Biomarkers based on imaging features of tumor tissue perfusion

Abstract

The present application relates to the field of tumor biomarkers, in particular to a biomarker based on tumor tissue perfusion imaging features. The present application constructs a biomarker OMP based on tumor tissue perfusion imaging features, which is obtained by segmenting the tumor enhanced image in a "peeling onion" manner and after data analysis with specific parameter setting. The biomarker OMP of the present application is used for immune-related treatment effect prediction, fully embodies the influence of tumor heterogeneity on tumor diagnosis and treatment, and has the advantages of simple operation, reliable diagnosis, safety and non-invasiveness, strong repeatability and the like. It can be dynamically evaluated with patient follow-up imaging data, and as a new marker, it has a wide range of application population and high clinical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of tumor biomarkers, in particular to a biomarker based on tumor tissue perfusion imaging features. BACKGROUND

[0002] The breakthrough of tumor immunotherapy, especially immune checkpoint inhibitors and CAR-T cell therapy, brings hope for human to cure malignant tumors, but at present, tumor immunotherapy only makes a small part of patients benefit for a long time. Therefore, the low efficiency has always plagued clinicians, and finding reliable biomarkers has always been a research hotspot in tumor immunotherapy. The existing biomarkers for predicting the effect of tumor immunotherapy mainly include: PD-L1 expression (TPS, CPS), tumor mutation load (TMB), microsatellite instability (MSI-H / dMMR), driver gene status, DNA damage repair related gene mutation, etc. Among them, the expression of tumor cell PD-L1-tumor cell positive score (TPS, calculated as any intensity PD-L1 membrane staining tumor cells / tumor cells total number x 100%) is the only biomarker approved by the clinic and recommended by the NCCN (National Comprehensive Cancer Network) guideline for predicting the effect of non-small cell lung cancer immunotherapy. But its limitations are also very obvious. First, the prediction effect is not ideal when PD-L1 is used alone. Second, some patients cannot be detected by PD-L1. The expression of tumor cell PD-L1 is a traumatic method based on tumor biopsy, which has high requirements for the physical condition of patients and the growth of tumors. Patients with poor physical condition, coagulation dysfunction or other comorbidities may not be able to undergo biopsy, and tumors close to large blood vessels or other important organs, or blocked by puncture approach, etc. may also lead to patients unable to undergo biopsy. Third, the malignant tumor biopsy cannot dynamically evaluate the index. Therefore, finding a non-invasive real-time monitoring marker is an urgent need to optimize tumor immunotherapy.

[0003] Enhanced CT is an imaging method widely used in clinic to detect tumor size, blood perfusion function, etc. It is an imaging examination method that increases the density difference between the lesion and the adjacent normal tissue by injecting water-soluble iodine contrast agent into the blood vessels and then performing CT scanning, so as to increase the iodine concentration in the blood and improve the display rate of the lesion. Therefore, enhanced CT is a routine examination method for patients with lung cancer and other tumors during follow-up, and the data availability is high and the data stability is good. There is no effective method for predicting the effect of immunotherapy in traditional imaging. In recent years, the emerging imageomics method extracts tumor CT image features and makes an immunotherapy effect prediction model, but the prediction effect is not good. These methods mainly analyze the correlation of image features, gene features, tumor immune features and heterogeneity features, but the internal tumor biological logic relationship of these connections is not clear. Moreover, the existing imageomics prediction model directly uses the features extracted from the enhanced CT, ignoring the typical feature of the high heterogeneity of the tumor itself, and not considering the temporal and spatial dynamic distribution of the contrast agent, which is a typical feature of enhanced CT. Such analysis method applied to tumor prediction is prone to large bias. Therefore, it has great practical significance and clinical application value to establish a biomarker that truly reflects the perfusion imaging features of tumor tissue. SUMMARY

[0004] Therefore, the technical problem to be solved by the present application is to provide a biomarker based on the perfusion imaging features of tumor tissue.

[0005] The present application provides an OMP marker, which is the sum of the single-layer relative CT average value or the multi-layer relative CT average value after normalizing the relative CT average value of each layer of the circle layer segmented image to the relative CT average value of the zero layer.

[0006] In the present application, the single-layer relative CT average value refers to any single-layer relative CT average value in the circle layer segmented image, which can be a first-layer relative CT average value, a second-layer relative CT average value, a third-layer relative CT average value, a fourth-layer relative CT average value, a fifth-layer relative CT average value, or a sixth-layer relative CT average value, etc. The arbitrary value is related to the number of layers of the circle layer segmented image, which is any value in the number of layers of the circle layer segmented image.

[0007] In the present application, the sum of the multi-layer relative CT average value includes but is not limited to the sum of 1-2 layers, the sum of 1-3 layers, the sum of 3-4 layers, the sum of 2-4 layers, the sum of 3 layers and 7 layers, or the sum of 1, 4, 5 layers, etc. The sum of the relative CT average values of any two or more layers in the circle layer segmented image.

[0008] In the present application, the sum of the multi-layer relative CT average value also includes the combination of the sum of any single-layer relative CT average value and any multi-layer relative CT average value.

[0009] Further, in some embodiments of the present application, the OMP value is the single-layer relative CT average value of the ring layer segmented image normalized by the relative CT average value of the zeroth layer.

[0010] In other embodiments, the OMP value of the present application is the sum of the first-layer relative CT average value and the second-layer relative CT average value of the ring layer segmented image normalized by the relative CT average value of the zeroth layer.

[0011] In other embodiments, the OMP value of the present application is the sum of the first-layer relative CT average value, the second-layer relative CT average value, and the third-layer relative CT average value of the ring layer segmented image normalized by the relative CT average value of the zeroth layer.

[0012] In other embodiments, the OMP value of the present application is the sum of the first-layer relative CT average value, the second-layer relative CT average value, the third-layer relative CT average value, and the fourth-layer relative CT average value of the ring layer segmented image normalized by the relative CT average value of the zeroth layer.

