Small cell lung cancer risk assessment method and system

Abstract

The present invention discloses a risk assessment method and system for small cell lung cancer, wherein the method comprises: step S1, image marker screening; step S2, comprehensive risk assessment; step S3, risk point scanning, which can solve the technical problems of lack of 0-3mm small cell lung cancer risk assessment method and lack of early identification and positioning technology of small cell lung cancer in related technologies.

Description

Technical Field

[0001] The present invention relates to the technical field of small cell lung cancer risk detection and assessment, and in particular to a small cell lung cancer risk assessment method and system. Background Art

[0002] Small cell lung cancer is a type of neuroendocrine tumor that is poorly differentiated, grows rapidly (doubling time is about 30-60 days), and is prone to early metastasis, especially lymphatic dissemination. Currently, only about 5-30% of small cell lung cancer patients are in the limited stage when diagnosed, and extensive-stage small cell lung cancer is no longer suitable for surgery. Systemic treatment has made slow progress in the past 30 years, and the mortality rate of patients is high, with an average survival period of only 6-10 months. The above facts make early intervention of small cell lung cancer particularly important.

[0003] Low-dose CT provides an important means for early diagnosis of lung cancer and is a recognized screening measure that can effectively reduce the overall mortality rate of lung cancer. However, for patients with small cell lung cancer, screening has not changed the patient's survival outcome. Many studies have shown that patients with small cell lung cancer have not benefited from low-dose CT lung cancer screening practices.

[0004] The key reason for the above problems is that the detection window of small cell lung cancer is too short. The process from "none" to mass and even widespread metastasis can be completed within 1 year. The current lung cancer screening detection standard is a nodule or mass larger than 3mm, and the recommendation for screening subjects with "negative" results is generally an annual review. This means that when the typical manifestations of small cell lung cancer are found, the patient has already lost the opportunity for surgery. The existing low-dose CT screening technology solutions are mainly aimed at the needs of non-small cell lung cancer, and it is difficult to capture small cell lung cancer that can be intervened in a timely and effective manner.

[0005] Lesions smaller than 3 mm are seriously affected by noise under low-dose CT and are easily hidden in normal lung texture images, making them difficult to identify with the naked eye, greatly increasing the difficulty of retrieval in the entire lung field. Although there are technologies dedicated to lung nodule recognition and segmentation that can help reduce the clinical burden, such technologies focus on suspicious lesions larger than 3 mm, and the location and analysis technology of "pre-nodule" lesions (i.e., less than 3 mm, the embryonic stage of lesions) is still blank, so it is impossible to solve the problem of identifying risk sites and early diagnosis of small cell lung cancer in the aforementioned "negative" screening subjects.

[0006] There are other technologies for screening and diagnosis, and their limitations for early identification of small cell lung cancer are briefly summarized below.

[0007] (1) Histopathology. It is the gold standard for the diagnosis of small cell lung cancer. However, the detection rate of sputum-based pathological smears is very low, and most of them appear in the extensive stage. Both puncture and surgical pathology are invasive operations, with low patient acceptance, and limited tissue samples are also prone to missed diagnoses.

[0008] (2) Molecular markers. Most of them are presented in the form of detection kits, including the detection of specific monoclonal antibodies, proteases, susceptibility genes, RNA and other molecules for small cell lung cancer. The performance of molecular markers for small cell lung cancer is still generally lacking in verification, especially the value of early identification still needs to be evaluated, and the clinical popularity is not as good as imaging methods.

[0009] (3) Prediction model. Statistical techniques, machine learning, and other artificial intelligence technologies are dedicated to integrating multi-dimensional or high-dimensional data information (such as combining image features, laboratory parameters, and basic patient information) and have developed prognostic assessment models for small cell lung cancer. However, they are not dedicated to the identification of high-risk groups for small cell lung cancer and risk assessment, and are unlikely to play a role in the early diagnosis of small cell lung cancer.

[0010] In summary: There is currently a lack of risk assessment methods for 0-3mm small cell lung cancer, and a lack of technology for early identification and localization of small cell lung cancer. Summary of the invention

[0011] The purpose of the present invention is to overcome the above technical deficiencies and provide a risk assessment method and system for small cell lung cancer to solve the technical problems in the related art of lacking a risk assessment method for 0-3mm small cell lung cancer and a lack of early identification and positioning technology for small cell lung cancer.

[0012] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0013] According to one aspect of the present invention, a method for risk assessment of small cell lung cancer is provided, comprising: step S1, screening of imaging markers; step S2, comprehensive risk assessment; step S3, risk point scanning; wherein:

[0014] The step S1 specifically includes:

[0015] Step S11, case and control selection: using case and control design to screen imaging markers, so that the screened imaging markers have the ability to sensitively detect and specifically identify small cell lung cancer; the controls include: small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control;

[0016] Step S12, front and back image registration: for all included cases and controls, define the CT examination in which the lesion is found as R1, and the CT examination before the lesion is found as R0, perform three-dimensional rendering of the whole chest image of R0 and R1, and then perform image registration of R1 and R0;

[0017] Step S13, delineation of voxels of interest: synchronizing the R0 and R1 lung windows, locating the R1 lesion, covering the lesion position, generating and storing a mask of voxels of interest;

