System for constructing prediction model for tumor accumulation of nano-particles based on CT and ultrasound imaging

Abstract

Disclosed is a system for constructing a prediction model for tumor accumulation of nano-particles based on CT and ultrasonic imaging, comprising: a nano-particle synthesis and detection apparatus, a model construction module, a CT and ultrasound imaging apparatus, a radiomics feature extraction module, a feature analysis module, a delivery and accumulation classification module, and a prediction model construction module. The present invention realizes non-invasive prediction of tumor accumulation of nano-particles based on CT and ultrasound imaging data. The method can exclude animals that are not suitable for nanomedicine before nanomedicine injection. The present invention is expected to be used for animal tumor stratification in the process of nanomedicine research and development, and to select suitable test animals, thereby accelerating nano-drug research and development, and reducing research and development costs.

Description

A system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images Technical Field

[0001] This invention belongs to the field of nanoparticle delivery, and specifically relates to a system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images. Background Technology

[0002] Nanomedicines are a class of drugs loaded with nanoparticles that utilize the selective accumulation of nanoparticles in tumors to achieve targeted drug delivery to tumors. However, compared to traditional small molecule formulations, nanomedicines are more expensive to develop and have a higher risk of clinical translation failure. A major problem is the significant inter-tumor variability in nanoparticle delivery efficiency. Because tumor heterogeneity significantly affects nanoparticle accumulation, traditional pharmacokinetic models cannot accurately predict nanoparticle accumulation in tumors, making it difficult to stratify subjects to reduce development costs and improve the success rate of nanomedicine translation. Currently developing nanomedicine accumulation prediction models mainly rely on pathological samples and cannot non-invasively and prospectively predict nanomedicine delivery efficiency.

[0003] Medical imaging can non-invasively reveal the heterogeneity of tumors. In particular, radiomics technology can extract quantitative features of tumor images in high throughput and use artificial intelligence models to screen tumor image features most closely related to the accumulation level of nanomedicines, which is expected to improve the accuracy of medical imaging in predicting the tumor accumulation level of nanomedicines.

[0004] However, to date, there is no research on using radiomics to predict the accumulation level of nanoparticles in animal tumors. Therefore, providing a method and tool based on CT and ultrasound radiomics technology to predict the accumulation of nanoparticles in animal tumors is a topic that urgently needs to be studied by those skilled in the art. This tool can be used for animal tumor stratification in the process of nanomedicine development, thereby accelerating the development speed of nanomedicine and saving development costs. Summary of the Invention

[0005] The purpose of this invention is to address some of the problems existing in the prior art and to provide a system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images, in order to solve the problem that there is currently no prospective and non-invasive prediction scheme for the delivery of nanoparticles to animal tumors.

[0006] To achieve the above objectives, the present invention employs the following technical solution: a system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images, comprising:

[0007] A nanoparticle synthesis and detection device for synthesizing nanoparticles and obtaining particle size data;

[0008] The model building module is used to build one or more tumor models and obtain tumor type data.

[0009] CT is used to perform CT imaging of tumors;

[0010] Ultrasound imaging device, used for ultrasound imaging of tumors;

[0011] The radiomics feature extraction module is used to segment tumor regions from tumor images obtained from CT and ultrasound imaging devices and extract radiomics features.

[0012] The feature analysis module is used to analyze the extracted radiomics features, remove highly collinear features, and screen out features closely related to the accumulation of nanoparticles in tumors.

[0013] The delivery and accumulation classification module is used to deliver nanoparticles to cause them to accumulate at the tumor site, collect the tumor and detect the amount of nanoparticles accumulated within the tumor, and classify the tumor based on the median accumulation amount.

[0014] A predictive model module was constructed to build a radiomics composite predictive model based on nanoparticle size, tumor type, CT and ultrasound radiomics characteristics, and tumor nanoparticle accumulation data, and to evaluate the performance of the composite predictive model.

[0015] Furthermore, the synthesized nanoparticles are gold nanoparticles, silica nanoparticles, nanoliposomes, or nanomicelles, and the surface of the nanoparticles is modified with polyethylene glycol. The hydrated particle size of the synthesized nanoparticles is 10–300 nm, and the morphology is spherical, rod-shaped, or multi-branched.