[0013] In the present application, the OMP marker can be used for predicting the effect of tumor immunotherapy. After numerous attempts, it is found that when the sum of the first-layer relative CT average value and the second-layer relative CT average value, the sum of the first to third-layer relative CT average values, or the sum of the first to fourth-layer relative CT average values is used as the OMP value after normalization by the relative CT average value of the zeroth layer, the effect of tumor immunotherapy can be predicted, but the sum of the first-layer relative CT average value and the second-layer relative CT average value is the best OMP for predicting the effect of tumor immunotherapy. In some specific embodiments of the present application, the predictive power of OMP can reach 0.791, indicating that about 79% of patients sensitive to treatment can be screened out, who will benefit from subsequent treatment.

[0014] The "tumor immunotherapy" of the present application refers to tumor immunotherapy or combined therapy of tumor immunotherapy with other tumor treatment methods. Other tumor treatment methods include but are not limited to chemotherapy, targeted therapy, anti-angiogenic therapy, radiotherapy, interventional therapy, nanodrug therapy, etc.

[0015] The OMP marker of the present application for predicting the effect of tumor immunotherapy is to obtain an OMP cutoff value by analyzing the correlation between the OMP value of the tumor imaging ring layer perfusion level and the clinical treatment response, and this OMP cutoff value can be applied to predict the effect of tumor immunotherapy. In some specific embodiments of the present application, the OMP cutoff value obtained by analyzing the correlation between the sum of the first-layer relative CT average value and the second-layer relative CT average value as OMP and the clinical treatment response is 0.32. If the obtained OMP value is greater than this cutoff value, it can be used to predict the effect of lung cancer immunotherapy, and the prediction result is accurate.

[0016] Further, the tumor includes a primary tumor and a metastatic tumor thereof.

[0017] Still further, the tumor includes at least one of lung cancer, breast cancer, liver cancer, esophageal cancer, colorectal cancer, pancreatic cancer, head and neck tumor, central nervous system tumor, gynecological tumor, and soft tissue sarcoma.

[0018] In the present application, the relative CT average value (Mean) is obtained by dividing the sum of the relative CT values of each voxel by the number of voxels. In some embodiments of the present application, the voxel size is 1mmx1mmx1mm, and other specific sizes can be obtained by resampling, which are well known to those skilled in the art.

[0019] The relative CT value is obtained by Min-max normalization of the original CT value, and the Min-max normalization formula is: Value normalization=(Value origin-Value min) / (Value max-Value min); wherein Value origin is the original CT value of the segmented image; Value max is the maximum original CT value in all segmented images; Value min is the minimum original CT value in all segmented images.

[0020] Further, the original CT value in the present application is the CT value (or gray value) of a single voxel. In some embodiments of the present application, a single voxel is resampled to 1mm 3 size, and the original CT value is well known to those skilled in the art.

[0021] In the present application, the relative CT average value normalization calculation formula of the zero layer is: OMP1=Mean1-Mean0, OMP2=Mean2-Mean0, OMP3=Mean3-Mean0, …; wherein OMP is the circle layer perfusion value, OMP1, OMP2, OMP3 are the circle layer perfusion values of the first, second and third layers of the tumor; Mean is the relative CT average value, Mean1, Mean2, Mean3 are the relative CT average values of the first, second and third layers of the tumor; and the rest is similar. Mean0 is the relative CT average value of the zero layer (baseline layer, i.e. the outermost layer of the tumor).

[0022] In the present application, the CT value refers to the different densities shown by the CT image under a specified acquisition condition. A higher CT value shows a larger density in the image, and a lower CT value shows a lower density. For CT examination of a tumor, enhanced CT examination is usually performed. Since the tumor has a relatively rich blood supply and some abnormal vascular function, the tumor body can be seen to have obvious diversity of enhancement in angiographic enhanced CT examination.

[0023] In the OMP marker of the application, the method for obtaining the ring layer segmentation image is: after resampling the enhanced CT image of the tumor solid part, the volume of interest is outlined to obtain a segmentation image, and then the ring layer segmentation image is obtained by layer-by-layer segmentation in the manner of peeling onions from the outer layer to the inner layer.

[0024] In the method for obtaining the ring layer segmentation image, the resampling method of the enhanced CT image includes but is not limited to any one of linear interpolation, nonlinear interpolation, nearest neighbor, bspline, hamming, cosine, welch, lanczos and blackman. The linear interpolation includes nearest neighbor interpolation, bilinear interpolation and bicubic interpolation. The use of the linear interpolation method for image resampling is conducive to clear imaging of the tumor solid part image. In the application, the bilinear interpolation method is used for resampling the enhanced CT image, and the image is collected to the same size, thereby facilitating analysis and comparison. In some specific embodiments of the application, the resampling size can be 1mm*1mm*1mm, or any other biologically meaningful numerical combination size. The original CT value of the tumor solid part at this time is recorded for OMP marker calculation.

[0025] In the method for obtaining the ring layer segmentation image, the outlining is performed in a combination of manual outlining and automatic outlining. The outlining method mainly uses automatic outlining and is supplemented by manual outlining, which can reduce errors in data analysis.

[0026] The selection of the size of the outlining threshold range is closely related to the accurate positioning of the tumor target area, and different tumors have different outlining threshold values. At present, there is no standard standard for the outlining threshold value of the tumor solid area. In some specific embodiments of the application, when the cancer is lung cancer, the automatic outlining threshold value is-50HU-200HU, which is a relatively accurate outlining threshold value for lung cancer obtained after numerous attempts.

[0027] Tumor segmentation is an important content in the field of medical image analysis. Due to the large differences in shape, texture and other reasons between individuals, tumor segmentation is difficult. Ordinary tumor ring layer segmentation often uses flat layer segmentation, which is difficult to obtain all the characteristics of the tumor lesion. In some failed embodiments of the application, the flat layer segmentation method is used to analyze the images of immune therapy with and without response, and no difference is obtained. In other embodiments of the application, the entire tumor is analyzed layer by layer from the outer layer to the inner layer in a three-dimensional angle like peeling onions, and a statistically significant difference is obtained.