[0018] Step S14, radiomics data extraction: applying the mask to R0 to form a VOI for R0, and extracting radiomics data for the lung texture where it is located;

[0019] Step S15, indicator evaluation and preliminary screening: stability evaluation of the indicators, differential test of the image features of cases and controls, wherein the indicators with significant differences between the small cell lung cancer self-control and the nodule-free patient same-point control are defined as risk warning image markers for risk point identification; the indicators with significant differences between the small cell lung cancer self-control, the nodule-free patient same-point control, the benign nodule patient control, and the non-small cell lung cancer control are defined as risk assessment image markers for comprehensive modeling;

[0020] The step S2 specifically includes:

[0021] Step S21, risk assessment mark selection: high correlation marks are eliminated;

[0022] Step S22, risk assessment model training: taking whether the patient progresses to small cell lung cancer as the outcome, taking the selected imaging markers as the predictive factors, and combining the known risk factors for small cell lung cancer, a small cell lung cancer risk assessment model is constructed;

[0023] Step S23, model evaluation and testing;

[0024] Further, step S2 also includes step S24, risk assessment report: using the positioning results of the risk point scanning in step S3, extracting the image index determined in step S21, calculating the risk probability of small cell lung cancer, and generating a small cell lung cancer risk assessment report;

[0025] The step S3 specifically includes:

[0026] Step S31, risk warning sign selection: for risk warning image signs, the AUC is calculated one by one based on whether a new disease change occurs as the outcome, and the top n candidate signs with the highest AUC are selected. The sensitivity under the same specificity is further calculated, and the top m image signs with the highest sensitivity are selected, and standard transformation is performed to define Score=max(S) for the identification of risk points;

[0027] Step S32, segmentation of the entire lung anatomical parts;

[0028] Step S33, multi-scale hierarchical detection: scan each site in the order of the common sites of small cell lung cancer, find potential candidate areas, calculate the score value, and define the threshold library at different scales for temporary storage and update of the maximum score value of different blocks at the same scale during the parallel calculation process, until the risk point with the highest score value is detected.

[0029] According to another aspect of the present invention, a risk assessment system for small cell lung cancer is provided, comprising: an image marker screening module, a comprehensive risk assessment module, and a risk point scanning module, wherein:

[0030] The image mark screening module is specifically used for:

[0031] Selection of cases and controls: The imaging markers were screened using a case and control design, so that the imaging markers screened had the ability to sensitively detect and specifically identify small cell lung cancer; the controls included: small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control;

[0032] Pre- and post-image registration: For all included cases and controls, the CT examination that found the lesion was defined as R1, and the CT examination before the lesion was defined as R0. The whole chest image of R0 and R1 was rendered in three dimensions, and then R1 and R0 were registered.

[0033] Voxel of interest delineation: synchronize R0 and R1 lung windows, locate R1 lesions, cover lesion locations, generate and store voxel of interest masks;

[0034] Radiomics data extraction: Apply the mask to R0 to form a VOI for R0, and extract radiomics data for the lung texture where it is located;

[0035] Index evaluation and initial screening: The indicators were evaluated for stability, and the image features of cases and controls were tested for differentiation. Among them, the indicators with significant differences for small cell lung cancer self-control and nodule-free patient same-point control were defined as risk warning imaging markers for risk point identification; the indicators with significant differences for small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control were defined as risk assessment imaging markers for comprehensive modeling;

[0036] The comprehensive risk assessment module is specifically used for:

[0037] Risk assessment sign selection: remove highly relevant signs;

[0038] Risk assessment model training: Taking progression to small cell lung cancer as the outcome and carefully selected imaging markers as predictive factors, combined with known risk factors for small cell lung cancer, a small cell lung cancer risk assessment model was constructed;

[0039] Model evaluation and testing;

[0040] Risk assessment report: Using the positioning results of the risk point scanning module, extract the determined imaging indicators, calculate the risk probability of small cell lung cancer, and generate a small cell lung cancer risk assessment report;

[0041] The risk point scanning module is specifically used for:

[0042] Selection of risk warning signs: For risk warning imaging signs, the AUC is calculated one by one based on whether new diseases occur or not. The top n candidate signs with the highest AUC are selected, and the sensitivity is further calculated under the same specificity. The top m imaging signs with the highest sensitivity are selected, and standard transformation is performed. Score=max(S) is defined for the identification of risk points.

[0043] Segmentation of the entire lung’s anatomical parts;

[0044] Multi-scale hierarchical detection: Scan each site in the order of the common sites of small cell lung cancer to find potential candidate areas, calculate the score value, and define threshold libraries at different scales for temporary storage and update of the maximum score values ​​of different blocks at the same scale during parallel calculation until the risk point with the highest score value is detected.

[0045] Furthermore, the image registration of R1 and R0 specifically includes:

[0046] The whole chest images of R0 and R1 were rendered in three dimensions. The sternal angle, scapula and ribs of the R0 image were used as landmarks, and the R1 image was translated and / or angled in the sagittal, coronal and axial planes until the landmarks overlapped.

