Method and system for constructing early diagnosis model of ovarian cancer based on multi-modal data
By combining pelvic MRI imaging data and serum sample data, an early diagnostic model for ovarian cancer was constructed, which solved the problems of insufficient sensitivity and specificity in the early diagnosis of ovarian cancer in existing technologies, achieved high-precision early diagnosis, and provided a reliable basis for clinical decision-making.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NO 5 AFFILIATED HOSPITAL OF XINJIANG MEDICAL UNIV
- Filing Date
- 2026-04-02
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies lack sufficient sensitivity and specificity in the early diagnosis of ovarian cancer, especially in differentiating early lesions from benign diseases. There is a lack of effective screening methods, and traditional imaging assessments rely on physician experience and lack objective evidence.
By comprehensively utilizing pelvic MRI imaging data and serum sample data, and extracting DCE-MRI quantitative parameters and ADC values, combined with serum HE4 and CA125 levels, a multivariate logistic regression model for early diagnosis of ovarian cancer was constructed.
It improves the sensitivity and specificity of early diagnosis of ovarian cancer, provides reliable clinical diagnostic evidence, can accurately distinguish early ovarian cancer lesions from benign tumors, has strong system practicality, and can quickly output diagnostic results.
Smart Images
Figure CN122067754A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of methods for constructing diagnostic models for ovarian cancer, and in particular to a method and system for constructing an early diagnostic model for ovarian cancer based on multimodal data. Background Technology
[0002] Ovarian cancer is one of the most common malignant tumors of the female reproductive system. Due to its insidious early symptoms and lack of effective screening methods, most patients are diagnosed at an advanced stage, resulting in a poor prognosis. Improving the early diagnosis rate of ovarian cancer is key to improving patient survival outcomes. Currently, serum tumor markers commonly used in the clinical diagnosis and monitoring of ovarian cancer, such as carbohydrate antigen 125 (CA125) and human epididymis protein 4 (HE4), have some value, but their sensitivity and specificity still need to be improved, especially in differentiating early lesions from benign diseases. Medical imaging examinations, especially magnetic resonance imaging (MRI), play an important role in differentiating the nature of ovarian tumors, but traditional imaging evaluation relies heavily on physician experience and lacks objective functional evidence. Dynamic contrast-enhanced MRI (DCE-MRI) combined with pharmacokinetic models can quantitatively reflect tumor microvascular permeability, perfusion status, and extracellular matrix characteristics. Key perfusion parameters include the volume transfer constant (Ktrans), rate constant (Kep), and extravascular extracellular space fraction (Ve); the apparent diffusion coefficient (ADC) can evaluate changes in cell density. These indicators are more objective and sensitive than traditional imaging findings. This application aims to obtain quantitative parameters and ADC values of ovarian lesions using DCE-MRI, and to construct a comprehensive diagnostic model by combining serum HE4 and CA125, providing a method and system for constructing an early diagnostic model for ovarian cancer based on multimodal data. Summary of the Invention
[0003] In view of this, the present invention addresses the deficiencies of the existing technology, and its main objective is to provide a method and system for constructing an early diagnostic model for ovarian cancer based on multimodal data. By comprehensively utilizing pelvic MRI imaging data and serum sample data, it constructs an accurate and effective early diagnostic model for ovarian cancer, improves the sensitivity and specificity of early diagnosis of ovarian cancer, and provides a reliable basis for clinical diagnosis.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for constructing an early diagnostic model for ovarian cancer based on multimodal data includes the following steps: S1. Obtain a sample dataset, which includes pelvic MRI images and serum samples from patients with pathologically confirmed ovarian cancer and patients with benign ovarian tumors. S2. Extract DCE-MRI quantitative parameters and ADC values from the pelvic MRI image data; S3. Detect the levels of serum human epididymal protein 4 and carbohydrate antigen 125 from the serum sample; S4. Using the DCE-MRI quantitative parameters, ADC value, serum human epididymal protein 4 level and carbohydrate antigen 125 level as input features, a multivariate logistic regression model for early diagnosis of ovarian cancer is constructed.