[0028] Further, in order to make the difference be embodied by specific numerical values and applied to immune-related treatment effect prediction, the application makes the following attempts:

[0029] The segmented images are analyzed by the correlation coefficient of the relative gray average broken line and the line Y=X, and the prediction effect has significant difference, but the prediction effect is not ideal;

[0030] The 0th layer is removed, the segmented images are analyzed by the correlation coefficient of the broken line with downward trend and the line Y=8-X, and the prediction effect is not ideal;

[0031] The 0th layer is removed, the segmented images are analyzed by the fitting slope k, and the prediction effect is not ideal;

[0032] The segmented images are analyzed by the sum of the relative gray values of the circle layers, and the results show that the sum of the relative CT values of the first layer and the second layer after the baseline normalization of the 0th layer segmented image is the highest prediction effect as the OMP value.

[0033] In the method for obtaining the segmented images of the circle layers, the thickness of the layer-by-layer segmentation is 0.5mm-10mm. The value can be determined according to observation focus, image resolution, biological interpretability and other factors. In some specific embodiments, the thickness of the layer-by-layer segmentation is 1mm.

[0034] In the OMP marker, the tumor image includes any one of lung cancer image, breast cancer image, liver cancer image, esophageal cancer image, colorectal cancer image, pancreatic cancer image, head and neck tumor image, central nervous system tumor image, gynecological tumor image, soft tissue sarcoma and metastatic tumor image.

[0035] The application provides a combined tumor marker, which comprises a tumor marker and the OMP marker.

[0036] Further, the tumor marker comprises a tumor biomarker and a tumor specific index; the tumor biomarker includes but is not limited to programmed death ligand 1 (PD-L1), programmed death receptor 1 (PD1), PSA, fPSA, alpha-fetoprotein, carcinoembryonic antigen, carbohydrate antigen 199, cancer antigen 125, tumor antigen 153, cancer antigen 50, carbohydrate antigen 242, serum ferritin, neuron-specific enolase, human chorionic gonadotropin, tumor necrosis factor, squamous cell carcinoma-related antigen, and an immune response-related index; the immune response-related index includes but is not limited to CD8 + T cells, CD4 +T cells, NK cells, Treg cells, dendritic cells, interferon gamma, granzyme, perforin, Foxp3, CXCL9, CXCL10, CXCL11, CCL8, TIGIT, ICOS, ICOSL, IFNGR1, IFNGR2, STAT1, STAT3, JAK1, JAK2, c-Cb1, CrKL, CrKII, Vav, C3G, Rap-1, IRF, IRF8, IRF9, ISGF3, antigen processing-related transporter TAP1, antigen processing-related transporter TAP2, MHC class II transcription activator (CIITA), beta 2-microglobulin, and the like; the tumor-specific indicators include, but are not limited to, high microsatellite instability (MSI-H), mismatch repair deficiency (dMMR), tumor mutation burden (TMB), tumor size, tumor load, intestinal flora, metabolomics, genomics, imagingomics, and the like. The present application does not make any limitation thereto.

[0037] Further, the tumor biomarker is selected from the group consisting of programmed death ligand 1 (PD-L1), programmed death receptor 1 (PD1), tumor mutation burden (TMB), high microsatellite instability (MSI-H), mismatch repair deficiency (dMMR), tumor size, tumor load, CD8 + T cells, CD4 + T cells, NK cells, Treg cells, dendritic cells, interferon gamma, granzyme, perforin, Foxp3, CXCL9, CXCL10, CXCL11, CCL8, TIGIT, ICOS, ICOSL, IFNGR1, IFNGR2, STAT1, STAT3, JAK1, JAK2, c-Cb1, CrKL, CrKII, Vav, C3G, Rap-1, IRF, IRF8, IRF9, ISGF3, antigen processing-related transporter TAP1, antigen processing-related transporter TAP2, MHC class II transcription activator (CIITA), beta 2-microglobulin, PSA, fPSA, alpha-fetoprotein, carcinoembryonic antigen, carbohydrate antigen 199, cancer antigen 125, tumor antigen 153, cancer antigen 50, carbohydrate antigen 242, beta 2-microglobulin, serum ferritin, neuron-specific enolase, human chorionic gonadotropin, and the like.

[0038] In the present application, the combination of tumor markers can be used for predicting the effect of tumor immunotherapy, and the combination of tumor biomarkers and OMP markers can further improve the prediction efficiency of tumor immunotherapy. In some specific embodiments, the combination of OMP markers and PD-L1 in the present application can achieve a prediction efficiency of 0.801 in predicting the effect of tumor immunotherapy, which means that about 80% of patients with treatment sensitivity can be screened out and will benefit from subsequent treatment. In the real world, the treatment-sensitive patients account for about 30% of lung cancer patients, and some tumor types are less than 10%. Therefore, the prediction method of the present application can benefit more patients and allow at least 50% of patients to avoid choosing a poor treatment plan as soon as possible.

[0039] The present application provides a method for constructing a tumor clinical prediction model, which comprises obtaining the OMP marker or the combination of tumor markers described in the present application to construct a prediction model. The method for constructing a tumor imaging prediction model provided by the present application requires analysis of imaging features. The imaging feature analysis mainly includes image data acquisition, image segmentation and reconstruction, feature extraction and screening, clinical model establishment and data information analysis. The imaging features are greatly affected by the acquisition device and acquisition parameters of the image acquisition, and the research results need to be verified by a large sample, therefore, the standardization of data, the repeatability and reliability of algorithm are crucial.

[0040] In the construction method of the present application, the tumor is at least one of lung cancer, breast cancer, liver cancer, esophageal cancer, colorectal cancer, pancreatic cancer, head and neck tumor, central nervous system tumor, gynecological tumor, soft tissue sarcoma and metastatic tumor.

[0041] The present application provides any one of I) to III) as follows in the application of tumor imaging, imaging, preparation of tumor efficacy prediction product or tumor clinical prediction model:

[0042] I) the OMP marker of the present application;

[0043] II) the combination of tumor markers of the present application;

[0044] III) the construction method of the present application.