[0016] Furthermore, the animal type of the tumor model is a rat or mouse, the tumor type is pancreatic cancer, liver cancer, breast cancer, lung cancer, bladder cancer, colon cancer or melanoma, and the tumor site is a subcutaneous tumor or an in situ tumor.

[0017] Furthermore, the tumor regions are segmented from tumor images obtained from CT and ultrasound imaging devices, and radiomics features are extracted using the following steps:

[0018] (1) The tumor region in the tumor image was delineated using ITK-SNAP software, in which the tumor region was delineated layer by layer in the CT image and the region of interest of the largest diameter layer of the tumor was delineated in the ultrasound image.

[0019] (2) Use the Pyradiomics package in Python to extract tumor CT and ultrasound image features.

[0020] Furthermore, radiomics features include first-order statistics, shape features (2D and 3D), texture features, and exponential, logarithmic, square, gradient, square root, Gaussian Laplace, and wavelet transform features.

[0021] Furthermore, features closely related to the accumulation of nanoparticles within tumors were screened out, specifically through the following steps:

[0022] (1) Some tumor features were randomly selected from all tumor features for a second delineation. The two delineations of regions of interest were one month apart. Radiomic features with a correlation coefficient (ICC) > 0.80 in the second delineation were included in the subsequent feature screening.

[0023] (2) By using the Pearson correlation coefficient analysis method, the feature set with a correlation coefficient exceeding 0.95 between image features is simplified into a single representative feature;

[0024] (3) Through LASSO regression analysis, feature selection and coefficient shrinkage were achieved for CT and ultrasound image omics features, and features closely related to the prediction of nanoparticle tumor accumulation were identified.

[0025] Furthermore, the amount of nanoparticles accumulated within the tumor was detected 24 hours after nanoparticle delivery.

[0026] Furthermore, the establishment of the radiomics-based composite prediction model adopts the following steps:

[0027] (1) Integrate tumor type, nanoparticle size, SWE average value and dynamic enhancement curve quantitative parameters, CT-ultrasound image omics characteristics and tumor nanoparticle accumulation data;

[0028] (2) The model was constructed using a logistic regression model;

[0029] (3) Establishing prediction models: Model 1 is a composite model based on radiomics, and Model 2 is the baseline model used for comparison performance.

[0030] Furthermore, the predictive model building module is also used to establish a baseline model for performance comparison.

[0031] Furthermore, the performance evaluation of the radiomics-based composite prediction model employs the following steps:

[0032] (1) Use ROC to evaluate the predictive performance of each model and compare the area under the curve (AUC) of each model;

[0033] (2) Using a confusion matrix, the classification performance of the model is further quantified to evaluate the accuracy of the model in tumor classification;

[0034] (3) Using DCA analysis, the clinical applicability or economic value of the model at different thresholds can be intuitively displayed by calculating the net benefit ratio. The additional benefits that each model can bring in actual application can be compared, so as to select the optimal prediction model.

[0035] The advantages of this invention are: based on CT and ultrasound imaging data, it achieves non-invasive prediction of tumor accumulation using nanoparticles; this method can exclude animals unsuitable for nanomedicine before injection. It is expected to be used for tumor stratification in animals during nanomedicine development, screening suitable test animals, thereby accelerating nanomedicine development and saving development costs. Attached Figure Description

[0036] Figure 1 is a flowchart of the method and tool for predicting nanoparticle tumor accumulation based on CT and ultrasound radiomics technology provided in an embodiment of the present invention.

[0037] Figure 2 is a statistical chart of the accumulation of gold nanoparticles of different sizes in different tumors according to the embodiments of the present invention.

[0038] Figure 3 shows the ROC curves and calibration curves of different prediction models provided in the embodiments of the present invention.

[0039] Figure 4 shows the confusion matrix of different prediction models provided in the embodiments of the present invention.

[0040] Figure 5 shows the DCA curves of different prediction models provided in the embodiments of the present invention.

[0041] Figure 6 shows the prediction results of typical cases provided by the embodiments of the present invention on different models. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be used to limit the scope of protection of the present invention.