[0005] As a preferred embodiment: in step S2, the DCE-MRI quantitative parameters are obtained by fitting dynamic contrast-enhanced MRI data to an extended Tofts pharmacokinetic model; the ADC value is obtained by acquiring and calculating using a diffusion-weighted imaging sequence.
[0006] As a preferred embodiment, step S2 involves extracting DCE-MRI quantitative parameters and ADC values as follows: A region of interest (ROI) or a three-dimensional volumetric ROI is delineated from pelvic MRI image data. A two-dimensional ROI is delineated on the diffusion-weighted imaging image at the level where the solid component of the lesion is most prominent for ADC value measurement. On conventional enhanced T1WI images of pelvic MRI image data, the lesion boundaries are delineated layer by layer to construct a three-dimensional volumetric ROI for DCE-MRI quantitative parameter measurement, excluding areas of necrosis, cystic degeneration, and hemorrhage.
[0007] As a preferred embodiment, the DCE-MRI quantitative parameters in step S2 include volume transport constant, reflux rate constant, and extracellular extravascular space fraction.
[0008] As a preferred method, the serum levels of human epididymal protein 4 and carbohydrate antigen 125 are quantitatively analyzed by enzyme-linked immunosorbent assay (ELISA).
[0009] As a preferred embodiment: the early diagnosis model for ovarian cancer is used to distinguish between early ovarian cancer lesions and benign ovarian tumors; the early diagnosis model for ovarian cancer is a multivariate logistic regression model, and its output diagnostic results include the probability value of the subject having ovarian cancer.
[0010] As a preferred embodiment, step S4 further includes: calculating the area under the curve, sensitivity, and specificity through receiver operating characteristic (ROC) curve analysis to evaluate the early diagnostic efficacy of the diagnostic model.
[0011] As a preferred embodiment, step S4 further includes: performing statistical analysis on the input features to select features with differences as model variables.
[0012] An early ovarian cancer diagnostic system constructed using the method described above includes: The image data acquisition module is used to acquire pelvic MRI image data of the subject under test; The serum data acquisition module is used to acquire the serum levels of human epididymal protein 4 and carbohydrate antigen 125 in the subjects to be tested. The feature extraction module is used to extract DCE-MRI quantitative parameters and ADC values from the pelvic MRI image data; The model processing module contains a diagnostic model trained by multivariate logistic regression based on the quantitative parameters of DCE-MRI, ADC value, serum human epididymal protein 4 level, and carbohydrate antigen 125 level, which is used to output the early diagnosis results of ovarian cancer for the test subjects.
[0013] As a preferred embodiment: the diagnostic results include at least one of the area under the receiver operating characteristic (ROC) curve, sensitivity, and specificity; the extraction of DCE-MRI quantitative parameters in the feature extraction module is based on fitting DCE-MRI data to an extended Tofts pharmacokinetic model.
[0014] Compared with existing technologies, this invention has significant advantages and beneficial effects. Specifically, as shown in the above technical solution, it comprehensively utilizes multimodal data to construct an early diagnostic model for ovarian cancer, and improves the accuracy and reliability of diagnosis through accurate feature extraction and model evaluation. Its advantages lie in the fusion of multimodal data, the precision of feature extraction, the scientific nature of model evaluation, and the practicality of the system, providing effective support for the early diagnosis of ovarian cancer. By integrating pelvic MRI imaging data and serum sample data, it overcomes the limitations of single diagnostic methods. The combination of imaging features and serum biomarker levels can reflect the characteristics of ovarian lesions from different dimensions, significantly improving the accuracy of early diagnosis and effectively distinguishing early ovarian cancer lesions from benign ovarian tumors. This diagnostic system is highly practical; its modules work collaboratively to quickly output diagnostic results, and the multiple diagnostic indicators it includes provide clinicians with intuitive and reliable decision-making basis.