[0045] In the present application, the tumor imaging refers to high-throughput extraction of a large amount of image information from images (CT, MRI, B-ultrasound, etc.), realization of tumor segmentation, feature extraction and model establishment, which assists physicians to make more reliable efficacy prediction through image data mining and analysis. The tumor images include, but are not limited to, any one of lung cancer images, breast cancer images, liver cancer images, esophageal cancer images, colorectal cancer images, pancreatic cancer images, head and neck tumor images, central nervous system tumor images, gynecological tumor images, soft tissue sarcoma images and metastatic tumor images of primary tumors.

[0046] The present application uses the circle layer perfusion level OMP as the tumor tissue perfusion imaging feature for the prediction of the efficacy of immunotherapy in the tumor imaging, the preparation of the tumor efficacy prediction product or the imageomics prediction model, which is obtained through numerous attempts. In some specific embodiments of the present application, the tumor CT shadow, the tumor CT texture feature, the single enhanced CT texture feature and the single enhanced CT gray scale composition are used as the tumor tissue perfusion imaging feature for the prediction of the efficacy of immunotherapy, and the results show that the above cannot realize the effective prediction of the efficacy of immunotherapy.

[0047] The OMP biomarker of the present application can be used for the prediction of the efficacy of immunotherapy alone, or in combination with other biomarkers and / or specific tumor indicators.

[0048] The OMP biomarker of the present application can also be used in tumor images generated by different imaging methods or different types of instruments and equipment for the prediction of the efficacy of immunotherapy.

[0049] The present application constructs the biomarker OMP based on the tumor tissue perfusion imaging feature, which is obtained by circle layer segmentation of the tumor enhanced CT image in the manner of "peeling onions" and data analysis with specific parameter setting. The biomarker OMP of the present application is used for the prediction of the efficacy of immunotherapy, fully reflects the influence of tumor heterogeneity on tumor diagnosis and treatment, and has the advantages of simple operation, reliable diagnosis, safety and non-invasiveness, strong repeatability, can be dynamically evaluated with patient follow-up image data, has a wide application population as a new biomarker, and has high clinical application value. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A circle layer segmentation schematic diagram of a tumor tissue enhanced CT image of a lung cancer patient is shown;

[0051] Figure 2 In lung cancer patients, the circle layer perfusion level (OMP) of the effective group of immunotherapy is significantly higher than that of the ineffective group;

[0052] Figure 3 The cut-off value of OMP was 0.32 found by the maximum Youden index in the lung cancer receiver operating characteristic (ROC) curve;

[0053] Figure 4 The individual predictive efficacy of OMP was up to 0.772 in the efficacy prediction model;

[0054] Figure 5 The individual predictive efficacy of OMP was higher than the approved biomarker PD-L1 TPS in the efficacy prediction model;

[0055] Figure 6 The individual predictive efficacy of OMP was up to 0.791 in patients with negative PD-L1 expression in tumor cells (TPS < 50%);

[0056] Figure 7 OMP can increase the net benefit of treatment by 25% in the patient population with negative PD-L1 expression;

[0057] Figure 8 The predictive efficacy of OMP combined with PD-L1 was up to 0.801 in the efficacy prediction model;

[0058] Figure 9 The combination of OMP and PD-L1 prediction greatly increased the proportion of net benefit of treatment;

[0059] Figure 10 The tumor blood perfusion function was detected by immunohistochemical method in mouse breast cancer;

[0060] Figure 11 The tumor blood perfusion function was detected by color Doppler ultrasound in mouse breast cancer;

[0061] Figure 12 The tumor blood perfusion function was detected by ultrasound microbubble contrast in mouse lung cancer;

[0062] Figure 13 The plain scan, enhanced CT and shadow images of liver metastases were shown;

[0063] Figure 14 The changes of texture features of liver metastases before and after enhancement were shown, wherein the X-axis was a specific case and the Y-axis was a specific feature;

[0064] Figure 15 The gray density histogram of part of lung cancer patients was shown, wherein the blue was the immune therapy responsive tumor and the red was the immune therapy non-responsive tumor; the X-axis was the relative CT value and the Y-axis was the proportion of the value in the total;

[0065] Figure 16Figure 6 shows the relative CT value ranking plot of some lung cancer patients, wherein blue represents immunotherapy responsive tumors, and red represents immunotherapy non-responsive tumors; the X-axis represents the percentile, and the Y-axis represents the relative CT value;

[0066] Figure 17 Figure 7 shows the gray density histogram of lung cancer, wherein blue represents immunotherapy responsive tumors, and red represents immunotherapy non-responsive tumors; the X-axis represents the relative CT value, and the Y-axis represents the proportion of the value in the total number;

[0067] Figure 18 Figure 8 shows the relative CT value ranking plot of lung cancer, wherein blue represents immunotherapy responsive tumors, and red represents immunotherapy non-responsive tumors; the X-axis represents the percentile, and the Y-axis represents the relative CT value;

[0068] Figure 19 Figure 9 shows the average value line graph of the OMP of the ring layer perfusion level of lung cancer, and the responsive group is higher than the non-responsive group;

[0069] Figure 20 Figure 10 shows the ROC curve of the correlation coefficient model of the ring layer gray line and the line Y=X;

[0070] Figure 21 Figure 11 shows the ROC curve of the correlation coefficient model of the ring layer gray line and the line Y=8-X;

[0071] Figure 22 Figure 12 shows the ROC curve of the fitting slope model of the ring layer gray line;

[0072] Figure 23 Figure 13 shows the ROC curve of the model constructed by the sum of the first to fourth layers;

[0073] Figure 24 Figure 14 shows the ROC curve of the model constructed by the sum of the first to third layers;

[0074] Figure 25 Figure 15 shows the ROC curve of the model constructed by the sum of the first to second layers. DETAILED DESCRIPTION

[0075] The present application provides a biomarker based on tumor tissue perfusion imaging features, and those skilled in the art can refer to the content herein to appropriately improve the process parameters. It should be particularly pointed out that all similar substitutions and changes are obvious to those skilled in the art, and they are all considered to be included in the present application. The method and application of the present application have been described by preferred embodiments, and relevant personnel can obviously modify or appropriately change and combine the method and application herein without departing from the content, spirit and scope of the present application, to realize and apply the present application technology.