[0043] This invention is not limited to the following embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0044] The present invention will now be described in detail with reference to specific embodiments and accompanying drawings: Example

[0045] To achieve the objectives of this invention, this embodiment provides a method and tool for predicting nanoparticle tumor accumulation based on CT and ultrasound imaging omics technology, as shown in Figure 1. The implementation steps include the following seven stages: synthesis, modeling, scanning, grouping, extraction and screening, prediction, and evaluation.

[0046] The synthesis step is used to obtain nanoparticles of different sizes;

[0047] The modeling step is used to obtain various tumor models;

[0048] The scanning process is used to obtain CT and ultrasound images of the tumor.

[0049] The grouping step is used to group nanoparticles according to their accumulation amount;

[0050] The extraction and screening steps are used for the extraction and screening of radiomics features;

[0051] The prediction step is used to construct various prediction models;

[0052] The evaluation step is used to evaluate the prediction model.

[0053] The nanoparticles synthesized in the aforementioned synthesis step are gold nanoparticles (GNPs). By controlling the reaction conditions, GNPs of 15 nm, 40 nm, and 70 nm were synthesized, respectively. To increase the blood circulation time of GNPs and maintain their surface potential inhibition, the surface of the GNPs was modified with polyethylene glycol to bring the potential close to electroneutrality. In addition, the required GNP size was ensured by detecting the hydrated particle size and using transmission electron microscopy.

[0054] The various tumor models constructed in the modeling process used C57BL / 6 mice as the animal model, with tumors implanted subcutaneously on the mouse's back. The constructed tumor types included melanoma (B16 cell line), breast cancer (E0771 cell line), liver cancer (Hepa1-6 cell line), lung adenocarcinoma (LLC cell line), bladder cancer (MB49 cell line), colon cancer (MC38 cell line), and pancreatic cancer (PAN02 cell line).

[0055] In the scanning process for acquiring tumor CT and ultrasound images, the CT scan requires control of parameters including slice thickness (1 mm), slice spacing (0 mm), tube voltage (120 kVp), and tube current-time product (100 mAs). For ultrasound imaging, an abdominal ultrasound probe is used at a frequency of 4-13 MHz; imaging modalities include B-mode US, SWE, and CEUS imaging. During CEUS imaging, continuous recording is performed for 90 seconds. From the CEUS images, a time-intensity curve is fitted, and the following parameters are further obtained from this curve: curve type, perfusion slope, peak intensity, time to peak, and area under the curve.

[0056] The grouping process utilizes the GNPs and tumor models synthesized in the synthesis and modeling stages. During the scanning stage, GNPs of different particle sizes are injected into mice with different tumor types via tail vein injection. After 24 hours, the tumors are dissected, and the gold mass in each tumor is obtained using inductively coupled plasma optical emission spectrometry (ICP-OES). The GNP accumulation in each tumor is calculated using the formula: Intratumoral accumulation (%ID / g) = Gold mass in tumor / (Gold mass injected intravenously × Tumor mass) × 100%. Based on the median accumulation, the tumors are divided into a high-accumulation group and a low-accumulation group (Figure 2).

[0057] The extraction and screening process utilizes CT and ultrasound images obtained from the scanning process. Regions of interest (ROIs) are manually delineated using the open-source software ITK-SNAP. After delineation, the original tumor images and mask files are stored. Radiomics features are extracted from each tumor ROI using the open-source Pyradiomics package, including first-order statistics, shape features (including 2D and 3D), texture features, and exponential, logarithmic, squared, gradient, square root, Laplace-Gaussian, and wavelet transform features. After extraction, each radiomics feature is standardized using the Z-score method. The Z-score calculation formula is: Z = X−μσ, where X is the value to be calculated, μ is the mean of the feature values, and σ is the standard deviation of the feature values. To assess inter-observer consistency, a subset of tumors is randomly selected from all tumors for two delineations, with a one-month interval between the two ROI delineations. Radiomics features with an ICC > 0.80 are included in subsequent feature screening. By using Pearson correlation coefficient analysis, features with correlation coefficients exceeding 0.95 were simplified into single representative features to reduce multicollinearity. LASSO regression analysis was then used to screen out a few CT and ultrasound radiomics features most critical for nanoparticle tumor accumulation, thereby improving the model's interpretability and predictive performance.