[0047] Furthermore, the model evaluation and testing specifically include: evaluating the model's discrimination for small cell lung cancer using the area under the receiver operating characteristic curve, evaluating the calibration of the small cell lung cancer risk prediction probability using the Brier score, recalibrating the model based on the probability of small cell lung cancer in the application scenario, and comprehensively evaluating the small cell lung cancer risk assessment model in an independent test set.

[0048] Furthermore, the whole lung anatomical part segmentation specifically includes: removing erroneous segmentation, filling in holes, ensuring the accuracy and continuity of segmentation, and further performing anatomical part segmentation on each part in turn according to the common sites of small cell lung cancer.

[0049] Furthermore, the scanning of each part specifically includes: scanning each part in turn quickly at a large scale and low resolution to find potential candidate areas, and then reducing the scale and increasing the resolution to perform block segmentation.

[0050] Furthermore, the risk point scanning module is also used for:

[0051] Risk point adjustment, the adjustment mechanism includes block window shape design, fine step sliding, and adjustment of threshold library definition rules.

[0052] Furthermore, the risk point scanning module is also used for:

[0053] Visualization of risk points: The detected risk points are displayed and recorded in the whole lung field of view and local field of view to guide the follow-up image positioning or guide the puncture operation.

[0054] According to another aspect of the present invention, there is also provided an electronic device, comprising: a processor and a memory;

[0055] The memory stores a computer-readable program executable by the processor;

[0056] When the processor executes the computer-readable program, the steps in the above method are implemented.

[0057] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the above method.

[0058] The present invention provides a risk assessment method and system for small cell lung cancer, which uses radiomics features to identify risk points and quantify the risk level. This computer vision far exceeds the scope of human visual assessment, and reduces the "positive" threshold of small cell lung cancer to below 3mm, thereby realizing the identification of 0-3mm small cell lung cancer risk points. This greatly expands the detection window period for small cell lung cancer, and then through refined risk assessment and follow-up, valuable time is gained for clinical intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic flow chart of a risk assessment method for small cell carcinoma is provided for an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of the structure of a risk assessment system for small cell lung cancer is provided for an embodiment of the present invention (mainly showing the workflow of an image marker screening module);

[0061] Figure 3 A schematic diagram of the structure of a risk assessment system for small cell lung cancer is provided for an embodiment of the present invention (mainly showing the workflow of a comprehensive risk assessment module);

[0062] Figure 4 A schematic diagram of the structure of a risk assessment system for small cell lung cancer is provided for an embodiment of the present invention (mainly showing the workflow of a risk point scanning module);

[0063] Figure 5 It is a structural block diagram of a terminal that can implement a risk assessment method for small cell carcinoma according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0065] Embodiment 1

[0066] According to an embodiment of the present invention, a method for evaluating small cell lung cancer is provided, combining Figure 1 , the method comprising:

[0067] Step S1: Image marker screening

[0068] In this embodiment, step S1 specifically includes:

[0069] Step S11, case and control selection

[0070] Small cell lung cancer accounts for a relatively low proportion of all lung cancer types (about 15%), and a case-control design was used for imaging marker screening. The following four forms of control were considered so that the imaging markers screened have the ability to sensitively detect and specifically identify small cell lung cancer:

[0071] ① Small cell lung cancer self-control (non-cancer site in the same lung lobe); ② Control of the same site in patients without nodules; ③ Control of patients with benign nodules; ④ Control of non-small cell lung cancer, including squamous cell carcinoma and adenocarcinoma. The principles for selecting controls: the same gender, similar age and smoking history, to exclude the interference of these common risk factors on the screening results; for controls of types ③-④, new nodules are selected (the probability of small cell lung cancer in non-new nodules is extremely low); in addition, for controls of type ④, the year of diagnosis is the same or similar to that of the case.

[0072] Step S12: front and back image registration

[0073] For all included cases and controls, the CT examination in which the lesion was found was defined as R1, and the CT examination before the lesion was found was defined as R0. The whole chest image of R0 and R1 was rendered in three dimensions, and the sternal angle, scapula, and ribs of the R0 image were used as landmarks. The R1 image was translated and / or angled in the sagittal, coronal, and axial planes until the landmarks overlapped, and then the images of R0 and R1 were registered. Image registration is a key step to ensure that the location of the new lesion in R1 is matched in R0 (i.e., the pre-nodular image).

[0074] Conventional radiomics analysis mainly analyzes images of lung nodules. However, for small cell lung cancer, which is the focus of the present invention, its progression rate is very fast, and it is difficult to find its "nodule stage" clinically. Therefore, the idea adopted by the present invention during research and development is to analyze the lung texture at the location of the nodule in the "early stage" or even in the stage where it cannot be identified by the naked eye. However, when the inventors were developing along this line of thought, they found two technical difficulties: (1) It is difficult to judge where nodules may grow in the future; (2) The imaging features and radiomics features that are of diagnostic significance for nodules are difficult to apply to the lung texture in the "early stage" of nodules.

[0075] Regarding technical difficulty (1), the innovation of the present invention lies in focusing on R0 (CT examination before the discovery of lesions) and R1 (CT examination after the discovery of lesions) at the same time, and utilizing the correspondence between R1 and R0 images in anatomical landmarks. Through three-dimensional image rendering and image registration, the location of the nodule in R1 is accurately matched to R0 with the help of the voxel mask of interest. This means that the point where the nodule will occur in the future can be found in a large range of lung texture images, and then the image features with risk point scanning and comprehensive risk assessment value can be found through image analysis technology.