[0015] To more clearly illustrate the structural features and effects of the present invention, a detailed description is provided below in conjunction with the accompanying drawings and specific embodiments. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the ROC curve for diagnosing early ovarian cancer using DCE-MRI quantitative parameters combined with serum HE4 and CA125 levels according to the present invention. Figure 2 This is a schematic diagram illustrating the steps of the method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to the present invention.
[0017] Figure 3 This is a flowchart illustrating the workflow of the ovarian cancer early diagnosis system of the present invention. Detailed Implementation
[0018] The present invention is as follows Figure 1 As shown in Figure 3, a method for constructing an early diagnostic model for ovarian cancer based on multimodal data includes the following steps: S1. Obtain the sample dataset, which includes pelvic MRI images and serum samples from patients with pathologically confirmed ovarian cancer and patients with benign ovarian tumors. S2. Extract DCE-MRI quantitative parameters and ADC values from the pelvic MRI image data; S3. Detect the serum levels of human epididymal protein 4 and carbohydrate antigen 125 from this serum sample; S4. Using the DCE-MRI quantitative parameters, ADC value, serum human epididymal protein 4 level, and carbohydrate antigen 125 level as input features, a multivariate logistic regression model for early diagnosis of ovarian cancer was constructed.
[0019] In step S2, the DCE-MRI quantitative parameters are obtained by fitting dynamic contrast-enhanced MRI data to an extended Tofts pharmacokinetic model; the ADC value is obtained by acquiring and calculating using diffusion-weighted imaging sequences.
[0020] The extraction of DCE-MRI quantitative parameters and ADC values in step S2 specifically involves: delineating the region of interest or three-dimensional volumetric region of interest from the pelvic MRI image data; delineating a two-dimensional region of interest at the level where the solid component of the lesion is most prominent on the diffusion-weighted imaging image for ADC value measurement; and delineating the lesion boundary layer by layer on the conventional sequence enhanced T1WI image of the pelvic MRI image data to construct a three-dimensional volumetric region of interest for DCE-MRI quantitative parameter measurement, while excluding necrotic, cystic, and hemorrhage areas.
[0021] In step S2, the DCE-MRI quantitative parameters include volume transport constant, reflux rate constant, and extracellular extravascular space fraction.
[0022] The serum levels of human epididymal protein 4 and carbohydrate antigen 125 were quantitatively analyzed by enzyme-linked immunosorbent assay (ELISA).
[0023] This early diagnosis model for ovarian cancer is used to distinguish early ovarian cancer lesions from benign ovarian tumors. The early diagnosis model for ovarian cancer is a multivariate logistic regression model, and its output diagnostic results include the probability value of the subject having ovarian cancer.
[0024] Step S4 also includes: calculating the area under the curve, sensitivity, and specificity through receiver operating characteristic (ROC) curve analysis to evaluate the early diagnostic efficacy of the diagnostic model.
[0025] Step S4 also includes: performing statistical analysis on the input features and selecting the features with differences as model variables.
[0026] An early diagnostic system for ovarian cancer constructed using this method includes: The image data acquisition module is used to acquire pelvic MRI image data of the subject under test; The serum data acquisition module is used to acquire the serum levels of human epididymal protein 4 and carbohydrate antigen 125 in the subjects to be tested. The feature extraction module is used to extract DCE-MRI quantitative parameters and ADC values from the pelvic MRI image data. The model processing module contains a diagnostic model trained by multivariate logistic regression based on the quantitative parameters of DCE-MRI, ADC value, serum human epididymal protein 4 level, and carbohydrate antigen 125 level, which is used to output the early diagnosis results of ovarian cancer for the test subjects.