[0058] The prediction process involves constructing multiple prediction models. First, all mouse tumor data are randomly divided into a training set and a test set (7:3). A model is constructed using logistic regression. Based on the source of features, prediction models are established: Model 1 is based on a radiomics composite model (tumor type, nanoparticle size, SWE average value and dynamic enhancement curve quantitative parameters, CT-US radiomics features), and Model 2 is the baseline model (tumor category and nanoparticle size).

[0059] The evaluation process assesses various prediction models established using different methods. As shown in Figure 3, ROC is used to evaluate the predictive efficacy of each model, and the models are compared based on the area under the curve. As shown in Figure 4, a confusion matrix is ​​used to further quantify the classification performance of the models and evaluate their accuracy in tumor classification. As shown in Figure 5, DCA curves are used to evaluate the models and, by calculating the net benefit ratio, to visually demonstrate the clinical applicability or economic value of the models at different thresholds, comparing the additional benefits that each model can bring in practical applications, thereby selecting the optimal prediction model.

[0060] Based on the established predictive models, two cases (A and B) were used as examples to compare the efficacy of different models in tumor stratification (Figure 6). The radiomics composite model predicted correctly, while the baseline model predicted incorrectly. The radiomics composite model accurately stratified tumors in animals by effectively combining cancer type with radiomics features such as US Radscore, CT Radscore, SWE mean, and peak intensity of the time-intensity curve. In contrast, the baseline model relied solely on GNP size and cancer type, leading to prediction errors. This highlights the importance of incorporating radiomics features to improve model performance.

[0061] In summary, the method and tool for predicting nanoparticle delivery in animal tumors based on CT and ultrasound radiomics technologies provided in this embodiment have the following technical advantages:

[0062] This embodiment provides a novel scheme for predicting the tumor delivery efficiency of gold nanoparticles in mouse tumors based on radiomics technology. The scheme includes synthesis, modeling, scanning, grouping, extraction and screening, prediction, and evaluation. Through these steps, radiomics features of mouse tumors from CT and ultrasound can be extracted. Then, based on the model, mouse tumors are classified to distinguish between tumors with high and low accumulation of gold nanoparticles, thereby achieving non-invasive prediction of nanoparticle tumor delivery efficiency. This can be used for animal model stratification in the nanomedicine development process, accelerating nanomedicine development and saving research costs.

[0063] As used in this article, the term "radiomics" refers to the extraction of a large number of quantitative features (such as shape features, texture features, etc.) from regions of interest in medical images (such as CT images, ultrasound images), and the application of statistical and machine learning methods to analyze them in order to find patterns related to outcomes (such as tumor growth rate, treatment response, survival rate, etc.).

[0064] As used herein, the term "nanoparticle" refers to tiny particles with a size between 10 and 300 nm. They can be made of a variety of materials, including metals, semiconductors, polymers, and biomolecules. The following provides further description of some of the nanoparticles covered by this invention:

[0065] Nanoparticle 1: As used herein, the term "gold nanoparticle" refers to an aggregate composed of gold atoms. Its size ranges from 10 to 300 nm. Based on size effects and surface plasmon resonance phenomena, gold nanoparticles possess unique optical, electrical, and catalytic properties.

[0066] Nanoparticles 2: As used herein, the term "silica nanoparticles" refers to tiny particles composed of silicon dioxide (SiO2). Their size ranges from 10 to 300 nm. Silica nanoparticles possess a porous structure, stable chemical properties, good biocompatibility, and easily modifiable surfaces.

[0067] Nanoparticles 3: As used herein, the term "nanoliposome" refers to a closed spherical structure composed of a phospholipid bilayer. The size ranges from 10 to 300 nanometers. Nanoliposomes mimic the structure of biological membranes, with an internal aqueous phase and an external hydrophobic phospholipid bilayer. They exhibit good biocompatibility, targeting ability, and controlled release.