[0076] The term "recipient," "patient" as used herein refers to any mammal or non-mammal. Mammals include, but are not limited to, humans, non-human primates, bovines, equines, canines, felines, porcines, ovines, caprines, murines.

[0077] The materials used in the present application are all ordinary commercially available products, which can be purchased in the market.

[0078] The present application is further described below in conjunction with examples:

[0079] Example 1: Establishment of biomarker OMP

[0080] 1. Establishment of biomarker OMP

[0081] The current study proposes the hypothesis that tumors that respond to immunotherapy have better perfusion of blood vessels, but how to use the perfusion characteristics of tumor blood vessels to predict the effect of tumor immunotherapy-related treatment and how to adjust the data parameters to obtain a prediction model that can accurately predict the effect of immunotherapy-related treatment still need to be explored. The present application establishes a biomarker that can effectively predict the effect of tumor immunotherapy-related treatment through analysis and exploration of a large number of clinical samples. The specific construction process is as follows:

[0082] In a non-small cell lung cancer immunotherapy-related treatment research cohort conducted by us in Chongqing Xinqiao Hospital, we screened out tumors with a solid part short diameter greater than 1 cm, clear and artifact-free arterial phase enhancement image data, and clear efficacy (n = 99) as research objects. We divided cases without PD-L1 expression into a training set (n = 39) and cases with PD-L1 expression into a validation set (n = 60). Through resampling, circle layer segmentation ( Figure 1 ), image data extraction, Min-Max normalization, baseline normalization, and other operations, the blood perfusion characteristics of non-small cell lung cancer tissue were extracted and named as: circle layer perfusion (Onion-mode perfusion, OMP). The data analysis results showed that the circle layer perfusion level (OMP) of the effective group of immunotherapy was significantly higher than that of the ineffective group ( Figure 2 , P = 0.0392, t test).

[0083] We made a receiver operating characteristic (ROC) curve in the training set through the OMP feature, and found the cut off value of OMP to be 0.3214 ( Figure 3 ) through the maximum Youden Index. The OMP was classified into high OMP and low OMP based on the cut off value, and the efficacy prediction model constructed by the two classification indexes ( Figure 4The area under the curve (AUC) on the training set was 0.649 (95% CI: 0.501–0.779, P = 0.013), and the AUC on the validation set was 0.772 (95% CI: 0.621–0.885, P < 0.0001). Furthermore, the predictive power of OMP on the validation set (AUC = 0.772, 95% CI: 0.621–0.885) was higher than that of the approved PD-L1TPS (AUC = 0.666, 95% CI: 0.508–0.801). Figure 5 Furthermore, OMP effectively screened out patients who responded to immunotherapy in tumor cells with negative PD-L1 expression (TPS < 50%, i.e., TPS indicates poor response to immunotherapy) (AUC = 0.791, 95% CI: 0.597–0.921, P = 0.0003). Figure 6 We used OMP to construct a clinical decision curve (DCA) in a PD-L1-negative patient population and found that administering immunotherapy based on OMP-suggested recommendations resulted in a net treatment benefit of up to 25% without compromising other patient benefits. Figure 7 Furthermore, we used logistic regression to build a joint model of OMP and TPS on the validation set. The AUC of the joint efficacy prediction model reached 0.801 (95% CI: 0.692–0.929, P<0.0001). Figure 8 The clinical benefits are also more obvious. Figure 9 These findings indicate that OMP is an effective, reliable, and non-invasive predictive biomarker for immune-related treatments. This method is simple to perform, provides reliable predictions, is safe, non-invasive, highly reproducible, and can be dynamically assessed alongside patient follow-up imaging data.

[0084] Furthermore, in mouse models of breast cancer, colon cancer, and lung cancer, immunofluorescence imaging (IFI) has shown significant efficacy. Figure 10 ), color Doppler ultrasound ( Figure 11 ) and ultrasound microbubble contrast imaging ( Figure 12 The results all showed that tumors that responded to immunotherapy had better blood perfusion.

[0085] The above results confirm the hypothesis that the degree of tumor tissue perfusion can predict the efficacy of immunotherapy, and through the analysis of non-small cell lung cancer clinical imaging data modeling, the biomarker OMP is obtained, which can effectively predict the efficacy of tumor immunotherapy. In addition, in clinical practice, using specific analysis methods to analyze different image data such as CT, MRI, ultrasound, and nuclear imaging of lung cancer, breast cancer, liver cancer, esophageal cancer, colorectal cancer, pancreatic cancer, head and neck tumors, central nervous system tumors, gynecological tumors, and soft tissue sarcomas, the perfusion characteristics OMP of the tumor can be obtained to stratify cancer patients and predict the efficacy of immunotherapy-related treatment, significantly improving the response rate of tumor immunotherapy-related treatment.2、OMP biomarker application scheme

[0086] 2.1 Extract tumor tissue perfusion characteristics using CT built-in segmentation software

[0087] (1) Perform chest enhanced CT scanning on non-small cell lung cancer patients who are considering immunotherapy. The CT model is not limited, and the image data collection is performed in accordance with the Chinese Health Industry Standard-CT Examination Operation Procedures (WST391-2012). The scanning position is supine position, the body is placed in the middle of the bed surface, the arms are raised, and the scan is performed at the end of inspiration. Scanning conditions: 120kV, automatic mA (100-300mA), reconstruction layer thickness ≤5mm. Total amount of iodine injection: 300-450mg iodine per kilogram of body weight, jet speed 3-4ML / s, start scanning time after injection: 20-30s in arterial phase.

[0088] (2) Use the CT built-in analysis software to resample the patient's CT Dicom data to 1mm x 1mm x 1mm using linear interpolation.

[0089] (3) Draw the volume of interest (VOI), draw the solid part of the lung tumor, and set the drawing threshold Threshold = -50-200HU to achieve semi-automatic drawing and improve the stability of the operation.

[0090] (4) Circle layer segmentation, use the three-dimensional shrinkage function in the CT built-in analysis software to "peel onions" from the outermost layer, and uniformly segment inward layer by layer, with each layer thickness of 1mm (i.e. one voxel thickness).