[0076] Regarding technical difficulty (2), because there are essential differences between lung nodule images and lung texture images, the image extraction methods are different (for example, it is impossible to extract geometric features such as edges and contours). In addition, the image features and radiomics features required to determine whether lung nodules will grow on the basis of lung texture at a specific location and the nature of the nodules that grow are very different from the features that traditional lung nodule analysis relies on. Therefore, whether it is subsequent analysis or subsequent modeling, the parameters relied on are different from traditional technologies.

[0077] In addition, in step S12, R0 and R1 are used for registration, which not only considers the axial image, but also includes the sagittal and coronal images, which increases the accuracy of the registration. Multi-angle registration can more comprehensively consider the anatomical location of new lesions, thereby improving the accuracy of small cell carcinoma lesion detection. The use of sternal angles, scapula, and ribs as registration benchmarks can serve as reliable reference points to ensure the alignment between the two scans. In addition, R0 and R1 are three-dimensionally rendered as the entire chest image, and then translated and angled in multiple planes until the registration benchmarks coincide, which is critical for the early positioning of lesions.

[0078] Step S13: Voxel of interest delineation

[0079] Synchronize the R0 and R1 lung windows, locate the R1 lesion, cover the lesion position with equal radius circles or squares layer by layer, generate and store cylindrical or cubic voxel of interest (VOI) masks.

[0080] Step S14, radiomics data extraction

[0081] The mask is applied to R0 to form a VOI for R0, and the imaging omics data is extracted for the lung texture. The imaging omics data includes but is not limited to the following category features: first-order grayscale features, grayscale co-occurrence matrix features, grayscale dependency matrix features, grayscale run matrix features, grayscale size area matrix features, domain grayscale difference matrix features, etc. These features extract potential small cell carcinoma microenvironment change information from different aspects. In some embodiments, wavelet changes and curvelet changes can also be combined to extract richer texture details.

[0082] In step S14, the radiomics data covers features of multiple dimensions, including first-order grayscale features, grayscale co-occurrence matrix features, grayscale dependency matrix features, grayscale run matrix features, grayscale size region matrix features, domain grayscale difference matrix features, etc. These features can capture subtle changes in the lesion area from different angles, which helps to more comprehensively evaluate the lesion condition. By combining techniques such as wavelet transform and curvelet transform, richer texture details can be extracted. These advanced transformation techniques can reveal complex patterns in images and further enhance the effect of feature extraction, especially for the detection of small changes.

[0083] Step S15: Index evaluation and initial screening

[0084] The stability of the indicators is evaluated, focusing on the stability of different indicators in the low-dose imaging mode, and the robustness to different imaging devices, scanning conditions and parameters. This part of the evaluation is used to exclude indicators that are not reliable in different scenarios, and can be performed through noise simulation room tests and body model room tests. Furthermore, the image features of cases and controls are tested for differentiation, where indicators that are significantly different from Class ①-② controls are defined as "risk warning" imaging signs, which are used for risk point identification, and indicators that are significantly different from Class ①-④ controls are defined as "risk assessment" imaging signs, which are used for comprehensive modeling, refined assessment of patients' small cell risk, and differentiation from the risks of other benign and malignant lesions.

[0085] For step S15, the traditional technical solution is to screen a set of differential indicators and then directly build a model for application. In step S15, it is divided into two steps, each corresponding to a module. (1) The risk identification module mainly considers that in specific applications, scanning the entire lung will be very time-consuming if too many indicators are used. Therefore, the risk warning image sign recognition is performed based on the comparison of categories ① and ② (i.e., judging whether new nodules will grow). At the same time, only the "first m items" indicators are calculated in this set of indicators to achieve the purpose of rapid positioning. (2) The risk assessment module uses the intersection of differential indicators under the comparison of categories ①-④ as the risk assessment image sign to perform a multi-dimensional and refined assessment (i.e., identify the type of future lesions). This part is only performed on high-risk points identified by the "risk identification module", so it can improve application efficiency.

[0086] Step S2: Comprehensive risk assessment

[0087] Step S21, risk assessment mark selection

[0088] Most of the "risk assessment" image markers are correlated (i.e., image information redundancy), which will affect the construction effect of the risk assessment model. In this embodiment, principal component analysis and other methods can be used for dimensionality reduction, minimum absolute shrinkage and selection operator regression method feature reduction, or combined with the above or other screening criteria to eliminate highly correlated markers to improve the accuracy of risk assessment model construction.