[0027] The diagnostic results include at least one of the area under the receiver operating characteristic (ROC) curve, sensitivity, and specificity; the extraction of quantitative parameters of DCE-MRI in the feature extraction module is based on fitting DCE-MRI data to an extended Tofts pharmacokinetic model.
[0028] Workflow of the early diagnosis system for ovarian cancer: Data acquisition: The image data acquisition module acquires the pelvic MRI image data of the subject, and the serum data acquisition module acquires the serum human epididymal protein 4 level and carbohydrate antigen 125 level of the subject.
[0029] Feature extraction: The feature extraction module extracts DCE-MRI quantitative parameters and ADC values from pelvic MRI image data. The DCE-MRI quantitative parameters are generated by fitting the DCE-MRI data with an extended Tofts pharmacokinetic model.
[0030] Diagnostic results output: The model processing module inputs the extracted features and serum biomarker levels into the built-in diagnostic model and outputs the early diagnosis results of ovarian cancer for the subject, including the probability value of the subject having ovarian cancer, as well as indicators such as area under the curve, sensitivity, and specificity.
[0031] Advantages of Multimodal Data Fusion: This invention comprehensively utilizes pelvic MRI imaging data and serum sample data, combining imaging features and serum biomarker levels to overcome the limitations of single diagnostic methods and improve the accuracy of early ovarian cancer diagnosis. Through the fusion of multimodal data, the characteristics of ovarian lesions can be reflected from different perspectives, providing more comprehensive information for diagnosis.
[0032] Feature extraction accuracy: A precise delineation method was employed when extracting quantitative parameters and ADC values from DCE-MRI, excluding areas of necrosis, cystic degeneration, and hemorrhage to ensure accurate feature extraction. Furthermore, by fitting the DCE-MRI data with an extended Tofts pharmacokinetic model, quantitative parameters were obtained more accurately, providing reliable input features for the diagnostic model.
[0033] Model performance evaluation: During the construction of the diagnostic model, receiver operating characteristic (ROC) curve analysis is used to calculate the area under the curve, sensitivity, and specificity to evaluate the model's early diagnostic efficacy. This helps to identify model shortcomings in a timely manner, allowing for optimization and improvement, thereby enhancing the model's diagnostic performance.
[0034] System Practicality: The constructed early ovarian cancer diagnostic system demonstrates excellent practicality. Through the collaborative work of its various modules, it can quickly and accurately output early ovarian cancer diagnostic results for the tested subjects. Diagnostic results include indicators such as area under the curve, sensitivity, and specificity, providing clinicians with intuitive and reliable diagnostic evidence and aiding in clinical decision-making.
[0035] Example: A method and system for constructing an early diagnostic model for ovarian cancer based on multimodal data. This application explores the diagnostic value of a combined model for early ovarian cancer constructed based on DCE-MRI quantitative parameters, serum human epididymal protein 4 (HE4), and carbohydrate antigen 125 (CA125). This application selected 102 patients with pathologically diagnosed ovarian cancer at the Fifth Affiliated Hospital of Xinjiang Medical University from June 2022 to June 2025 as the observation group, and simultaneously selected 50 patients with benign ovarian tumors confirmed by surgery and pathology during the same period as the control group. Inclusion criteria were: ① Patients in the observation group met the diagnostic criteria for ovarian cancer, and all patients had complete preoperative pelvic MRI imaging data; ② Serum samples were collected preoperatively and HE4 and CA125 tests were completed; ③ Clinical medical records were complete, and all patients signed informed consent forms. Exclusion criteria were: ① Patients with other malignant tumors or severe organ dysfunction; ② Patients who had recently received radiotherapy, chemotherapy, or immunotherapy, which may affect serological indicators; ③ Patients with severe artifacts or poor quality MRI images affecting DCE-MRI quantitative parameter measurement; ④ Patients with incomplete clinical information or who refused to participate in the study. This application was approved by the hospital's medical ethics committee (K-2025116). This application is a retrospective analysis. All patients' age, body mass index, pathological type, and FIGO staging (International Federation of Gynecology and Obstetrics Staging) clinical and pathological data were retrieved from the electronic medical record system of the Fifth Affiliated Hospital of Xinjiang Medical University. Pathological type and FIGO staging information were confirmed and recorded using the patient's final postoperative pathology report and surgical record as the gold standard. Early-stage ovarian cancer refers to FIGO stages I and II. Stage I tumors are completely confined to one or both ovaries, while stage II tumors have invaded other pelvic structures such as the uterus or bladder. Advanced-stage ovarian cancer refers to widespread disease spread, including FIGO stages III and IV. Stage III means that cancer cells have spread to the peritoneum or retroperitoneal lymph nodes, while stage IV shows metastasis to distant organs such as pleural effusion.