[0091] (5) Perform Min-max normalization on the data within the VOI, using the formula: Value_normalization = (Value_origin - Value_min) / (Value_max - Value_min). Here, value_origin is the original CT value at a certain location within the target area, and value_max and value_min are the maximum and minimum original CT values ​​within that target area, respectively. This operation mainly normalizes the relative difference between the maximum and minimum values ​​in the target area to a range of 0 to 1, thus reducing the inconsistency in CT value ranges between different tumors. Calculate the arithmetic mean of Value_normalization for each layer in the layered segmentation. After normalization using the zeroth layer as the baseline, the sum of the relative CT averages of the first and second layers is taken as the OMP.

[0092] (6) According to the current diagnostic criteria, the cutoff value for OMP is 0.32. If the OMP value of the case is greater than 0.32, it is considered as high perfusion, and the possibility of effective immunotherapy is high. Immunotherapy is recommended. Conversely, if the OMP value is less than 0.32, it is considered as low perfusion, and the effectiveness of immunotherapy is low. Immunotherapy is not recommended.

[0093] (7) Regardless of the assessment results, if the patient undergoes immunotherapy, routine follow-up should be conducted, the efficacy should be recorded and reported, and the model should be continuously improved.

[0094] 2.2 Using third-party software in conjunction with Python to extract tumor tissue perfusion characteristics

[0095] (1) For non-small cell lung cancer patients considered for immunotherapy, enhanced chest CT scans were performed. The CT scanner model was not limited. Image acquisition was performed according to the Chinese Health Industry Standard - CT Examination Operating Procedures (WST391-2012). The patient position was supine with the body in the middle of the bed and arms raised. Breath-holding was performed at the end of inspiration. Scanning conditions were 120kV, automatic mA (100-300mA), and reconstruction slice thickness ≤5mm. The total iodine injection volume was 300-450mg iodine / kg body weight, with a jet rate of 3-4ML / s. The scan started 20-30s after injection during the arterial phase.

[0096] (2) Import the patient's CT Dicom data into 3DSlicer (version 4.11.0) and resample the image data to 1mm×1mm×1mm using linear interpolation.

[0097] (3) Delineate the volume of interest (VOI). Use 3DSlicer (version 4.11.0) to delineate the solid part of the lung tumor. By setting the delineation threshold Threshold = -50 to 200HU, semi-automatic delineation can be achieved, improving the stability of the operation.

[0098] (4), ring layer segmentation, using 3DSlicer (version 4.11.0), in the segmentation module, using the shrink operation in the margin function to "peel onions" VOI, starting from the outermost layer, evenly segmented layer by layer inward, each layer thickness is 1mm (i.e. one voxel thickness).

[0099] (5), based on python (version 3.7.4) using SimpleITK package (Version 1.2.4) to extract the data in VOI, and Min-max normalization, formula: Value_normalization = (Value_origin-Value_min) / (Value_max-Value_min). Value_origin is the original CT value of a certain position in the target area, and value_max and value_min are the maximum and minimum original CT values in the target area respectively. After Min-max normalization, the relative difference between the maximum and minimum values in the target area is normalized to 1~0, so that the difference between the inconsistent CT values of each tumor can be weakened. The arithmetic mean of each layer in the ring layer segmentation is calculated, and the sum of the relative CT average values of the first and second layers after normalization with the respective zero layer as the baseline is taken as OMP.

[0100] (6), according to the current diagnostic standard, the cutoff value of OMP is 0.32, if the OMP value of the case is greater than 0.32, it is considered as high perfusion, and the possibility of immunotherapy is large, and immunotherapy is recommended, otherwise it is considered as low perfusion, and the effectiveness of immunotherapy is low, and immunotherapy is not recommended.

[0101] (7), regardless of the evaluation result, if the patient undergoes immunotherapy, routine follow-up is carried out, and the effect is recorded and fed back, and the model is continuously improved.

[0102] 2.3 Other application schemes

[0103] The biomarker OMP of the application can also be combined with other clinical indicators such as PD-L1 expression to construct a more effective efficacy prediction model. The OMP extraction method is the same as the above scheme, combined with the PD-L1 expression (TPS) of the patient's tumor tissue, microsatellite instability, TMB and other indicators, a joint prediction model can be established using logistic regression to significantly improve the prediction efficiency of the model. The prediction result of the prediction model established by the biomarker OMP combined with PD-L1 is shown in the table as follows, and the prediction efficiency of this simple model is as high as 0.801. Figure 8

[0104] At the same time, the tumor tissue perfusion characteristics OMP can also be specifically extracted based on different imaging methods of different tumor types to predict the effect of immunotherapy. At the same time, the tumor tissue perfusion characteristics OMP can also be specifically extracted based on different imaging methods of different tumor types to predict the effect of immunotherapy.

[0105] Exploration of CT imaging to evaluate vascular perfusion features of tumors

[0106] I. Using the difference between plain and enhanced CT to reflect the perfusion features of tumor tissues

[0107] 1. CT Silhouette image method

[0108] In enhanced CT, the CT value increase caused by contrast agent is positively correlated with blood supply, so the most intuitive method to evaluate the tumor blood perfusion is the difference method, that is, the CT value of the enhanced image minus the CT value of the pre-enhanced image. We used this method to analyze the perfusion status of 19 cases of liver metastatic carcinoma (10 cases did not respond to immunotherapy, 9 cases responded to immunotherapy). We found that in the plain and enhanced CT images ( Figure 13 top and middle images), liver metastases generally showed central low density (low perfusion / necrosis area inside the tumor), peripheral high density (tumor peripheral cell activity, high perfusion area and squeezed liver sinusoids). After rigid registration based on linear algorithm, we used the enhanced CT value minus the plain CT value to generate a new CT silhouette image ( Figure 13 bottom Silhouette image). We observed from it that the low perfusion area inside the patients who responded to immunotherapy had more gray scale increase areas ( Figure 13 Silhouette image in the lower figure, white spots in the green line range), which suggested that the tumor inside the patients who responded to immunotherapy might have better blood supply.