[0089] Step S22: Risk assessment model training

[0090] A small cell lung cancer risk assessment model is constructed with whether or not the disease progresses to small cell lung cancer as the outcome, and with selected imaging markers as predictive factors, combined with known small cell lung cancer risk factors (such as smoking, family history, and chronic obstructive pulmonary disease). The model is constructed using random forests to fully mobilize the ability of complex interactions between variables to improve the accuracy of the assessment, and to ensure the generalization performance of the model. In some other embodiments, other statistical modeling or machine learning methods can also be considered based on the model performance. Depending on the sample accessibility, splitting the training and validation samples or performing cross-validation can be considered. If necessary, measures such as oversampling and weight adjustment can be used to deal with inter-class imbalance problems during the model training stage. For example, oversampling increases the amount of data for the minority class by duplicating minority class samples or synthesizing new minority class samples, so that the risk assessment model can better learn the characteristics of the minority class, which helps to improve the recognition rate of the minority class and reduce false negatives. Weight adjustment can assign different loss weights to samples of different categories during the training process to compensate for the imbalance in the number of samples, so that the original data set does not need to be changed, and the weights in the loss function are adjusted to affect the learning process of the model. Giving higher weights to minority classes can encourage the risk assessment model to pay more attention to the prediction accuracy of minority classes.

[0091] Step S23, model evaluation and testing

[0092] The area under the receiver operating characteristic curve (AUC) was used to evaluate the model's discrimination of small cell lung cancer, and the Brier score was used to evaluate the calibration of the small cell lung cancer risk prediction probability. The model was recalibrated based on the probability of small cell lung cancer in the application scenario. A comprehensive evaluation of the small cell lung cancer risk assessment model was conducted in an independent test set. A comprehensive evaluation of the model in an independent test set can verify the generalization ability of the risk assessment model on unseen data, and can more realistically reflect the actual performance of the risk assessment model in the assessment of small cell cancer risk points, ensuring the reliability and stability of the model in practical applications.

[0093] Furthermore, step S2 also includes step S24, risk assessment report

[0094] The positioning results of the "risk point scanning" module below are used to extract the imaging indicators determined in "step S21, risk assessment marker selection", and the risk probability of small cell lung cancer can be calculated in combination with the small cell lung cancer risk factors reported by the patient (such as smoking, family history, chronic obstructive pulmonary disease), so as to generate a refined and individualized small cell lung cancer risk assessment report.

[0095] In step S24, not only does it rely on image data, but it also combines risk factors reported by patients (such as smoking, family history, chronic obstructive pulmonary disease, etc.) to achieve the fusion of multi-source data. This method can more comprehensively assess the patient's risk level and improve the accuracy and reliability of the assessment. In addition, step S24 can also perform refined risk assessment. The lesion position located by the "risk point scanning" module and combined with image indicators can be used to perform a refined risk assessment of the lesion area. This means that the embodiment of the present invention not only simply gives an overall risk probability, but also conducts a detailed analysis of each point where there may be a risk. Furthermore, the report generated by the embodiment of the present invention is for the specific situation of individual patients, rather than a general conclusion. This individualized report helps to provide patients with more targeted advice and treatment plans.

[0096] Step S3, risk point scanning

[0097] Step S31: Select risk warning signs

[0098] For the "risk warning" imaging markers, the AUC was calculated one by one based on the occurrence of new morbidity as the outcome, and the top 10 candidate markers with the highest AUC were selected to ensure the accuracy of the selected markers. The sensitivity (SEN) was further calculated under the same specificity (SPE), and the top 3 imaging markers with the highest SEN were selected for standard transformation, labeled as S1, S2, and S3 respectively, and Score=max(S) was defined for the identification of risk points.

[0099] For example, in one embodiment of the present invention, a total of 837 radiomics indicators are extracted, among which

[0100] There are 222 “risk warning” image signs (26.5%), including: 7 first-order grayscale features, 11 grayscale co-occurrence matrix features, 5 grayscale dependence matrix features, 3 grayscale run matrix features, 7 grayscale size area matrix features, 1 neighborhood grayscale difference matrix feature, and 188 wavelet transform features.

[0101] The “risk assessment” image markers mentioned in the previous step S15 are 56 (6.7%), including: 2 grayscale size area matrix features and 54 wavelet transform features.

[0102] In step S31, given the same specificity (SPE), the sensitivity (SEN) is further calculated, and the top three image markers with the highest SEN are selected. This method ensures that the false negative rate is minimized while the false positive rate is controlled, thereby improving the accuracy of risk point identification. In this way, the best balance point can be found, so that the risk assessment model can effectively identify the real risk points and reduce unnecessary alarms.

[0103] For example, in one embodiment of the present invention, three wavelet change features are selected from the "risk warning" image mark: w_LLLglcmImc2, w_LLLngtdmContrast, and w_LLLfirstorderEntropy, and the SENs of the risk points for judging the occurrence of small cell lung cancer in the future are all greater than 0.8.

[0104] Step S32: Segmentation of the whole lung anatomical parts

[0105] For fast segmentation of the whole lung structure, methods such as region growth and edge detection can be used to remove erroneous segmentation and fill in holes to ensure the accuracy and continuity of segmentation. Further, according to the common sites of small cell lung cancer (main bronchus and its branches, upper lobe tip, lower lobe base, hilar area and other areas), anatomical segmentation is performed on each site in turn, and specific methods such as semantic segmentation and template segmentation can be used.

[0106] In step S32, methods such as region growing and edge detection are used to quickly segment the whole lung structure. These methods can quickly locate different areas of the lungs, improve the speed and efficiency of segmentation, and maintain the accuracy of segmentation. By removing erroneous segmentation and filling holes, the accuracy and continuity of the segmentation results are ensured, the deviation caused by segmentation errors can be reduced, and the quality of the data can be improved. In addition, according to the common sites of small cell lung cancer, anatomical site segmentation is performed on each site, making full use of the anatomical distribution characteristics of small cell lung cancer, and potential lesion areas can be identified more specifically.