[0068] Nanoparticles 4: As used herein, the term "nanomimus" refers to aggregates of amphiphilic molecules (i.e., molecules with both hydrophilic and hydrophobic ends) that spontaneously form in water. Their size ranges from 10 to 300 nanometers. Their structure is similar to miniature soap bubbles, with the hydrophobic ends agglomerating inward to form a core, while the hydrophilic ends contact the water outward, forming a stable interface. Nanomicelles are characterized by self-assembly, uniform size, and drug-carrying capacity.

[0069] As used in this article, the term "aggregation" in the field of nanomedicine refers to the concentrated distribution of nanoparticles in a specific biological environment or target site. In tumor diagnosis and treatment, a high aggregation level of nanoparticles means that more of them reach the tumor area, enhancing the therapeutic effect while reducing side effects on normal tissues. This process depends not only on the physicochemical properties of the nanoparticles themselves but also on the physiological and pathological characteristics of the tumor.

[0070] As used herein, the term “prediction” refers to a guess about the amount of nanoparticles accumulating in a tumor, and for the purposes of this invention, it refers to a guess about the amount of nanoparticles accumulating at the tumor site (e.g., high accumulation, low accumulation).

[0071] As used herein, the term "hydrated particle size" refers to the particle size measured in an aqueous solution after water molecules adsorb onto its surface to form a hydration layer. Furthermore, in this invention, when nanoparticles are dispersed in water, water molecules surround the particle surface to form one or more hydration shells, which increase the actual size of the particles. The measurement of hydrated particle size reflects the true dispersion state and dynamic size of the nanoparticles in aqueous solution, and is closely related to the stability, dispersibility, and behavior of the nanoparticles in organisms.

[0072] As used herein, the term "polyethylene glycol" refers to a linear polymer (PEG) composed of repeating ethylene oxide units (-CH2-CH2-O-). The molecular weight of PEG ranges from several hundred to several million Daltons, depending on the number of repeating units. PEG is a colorless, odorless, and non-toxic water-soluble polymer with excellent chemical stability and biocompatibility. Furthermore, in this invention, PEG is modified onto the surface of nanoparticles to increase their stability and cycle time, reduce the rate of removal, and simultaneously improve the bioavailability of the nanoparticles.

[0073] As used herein, the term "stratification" refers to dividing things into different levels or categories according to certain criteria or attributes. For the purposes of this invention, it refers to classifying tumors into high-accumulation and low-accumulation groups based on the amount of nanoparticles accumulated in the tumor.

[0074] As used in this article, the term "CT" refers to Computed Tomography, a medical imaging method that uses X-rays and computer processing technology to generate images of tissue structures based on differences in tissue density.

[0075] As used in this article, the term "US" refers to Ultrasound Imaging, a medical imaging method that uses the differences in the propagation speed and reflection characteristics of high-frequency sound waves in different tissues of the human body to generate images of tissue structures.

[0076] As used in this article, the term "B-mode US" refers to B-mode grayscale imaging, which is one of the most common types of ultrasound imaging. It is an imaging method that generates the brightness (i.e., grayscale value) of each point in the image based on the signal intensity reflected back when ultrasound waves pass through human tissue.

[0077] As used in this article, the term "SWE" refers to Shear Wave Elastography (SWE), an imaging method that quantifies tissue stiffness based on the propagation speed of shear waves.

[0078] As used in this article, the term "CEUS" refers to contrast-enhanced ultrasound, an imaging method that uses microbubble contrast agents to enhance the quality of ultrasound imaging, primarily used to improve the visibility of vascular structures and reflect the blood supply to tissues.

[0079] It should be understood that the aforementioned technical means can be implemented individually in hardware or software form, or through an integration of both. Therefore, the methods and apparatus involved in this invention, as well as some of their constituent elements or aspects, can be embedded in a physical medium, such as a floppy disk, CD-ROM, hard disk, or any other machine-readable storage medium carrying program code (i.e., a series of instructions). When this program is loaded onto a device such as a computer and executed, the device becomes a tool for implementing this invention.

[0080] The foregoing description reveals the basic principles, core features, and advantages of this invention. It should be understood that, for those skilled in the art, the scope of this invention is not limited to the exemplary embodiments described above. These examples and descriptions are intended to illustrate the core principles of this invention. However, without departing from the core spirit and overall framework of this invention, various variations and optimizations are possible. All these variations and improvements fall within the scope of protection sought by this invention, and the specific scope of protection is defined by the appended claims and their equivalents.