[0109] 2. CT texture feature method

[0110] We also established a method to evaluate the perfusion features of tumor tissues by changes in CT texture features before and after enhancement ( Figure 14 ). We found that regardless of the response to immunotherapy, the low perfusion area of liver tumors showed a trend of heterogeneity increase and homogeneity decrease after enhancement, which we considered to be related to the abnormal proliferation and dysfunction of blood vessels in malignant tumors, consistent with the general characteristics of malignant tumors. However, it is worth noting that the tumors that responded to immunotherapy showed a significant increase in high gray level related features and a more general decrease in low gray level tumors ( Figure 14 ), which was exactly the opposite in tumors that did not respond to immunotherapy. This phenomenon is consistent with the gray scale distribution features of the silhouette image and also supports the theory that the response of malignant tumors to immunotherapy is positively correlated with tumor blood perfusion.

[0111] Although both methods can be used to compare the vascular perfusion features of different tumors, their limitations in predicting efficacy are also obvious, mainly summarized as: the uncertainty of the boundary of the low perfusion area and the limitations of analysis, the problem of registration accuracy and the problem of incomplete image data.

[0112] Firstly, the boundary of this hypoperfusion region is difficult to determine, and the analysis results are limited. From plain and enhanced images, the most easily identifiable area is the region of intratumoral hypoperfusion / ischemic necrosis in the enhanced image. Figure 13 (Low grayscale area within the medium green line). In both sets of images, the enhanced low grayscale area was generally larger than that on the plain scan. This may be because the plain scan only included the tissue with more necrosis due to low perfusion, while the enhanced scan included the entire low perfusion area. However, the definition of the low perfusion area is controversial in different tumors, and some small tumors do not even have obvious low perfusion areas. Furthermore, as a heterogeneous whole, the analysis of low perfusion areas alone cannot objectively and comprehensively assess the overall blood supply of the tumor.

[0113] Secondly, accurate registration is difficult to guarantee. CT scans are typically performed after a deep inspiration and breath-holding, with the plain and enhanced images obtained during two separate breath-holding periods. This makes it difficult to ensure the consistency of tumor location and morphology between the two scans. Therefore, registration is necessary before obtaining silhouette images. Common registration methods include rigid and non-rigid registration, but neither currently provides accurate results. Finally, imaging data is often incomplete. Due to the complexity and diversity of clinical practice, many CT scans for malignant tumors lack accompanying plain images, or the time interval between the plain and enhanced scans is too long, resulting in significant changes in the tumor that make registration and comparison impossible.

[0114] II. Texture features in single-enhanced CT reflect lung cancer vascular perfusion characteristics.

[0115] The incompleteness of imaging data was clearly exposed in a non-small cell lung cancer clinical study we collected. Because this study was multicenter, and different institutions had different work practices and clinical focuses, only about 10% of cases had plain CT images, about 60% had standard arterial phase images, about 30% had standard venous phase images, and about 20% had contrast-enhanced images between the arterial and venous phases. We focused our attention on cases with standard arterial phase images. Without a baseline from the plain CT scan, it is impossible to assess tumor perfusion; therefore, we focused on exploring the characteristics of tumor contrast-enhanced grayscale distribution, such as homogeneity and high grayscale features.

[0116] By comparing the uniformity and high grayscale characteristics of lung cancer tumors, we did not find significant differences. Based on our previous experience in analyzing liver tumors, we analyzed the possible main reasons: uniformity alone cannot reflect the characteristics of tumor vascular perfusion. Tumors with good perfusion may show uniformly high or unevenly high values, while tumors with poor perfusion may show uniformly low or unevenly low values.

[0117] III. Gray-scale composition characteristics of single-contrast CT reflect the vascular perfusion characteristics of lung cancer.

[0118] We believe that the higher the degree of vascular normalization, the closer the gray-level distribution within the tumor tends to a normal distribution. We hypothesized that this phenomenon might be reflected in the gray-level histogram, so we performed a preliminary analysis on a subset of cases. To offset the heterogeneity of different tumor baselines, we performed Min-Max normalization on the tumor CT values, placing the minimum and maximum CT values ​​of all tumors within the range of 0–100 or 0–1, allowing their gray-level density histograms to be compared at the same level. Figure 15 In the preliminary analysis, we found that the distribution of CT values ​​in tumor CT images of patients who responded to and did not respond to immunotherapy tended to be normal. The grayscale distribution in the responsive group was biased to the right, meaning that tumors responding to immunotherapy had more mid-to-high grayscale voxels and fewer mid-to-low grayscale voxels after CT enhancement. This suggests that the CT enhancement effect on immunotherapy-responsive tumors is more generalized, indicating greater effective vascular perfusion. Subsequently, we sorted the grayscale values ​​of all voxels in each tumor from low to high and plotted the relative CT values ​​at the percentile position. Figure 16 We found that in the low-level range of voxel grayscale (0.2–0.4), the responding group had fewer voxels; while in the high-level range of voxel grayscale (0.5–0.7), the responding group had more voxels. This result is consistent with the grayscale density histogram results, and we believe that this feature can reflect the vascular perfusion status of lung cancer to some extent. In subsequent experiments, we expanded the sample size and analyzed images that met the requirements from the entire clinical trial cohort. Unfortunately, this feature was not reflected in the large sample size analysis. Figure 17 , Figure 18 Furthermore, our analysis of feature layers such as the intermediate layer and the maximum layer failed to yield significant differences, indicating that this attempt was unsuccessful.

[0119] In summary, we believe that the concept of grayscale composition in enhanced lung cancer CT is too simplistic. It is similar to kurtosis in the first-order grayscale features of radiomics. In previous texture analysis, the kurtosis feature did not show significant differences, so manual adjustment and calculation did not show positive results.

[0120] Example 3: Exploration of parameters in the process of OMP reflecting the vascular perfusion characteristics of lung cancer.