[0107] Step S33: multi-scale hierarchical detection

[0108] According to the order of the common sites of small cell lung cancer, each site is quickly scanned at a large scale and low resolution to find potential candidate areas, and then the scale is reduced and the resolution is increased to segment the block and calculate the score value. The threshold library at different scales is defined to temporarily store and update the maximum score value of different blocks at the same scale during the parallel calculation process until the risk point with the highest score value is detected.

[0109] In step S33, firstly, a rapid scan is performed on the high-incidence sites of small cell lung cancer at a large scale and low resolution to find potential candidate areas, so that the areas where problems may exist can be quickly locked, and then the resolution is gradually refined to a higher resolution for a more detailed analysis. This saves computing resources and improves the efficiency of detection. After the candidate areas are initially locked, the scale is reduced and the resolution is increased to perform block segmentation, and the Score value of each block is calculated, so that a more detailed analysis can be performed in the local area, thereby improving the accuracy of detection. By calculating the Score value, the risk level of the potential lesion area can be quantified. Threshold libraries at different scales are defined to temporarily store and update the maximum Score values ​​of different blocks at the same scale during parallel computing, which not only improves the computing speed, but also accelerates the entire detection process through parallel processing. The existence of the threshold library enables the system to dynamically adjust the detection standard to ensure the reliability of the detection results.

[0110] For example, in one embodiment of the present invention, the Score values ​​calculated based on the three wavelet change features w_LLLglcmImc2, w_LLLngtdmContrast, and w_LLLfirstorderEntropy both have an AUC and SEN of more than 0.8 for predicting the occurrence of small cell lung cancer at the risk point.

[0111] Step S34, risk point adjustment (optional)

[0112] On the basis of the above, different mechanisms can be introduced to achieve fine adjustment of risk points, such as block window shape design, fine step sliding, and adjustment of threshold library definition rules (for example, selecting the top 2-5 score values ​​to be included in the threshold library to avoid the scanning results falling into the local maximum position).

[0113] For example, by designing different block window forms, the characteristics of different parts can be better adapted. As an example, the window size and shape suitable for the area can be designed according to the different anatomical structures of the common sites of small cell lung cancer. This method can improve the flexibility and pertinence of detection, making the detection results more accurate. Alternatively, the fine step sliding method can be used to search for potential risk points more carefully. By gradually reducing the step size, more accurate scanning can be performed within the approximate area that has been determined, thereby avoiding missing any possible lesion areas. This method improves the accuracy of detection and reduces the possibility of false negatives. For another example, by adjusting the definition rules of the threshold library, the model can be prevented from falling into the local optimal solution, thereby ensuring the global optimality of the detection results.

[0114] Step S35: Risk point visualization

[0115] The detected risk points are displayed and recorded in the whole lung field of view and local field of view to guide follow-up image positioning or guide puncture operations.

[0116] In step S35, through the dual display of the whole lung field of view and the local field of view and detailed visual records, this technical solution is not only innovative in theory, but also has high value in practical applications. This method can not only improve the efficiency and accuracy of diagnosis, but also support the formulation of personalized treatment plans and provide reliable data support for long-term follow-up management and scientific research. Through these technical means, the accuracy and reliability of small cell lung cancer risk assessment can be significantly improved, providing strong technical support for clinical diagnosis.

[0117] This embodiment can fill the gap in the early identification and positioning technology of small cell lung cancer, especially for the risk assessment of 0-3mm small cell lung cancer. This technology advances the detection window of small cell lung cancer, which can improve the early detection rate, diagnosis rate and treatment rate of small cell lung cancer, improve the survival rate and survival period of patients, and thus change the situation that such patients cannot benefit from lung cancer screening.

[0118] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0119] Embodiment 2

[0120] According to an embodiment of the present invention, a small cell lung cancer evaluation system is provided. Figure 2-4 The system includes an image sign screening module (referred to as module 1), a comprehensive risk assessment module (referred to as module 2), and a risk point scanning module (referred to as module 3), among which:

[0121] The image mark screening module is specifically used for:

[0122] Selection of cases and controls: The imaging markers were screened using a case and control design, so that the imaging markers screened had the ability to sensitively detect and specifically identify small cell lung cancer; the controls included: small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control;

[0123] Pre- and post-image registration: For all included cases and controls, the CT examination that found the lesion was defined as R1, and the CT examination before the lesion was defined as R0. The whole chest image of R0 and R1 was rendered in three dimensions, and then R1 and R0 were registered.