[0036] MRI imaging examination All subjects were placed in a supine position and scanned using a GE Discovery MR750 3.0T magnetic resonance scanner. Multiparameter imaging of the pelvic cavity was performed. Standard sequences included transverse T1WI and T2W (I-parameters: TR 4000 ms, TE 102 ms, slice thickness 5 mm, FOV 280 mm × 280 mm). Diffusion-weighted imaging (DWI) used a single-shot excitation planar echo imaging sequence with the following parameters: TR 2400 ms, TE 59 ms, slice thickness 5 mm, interslice spacing 1 mm, matrix 256 × 256, FOV 306 mm × 306 mm, 4 excitations, and a b-value of 50 s / mm. 2 With 1000 s / mm 2 DCE-MRI employed a three-dimensional volumetric interpolation fast gradient echo sequence (TR 4.5 ms, TE 2.2 ms, slice thickness 3 mm). Gadolinium contrast agent (0.1 mmol / kg) was injected via the antecubital vein at a flow rate of 2.5 mL / s before multi-phase acquisition was initiated. Raw data were uploaded to an AW VolumeShare 5 workstation for post-processing. The system automatically generated ADC maps, which were independently measured by two associate chief physicians with over 10 years of experience in female pelvic imaging diagnosis, without knowledge of clinical grouping information. On DWI images, necrotic, cystic, and vascular artifacts were avoided, and a two-dimensional region of interest (ROI) was delineated at the level where the solid component of the lesion was most prominent. The software automatically calculated the average ADC value.
[0037] DCE-MRI data were analyzed for pharmacokinetic fitting using an extended Tofts model. Three-dimensional regions of interest (VOIs) were manually delineated layer by layer on enhanced T1WI images to maximize coverage of the entire tumor volume, excluding necrotic, cystic, and hemorrhage areas. Workstation software automatically generated functional quantitative perfusion parameters, including K... trans The mean values of Kep and Ve were calculated. To assess the repeatability and consistency of parameter measurements, 20 patients were randomly selected and the measurements were repeated once. The intraclass correlation coefficient (ICC) was calculated. A ICC > 0.80 was considered to indicate good consistency in parameter extraction, which was then used for subsequent statistical analysis.
[0038] Serum HE4 and CA125 level detection Three mL of venous blood was collected from the patient, and the serum was separated by centrifugation at 3000 rpm for 10 minutes and stored at -20°C. The levels of CA125 and HE4 were quantitatively analyzed by enzyme-linked immunosorbent assay.
[0039] Grouping Patients in the observation group were divided into an early group and a late group according to the FIGO staging system. The early group consisted of FIGO stages I and II, and the late group consisted of FIGO stages III and IV.
[0040] Statistical methods SPSS 27.0 statistical software was used for data processing and analysis. Normally distributed measurement data were expressed as mean ± standard deviation and analyzed using t-tests. Comparisons of rates for categorical data were performed using chi-square tests. 2 Risk factors influencing early diagnosis of ovarian cancer were analyzed using logistic regression. Receiver operating characteristic (ROC) curves were plotted using R3.4.3 software based on the multivariate analysis results, and the area under the curve (AUC) was calculated to assess predictive power. P < 0.05 was considered statistically significant.