[0121] Although malignant lung tumors are predominantly characterized by heterogeneous enhancement, based on the general characteristics of tumors, we hypothesized that tumor perfusion should gradually decrease from the outside in. Therefore, we attempted to analyze the tumor layer by layer from the outside in, as described above. Figure 1The results are shown in the graph. We first performed traditional circle layer segmentation on the maximum layer of the tumor, and plotted the average gray value of each layer for comparison, and found no obvious difference. Considering that a single layer cannot show the overall tumor, we then analyzed the entire tumor layer by layer like "peeling onions" in three dimensions, and obtained statistically significant difference results (the graph of the broken line). In order to make this difference reflected in specific numerical values, we made the following attempts: Figure 19

[0122] 1. Correlation coefficient of circle layer gray value broken line and line Y=X

[0123] We found that the lung cancer CT image after normalization from the outermost layer (0th layer) showed a trend from high to low from outside to inside, and the tumors with effective immunotherapy showed a higher level. We first thought of using the correlation coefficient of the two groups of broken lines and the straight line Y=X to quantify this difference, and using this as a model to evaluate the efficacy of immunotherapy (AUC=0.622). Although this prediction effect has a significant difference, the prediction power is not ideal. Figure 20

[0124] 2. Remove the 0th layer, the correlation coefficient of the downward trend broken line and the line Y=8-X

[0125] Since the circle layer gray value broken line is an inverted U shape, its correlation with Y=X may not truly quantify the characteristics of the broken line, so we removed the 0th layer value to get two groups of gradually decreasing broken line graphs, and quantified the broken line characteristics by its correlation coefficient with Y=8-X to construct the model (AUC=0.603). The prediction power of this method is also not ideal. Figure 21

[0126] 3. Remove 0 layer, fit slope k

[0127] The correlation coefficient cannot effectively improve the prediction power, so we try to quantify the characteristics of the broken line by its slope, and construct the model by this method (AUC=0.633). The prediction power of this method has improved, but it is still not ideal. Figure 22

[0128] 4. Sum of circle layer relative gray value average

[0129] The above methods cannot effectively improve the prediction power of the model, we consider that the reason may be that in different tumors, the promotion trend of different circle layers and the overall promotion trend of circle layer perfusion are different, so the performance in correlation coefficient and slope is poor.

[0130] ​​​​In the next exploration, we went back and carefully considered the tumor microenvironment characteristics, and found a simpler and more tumor biology interpretable quantification method: the relative advantage of the outer layer of the tumor. Therefore, we tried to directly add the relative gray values of the circle layer perfusion. Since the circle layer perfusion of the reactive and non-reactive tumors differs significantly in 1-4 layers, we constructed a model with the sum of 1-4 layers ( Figure 23 ), and the AUC was 0.662. By using this method, the prediction efficiency was significantly improved.

[0131] Figure 23 The results show that the relative advantage of the outer layer of the tumor is correct. Encouraged by this result, we tried to construct a model with the sum of 1-3 and 1-2 layers ( Figure 24 and Figure 25 ), and found that the sum of 1-2 layers has the best prediction efficiency, with an AUC of 0.703.

[0132] The above is only a preferred embodiment of the present application, and it should be pointed out that for ordinary skilled persons in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. An OMP marker characterized in that, a sum of single-layer relative CT average values or multi-layer relative CT average values normalized by zero-layer relative CT average values relative to CT average values of each layer of the annular layer segmentation image; the relative CT average value of each layer is a ratio of a sum of relative CT values of voxels of each layer to a number of voxels of each layer; the method for obtaining the annular layer segmentation image comprises: obtaining a segmentation image by manually or automatically delineating a volume of interest after resampling a solid part enhanced CT image of a tumor; and obtaining the annular layer segmentation image by layer-by-layer segmentation in an onion-peeling manner from an outer layer to an inner layer.

2. The OMP marker according to claim 1, wherein the relative CT value is obtained by Min-max normalization of an original CT value, and the Min-max normalization formula is: Value normalization = (Value origin-Value min) / (Value max-Value min); wherein Value origin is an original CT value of the segmentation image; Value max is a maximum original CT value in all segmentation images; and Value min is a minimum original CT value in all segmentation images.

3. The OMP marker of claim 1, wherein, the delineation is a combination of manual delineation and automatic delineation; and when the tumor is lung cancer, the automatic delineation threshold is -50HU-200HU.

4. The OMP marker of claim 1, wherein, the thickness of the layer-by-layer segmentation is 0.5mm-10mm.

5. The OMP marker of claim 1, wherein, the image of the tumor includes any one of a lung cancer image, a breast cancer image, a liver cancer image, an esophageal cancer image, a colorectal cancer image, a pancreatic cancer image, a head and neck tumor image, a central nervous system tumor image, a gynecological tumor image, a soft tissue sarcoma image, and an image of a primary tumor and metastatic tumor thereof.

6. A combination tumor marker characterized in that, the tumor marker and the OMP marker according to any one of claims 1-5.

7. The combination tumor marker according to claim 6, characterized in that, the tumor marker is selected from the group consisting of programmed death-ligand 1, programmed death-receptor 1, tumor mutational burden, mismatch repair deficiency, high microsatellite instability, tumor size, tumor burden, CD8 + T cells, CD4 + T cells, NK cells, Treg cells, dendritic cells, interferon gamma, granzyme, perforin, Foxp3, CXCL9, CXCL10, CXCL11, STAT1, STAT3, JAK1, JAK2, IRF, CCL8, TIGIT, ICOS, ICOSL, antigen processing associated transporter TAP1, antigen processing associated transporter TAP2, MHC class II transcription activator, beta 2-microglobulin, PSA, fPSA, alpha-fetoprotein, carcinoembryonic antigen, carbohydrate antigen 199, cancer antigen 125, tumor antigen 153, cancer antigen 50, carbohydrate antigen 242, beta 2-microglobulin, serum ferritin, neuron-specific enolase, or human chorionic gonadotropin.

8. A method for constructing a tumor clinical prediction model, comprising: obtaining the OMP marker according to any one of claims 1-5 or the joint tumor marker according to claim 6 or 7; and constructing a therapeutic effect prediction model.

9. Any one of I)-III) as shown below in the application of tumor imaging, imaging genomics, preparation of a tumor therapeutic effect prediction product or a tumor clinical prediction model: I) the OMP marker according to any one of claims 1-5; II) the joint tumor marker according to claim 6 or 7; III) the method for constructing according to claim 8.

Images