[0124] Voxel of interest delineation: synchronize R0 and R1 lung windows, locate R1 lesions, cover lesion locations, generate and store voxel of interest masks;

[0125] Radiomics data extraction: Apply the mask to R0 to form a VOI for R0, and extract radiomics data for the lung texture where it is located;

[0126] Index evaluation and initial screening: The indicators were evaluated for stability, and the image features of cases and controls were tested for differentiation. Among them, the indicators with significant differences for small cell lung cancer self-control and nodule-free patient same-point control were defined as risk warning imaging markers for risk point identification; the indicators with significant differences for small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control were defined as risk assessment imaging markers for comprehensive modeling;

[0127] The comprehensive risk assessment module is specifically used for:

[0128] Risk assessment sign selection: remove highly relevant signs;

[0129] Risk assessment model training: Taking progression to small cell lung cancer as the outcome and carefully selected imaging markers as predictive factors, combined with known risk factors for small cell lung cancer, a small cell lung cancer risk assessment model was constructed;

[0130] Model evaluation and testing;

[0131] Risk assessment report: Using the positioning results of the risk point scanning module, extract the determined imaging indicators, calculate the risk probability of small cell lung cancer, and generate a small cell lung cancer risk assessment report;

[0132] The risk point scanning module is specifically used for:

[0133] Selection of risk warning signs: For risk warning imaging signs, the AUC is calculated one by one based on whether new diseases occur or not. The top n candidate signs with the highest AUC are selected, and the sensitivity is further calculated under the same specificity. The top m imaging signs with the highest sensitivity are selected, and standard transformation is performed. Score=max(S) is defined for the identification of risk points.

[0134] Segmentation of the entire lung’s anatomical parts;

[0135] Multi-scale hierarchical detection: Scan each site in the order of the common sites of small cell lung cancer to find potential candidate areas, calculate the score value, and define threshold libraries at different scales for temporary storage and update of the maximum score values ​​of different blocks at the same scale during parallel calculation until the risk point with the highest score value is detected.

[0136] Optionally, the specific examples in this embodiment can refer to the examples described in the above embodiments, and this embodiment will not be repeated here. Specifically, the execution process of calling module 1 can refer to the process of step S1 and each sub-step in step S1 in embodiment 1, the execution process of calling module 2 can refer to the process of step S12 and each sub-step in step S2 in embodiment 1, and the execution process of calling module 3 can refer to the process of step S3 and each sub-step in step S3 in embodiment 1.

[0137] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the contents disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can be run in a corresponding hardware environment, can be implemented by software, and can also be implemented by hardware, wherein the hardware environment includes a network environment.

[0138] Figure 5 is a structural block diagram of a terminal according to an embodiment of the present application, such as Figure 5 As shown, the terminal may include: one or more (only one is shown) processors 101, a memory 103, and a transmission device 105, such as Figure 5 As shown, the terminal may also include an input and output device 107 .

[0139] Among them, the memory 103 can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor 101 executes various functional applications and data processing by running the software programs and modules stored in the memory 103, that is, implementing the above-mentioned method. The memory 103 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 103 may further include a memory remotely arranged relative to the processor 101, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0140] The transmission device 105 is used to receive or send data via a network, and can also be used for data transmission between a processor and a memory. Specific examples of the above-mentioned network may include a wired network and a wireless network. In one example, the transmission device 105 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers via a network cable so as to communicate with the Internet or a local area network. In one example, the transmission device 105 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0141] Specifically, the memory 103 is used to store application programs.

[0142] The processor 101 may call the application program stored in the memory 103 through the transmission device 105 to execute each step of the method in the above-mentioned embodiment 1.

[0143] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.

[0144] Those skilled in the art will appreciate that the above terminal structures are for illustration only, and the terminals may be smart phones (such as Android phones, iOS phones, etc.), tablet computers, PDAs, mobile Internet devices (MID), PADs, and other terminal devices. Figure 5 It does not limit the structure of the above electronic device. For example, the terminal may also include Figure 5 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 5 Different configurations shown.

[0145] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for risk assessment of small cell lung cancer, characterized in that: include: Step S1, image marker screening; Step S2, comprehensive risk assessment; Step S3, risk point scanning; wherein: The step S1 specifically includes: Step S11, case and control selection: adopt case-control design to screen imaging markers, so that the screened imaging markers have the ability to discover and identify small cell lung cancer risk points; the controls include: small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, non-small cell lung cancer control; Step S12, front and back image registration: for all included cases and controls, define the CT examination in which the lesion is found as R1, and the CT examination before the lesion is found as R0, perform three-dimensional rendering of the whole chest image of R0 and R1, and then perform image registration of R1 and R0; Step S13, delineation of voxels of interest: synchronizing the R0 and R1 lung windows, locating the R1 lesion, covering the lesion position, generating and storing a mask of voxels of interest; Step S14, radiomics data extraction: applying the mask to R0 to form a VOI for R0, and extracting radiomics data for the lung texture where it is located; Step S15, indicator evaluation and preliminary screening: stability evaluation of the indicators, differential test of the image features of cases and controls, wherein the indicators with significant differences between the small cell lung cancer self-control and the nodule-free patient same-point control are defined as risk warning image markers for risk point identification; the indicators with significant differences between the small cell lung cancer self-control, the nodule-free patient same-point control, the benign nodule patient control, and the non-small cell lung cancer control are defined as risk assessment image markers for comprehensive modeling; The step S2 specifically includes: Step S21, risk assessment mark selection: high correlation marks are eliminated; Step S22, risk assessment model training: taking whether the patient progresses to small cell lung cancer as the outcome, taking the selected imaging markers as the predictive factors, and combining the known risk factors for small cell lung cancer, a small cell lung cancer risk assessment model is constructed; Step S23, model evaluation and testing; The step S3 specifically includes: Step S31, risk warning sign selection: for risk warning image signs, the AUC is calculated one by one based on whether a new disease change occurs as the outcome, and the top n candidate signs with the highest AUC are selected. The sensitivity under the same specificity is further calculated, and the top m image signs with the highest sensitivity are selected, and standard transformation is performed to define Score=max(S) for the identification of risk points; Step S32, segmentation of the entire lung anatomical parts; Step S33, multi-scale hierarchical detection: scan each site in the order of the common sites of small cell lung cancer, find potential candidate areas, calculate the score value, and define the threshold library at different scales for temporary storage and update of the maximum score value of different blocks at the same scale during the parallel calculation process, until the risk point with the highest score value is detected.

