictly implemented, including the use of specified software libraries and algorithms, dividing the data set according to the specified proportion, and performing parameter screening and calculation operations.

[0057] According to the model prediction formula, patient-related data are input in actual applications to calculate Radscore, and the prediction results of survival time and immunotherapy efficacy are obtained through nomograms and other methods to provide a reference for clinical decision-making.

[0058] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging, characterized in that: The following steps are involved: 1) For patients with small cell lung cancer receiving immunotherapy, at least one lung enhanced CT scan was performed before treatment and 6-8 weeks after treatment. Radiomic features (RF) were extracted from the baseline and 6-8 weeks after treatment, and delta-RF was calculated. The delta-RF was defined as the relative net change of RF between the two time points, and the calculation method was relative net change = (post-treatment feature - pre-treatment feature) / pre-treatment feature; 2) Model construction: CT radiomics feature extraction: A third-party Python library pyRadiomics was used to extract 1037 radiomics features from each volume of interest, including direct features, wavelet transform features, logarithmic transform features, and gradient filter features. They were divided into 93 gradient features, 93 LoG features, 107 original features, and 744 wavelet features according to the image type. For the feature class, there were 198 first-order features, 264 GLCM features, 154 GLDM features, 176 GLRLM features, 176 GLSZM features, 55 NGTDM features, and 14 shape features. All features were extracted from 3D volumes. Thirty randomly selected CT images were used to analyze intra-observer and inter-observer differences. An intraclass correlation coefficient (ICC) greater than 0.8 was used to identify stable and repeatable features. The Radscore of each patient in the model was calculated. The score represents the weighted sum of the selected features, and the weights are the coefficients learned during model training. 3) Construction, verification and statistical data analysis of Delta-radiomics: The hospital database was screened for patients with small cell lung cancer who had undergone immunotherapy. The clinical and imaging data information of eligible patients was finally used to participate in the model construction. The data was processed and analyzed using R language. The data was randomly divided into a training set and a validation set with a ratio of 7:

3. The training set was used for model training, and the validation set was used for model verification. The parameters in each feature were normalized, and a rank sum test was performed on the outcome in the training set data to screen out parameters with P < 0.

1. The least absolute shrinkage and selection operator logistic regression algorithm was used to select significant features from the screened parameters. A 10-fold cross-validation was used to determine the optimal parameter configuration, and the coefficients were determined based on the lambda value corresponding to the minimum deviation from the distance. Parameters with non-zero coefficients were screened out, and finally the non-zero parameters and their coefficients were used in the training set. The radiomics score was calculated in the data of the first and validation sets using the formula Radscore = b0 + b1X1 + b2X2 + b3*X3..., where b0 is a constant term, X1, X2... are the non-zero coefficient parameters selected, and b1, b2... are the values ​​corresponding to the parameters. The delta radiomics combination was used to construct a model, and a nomogram was drawn. The constructed nomogram was validated by 1000 bootstrap samplings, and a calibration curve was drawn to evaluate the calibration of the model. The receiver operating characteristic curve was further analyzed to calculate the AUC, sensitivity, specificity, accuracy and other indicators to evaluate its discrimination efficiency. A clinical decision curve was constructed to evaluate the clinical application value of the model, and the net benefit within the threshold probability range was quantified. Finally, the constructed model was validated in the validation set. All tests were two-sided tests, and differences were considered statistically significant when P < 0.

05.

2. The method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging according to claim 1, characterized in that: The predictive efficacy of the method for survival time is as follows: for the prediction of PFS, the training set C index is 0.824, and the AUCs of 6-month, 12-month, and 18-month PFS in the ROC curve are 0.888, 0.881, and 0.859, respectively; for the prediction of OS, the training set C index is 0.862, and the AUCs of 6-month, 12-month, and 18-month OS in the ROC curve are 0.975, 0.890, and 0.930, respectively.

3. The method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging according to claim 1, characterized in that: The predictive efficacy of the method for the efficacy of immunotherapy is as follows: the AUC of the ROC curve of the training set is 0.913, the calibration curve is close to the ideal line, and the net benefit within the threshold probability range of the decision curve is higher than the all line and the none line.

4. The method for predicting the effect of immunotherapy for small cell lung cancer based on CT imaging according to claim 1, characterized in that: The model prediction formula is: PFS prediction: The total score is calculated by Radscore PFS, corresponding to the predicted probability of PFS at different time points; OS prediction: The total score is calculated by Radscore_OS, corresponding to the OS prediction probability at different time points; Efficacy prediction: The total score is calculated by Radscore, which corresponds to the predicted probability of immunotherapy efficacy.

Images