Claims

1. A system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images, characterized in that... include: A nanoparticle synthesis and detection device for synthesizing nanoparticles and obtaining particle size data; The model building module is used to build one or more tumor models and obtain tumor type data. CT is used to perform CT imaging of tumors; Ultrasound imaging device, used for ultrasound imaging of tumors; The radiomics feature extraction module is used to segment tumor regions from tumor images obtained from CT and ultrasound imaging devices and extract radiomics features. The feature analysis module is used to analyze the extracted radiomics features, remove highly collinear features, and screen out features closely related to the accumulation of nanoparticles in tumors. The delivery and accumulation classification module is used to deliver nanoparticles to cause them to accumulate at the tumor site, collect the tumor and detect the amount of nanoparticles accumulated within the tumor, and classify the tumor based on the median accumulation amount. A predictive model module was constructed to build a radiomics composite predictive model based on nanoparticle size, tumor type, CT and ultrasound radiomics characteristics, and tumor nanoparticle accumulation data, and to evaluate the performance of the composite predictive model.

2. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The synthesized nanoparticles are gold nanoparticles, silica nanoparticles, nanoliposomes, or nanomicelles. The surface of the nanoparticles is modified with polyethylene glycol. The hydrated particle size of the synthesized nanoparticles is 10–300 nm, and the morphology is spherical, rod-shaped, or multi-branched.

3. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The tumor model is based on rats or mice, and the tumor type is pancreatic cancer, liver cancer, breast cancer, lung cancer, bladder cancer, colon cancer, or melanoma. The tumor site is either a subcutaneous tumor or an in situ tumor.

4. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The specific steps for segmenting tumor regions from tumor images obtained from CT and ultrasound imaging devices and extracting radiomics features are as follows: (1) The tumor region in the tumor image was delineated using ITK-SNAP software, in which the tumor region was delineated layer by layer in the CT image and the region of interest of the largest diameter layer of the tumor was delineated in the ultrasound image. (2) Use the Pyradiomics package in Python to extract tumor CT and ultrasound image features.

5. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The image omics features include first-order statistics, shape features (2D and 3D), texture features, and exponential, logarithmic, square, gradient, square root, Gaussian Laplace, and wavelet transform features.

6. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The screening process for features closely related to the accumulation of nanoparticles within tumors involves the following steps: (1) Some tumor features were randomly selected from all tumor features for a second delineation. The two delineations of regions of interest were one month apart. Radiomic features with a correlation coefficient (ICC) > 0.80 in the second delineation were included in the subsequent feature screening. (2) By using the Pearson correlation coefficient analysis method, the feature set with a correlation coefficient exceeding 0.95 between image features is simplified into a single representative feature; (3) Through LASSO regression analysis, feature selection and coefficient shrinkage were achieved for CT and ultrasound image omics features, and features closely related to the prediction of nanoparticle tumor accumulation were identified.

7. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The amount of nanoparticles accumulated within the tumor was detected 24 hours after nanoparticle delivery.

8. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The radiomics-based composite prediction model was established using the following steps: (1) Integrate tumor type, nanoparticle size, SWE average value and dynamic enhancement curve quantitative parameters, CT-ultrasound image omics characteristics and tumor nanoparticle accumulation data; (2) The model was constructed using a logistic regression model; (3) Establishing prediction models: Model 1 is a composite model based on radiomics, and Model 2 is the baseline model used for comparison performance.

9. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The predictive model building module is also used to establish a baseline model for performance comparison.

10. The system for constructing a nanoparticle tumor accumulation prediction model based on CT and ultrasound images according to claim 1, characterized in that: The performance evaluation of the radiomics-based composite prediction model is carried out using the following steps: (1) Use ROC to evaluate the predictive performance of each model and compare the area under the curve (AUC) of each model; (2) Using a confusion matrix, the classification performance of the model is further quantified to evaluate the accuracy of the model in tumor classification; (3) Using DCA analysis, the clinical applicability or economic value of the model at different thresholds can be intuitively displayed by calculating the net benefit ratio. The additional benefits that each model can bring in actual application can be compared, so as to select the optimal prediction model.

Images