2. A risk assessment system for small cell lung cancer, characterized in that: include: Image sign screening module, comprehensive risk assessment module, risk point scanning module, including: The image mark screening module is specifically used for: Selection of cases and controls: The imaging markers were screened using a case and control design, so that the imaging markers screened had the ability to detect and identify small cell lung cancer; the controls included: small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control; Pre- and post-image registration: For all included cases and controls, the CT examination that found the lesion was defined as R1, and the CT examination before the lesion was defined as R0. The whole chest image of R0 and R1 was rendered in three dimensions, and then R1 and R0 were registered. Voxel of interest delineation: synchronize R0 and R1 lung windows, locate R1 lesions, cover lesion locations, generate and store voxel of interest masks; Radiomics data extraction: Apply the mask to R0 to form a VOI for R0, and extract radiomics data for the lung texture where it is located; Index evaluation and initial screening: The indicators were evaluated for stability, and the image features of cases and controls were tested for differentiation. Among them, the indicators with significant differences for small cell lung cancer self-control and nodule-free patient same-point control were defined as risk warning imaging markers for risk point identification; the indicators with significant differences for small cell lung cancer self-control, nodule-free patient same-point control, benign nodule patient control, and non-small cell lung cancer control were defined as risk assessment imaging markers for comprehensive modeling; The comprehensive risk assessment module is specifically used for: Risk assessment sign selection: remove highly relevant signs; Risk assessment model training: Taking progression to small cell lung cancer as the outcome and carefully selected imaging markers as predictive factors, combined with known risk factors for small cell lung cancer, a small cell lung cancer risk assessment model was constructed; Model evaluation and testing; Risk assessment report: Using the positioning results of the risk point scanning module, extract the determined imaging indicators, calculate the risk probability of small cell lung cancer, and generate a small cell lung cancer risk assessment report; The risk point scanning module is specifically used for: Selection of risk warning signs: For risk warning imaging signs, the AUC is calculated one by one based on whether new diseases occur or not. The top n candidate signs with the highest AUC are selected, and the sensitivity is further calculated under the same specificity. The top m imaging signs with the highest sensitivity are selected, and standard transformation is performed. Score=max(S) is defined for the identification of risk points. Segmentation of the entire lung’s anatomical parts; Multi-scale hierarchical detection: Scan each site in the order of the common sites of small cell lung cancer to find potential candidate areas, calculate the score value, and define threshold libraries at different scales for temporary storage and update of the maximum score values ​​of different blocks at the same scale during parallel calculation until the risk point with the highest score value is detected.

3. The risk assessment system for small cell lung cancer according to claim 2, characterized in that: The image registration of R1 and R0 specifically includes: The whole chest images of R0 and R1 were rendered in three dimensions. The sternal angle, scapula and ribs of the R0 image were used as landmarks, and the R1 image was translated and / or angled in the sagittal, coronal and axial planes until the landmarks overlapped.

4. The risk assessment system for small cell lung cancer according to claim 2, characterized in that: The model evaluation and testing specifically include: evaluating the model's discrimination for small cell lung cancer using the area under the receiver operating characteristic curve, evaluating the calibration of the small cell lung cancer risk prediction probability using the Brier score, recalibrating the model based on the probability of small cell lung cancer in the application scenario, and comprehensively evaluating the small cell lung cancer risk assessment model in an independent test set.

5. The risk assessment system for small cell lung cancer according to claim 2, characterized in that: The whole lung anatomical part segmentation specifically includes: removing erroneous segmentation, filling in holes, ensuring the accuracy and continuity of segmentation, and further performing anatomical part segmentation on each part in turn according to the common sites of small cell lung cancer.

6. The risk assessment system for small cell lung cancer according to claim 2, characterized in that: The scanning of each part specifically includes: scanning each part in turn quickly at a large scale and low resolution to find potential candidate areas, and then reducing the scale and increasing the resolution to perform block segmentation.

7. The risk assessment system for small cell lung cancer according to claim 2, characterized in that: The risk point scanning module is also used for: Risk point adjustment, the adjustment mechanism includes block window shape design, fine step sliding, and adjustment of threshold library definition rules.

8. The risk assessment system for small cell lung cancer according to claim 2, characterized in that: The risk point scanning module is also used for: Visualization of risk points: The detected risk points are displayed and recorded in the whole lung field of view and local field of view to guide the follow-up image positioning or guide the puncture operation.

9. An electronic device, characterized in that: include: Processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in the method according to claim 1 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the method according to claim 1.

Images