[0041] Comparison of clinical data between the two groups of patients Comparison of clinical data between the two groups showed no statistically significant differences in age and body mass index (P > 0.05). In the observation group of 102 ovarian cancer patients, the pathological types were distributed as follows: serous adenocarcinoma 32 cases (31.4%), endometrioid adenocarcinoma 25 cases (24.5%), mucinous adenocarcinoma 29 cases (28.4%), and other types 16 cases (15.7%). According to the FIGO staging system, 74 patients (72.55%) were in stage I-II, and 28 patients (27.45%) were in stage III-IV. See Table 1.
[0042] Table 1: Comparison of clinical data between the two groups of patients
[0043] Comparison of serum HE4 and CA125 levels between the observation group and the control group The serum HE4 and CA125 levels in the observation group were higher than those in the control group (P < 0.05). See Table 2.
[0044] Table 2: Comparison of serum HE4 and CA125 levels between the observation group and the control group
[0045] Comparison of serum HE4 and CA125 levels between early-stage and late-stage ovarian cancer patients; serum HE4 and CA125 levels were higher in the late-stage group than in the early-stage group (P<0.05); see Table 3.
[0046] Table 3: Comparison of serum HE4 and CA125 levels in ovarian cancer patients in the early and late stages.
[0047] Comparison of DCE-MRI quantitative parameters between the observation group and the control group: K in patients of the observation group trans The levels of Kep and Ve were higher than those in the control group, while the ADC value was lower than that in the control group (P < 0.05); see Table 4.
[0048] Table 4: Comparison of DCE-MRI quantitative parameters between the observation group and the control group (x±s)
[0049] Comparison of DCE-MRI quantitative parameters between early-stage and late-stage ovarian cancer patients; K in late-stage patients trans The levels of Kep and Ve were higher in the early group than in the early group, while the ADC value was lower in the early group (P < 0.05); see Table 5.
[0050] Table 5: Comparison of DCE-MRI quantitative parameters between early-stage and late-stage ovarian cancer patients
[0051] Efficacy analysis of DCE-MRI quantitative parameters combined with serum HE4 and CA125 levels for early diagnosis of ovarian cancer: Receiver operating characteristic (ROC) curves were constructed using R3.4.3 software, with DCE-MRI quantitative parameters, serum HE4, and CA125 levels as indicators, to evaluate their efficacy in early diagnosis of ovarian cancer. Results showed that each individual indicator had a certain ability to identify early ovarian cancer, with the combined model exhibiting the highest area under the curve (AUC) of 0.986 (95% CI: 0.967–1.000), a sensitivity of 95.9% (95% CI: 88.6–99.2) and a specificity of 96.4% (95% CI: 81.7–99.9), and a Youden index of 0.923, significantly superior to individual indicators. (See Figure 1 and Table 6.)
[0052] Table 6: Efficacy analysis of DCE-MRI quantitative parameters combined with serum HE4 and CA125 levels in the early diagnosis of ovarian cancer
[0053] Discussion of Results: Ovarian cancer is one of the most common malignant tumors of the female reproductive system with a poor prognosis. Its early clinical symptoms are often subtle, and approximately 70% of patients are diagnosed at an advanced stage, severely impacting treatment outcomes and survival rates. Existing research indicates that early diagnosis of ovarian cancer is crucial for improving patient prognosis, significantly reducing recurrence rates and increasing overall survival. Therefore, there is an urgent need to explore more precise and non-invasive detection methods. Currently, commonly used serological markers such as HE4 and CA125 play an important role in the development and progression of ovarian cancer, but their sensitivity and specificity in early diagnosis remain limited when used alone. MRI, in addition to displaying tumor morphology and extent, can also obtain functional quantitative parameters reflecting changes in tumor microvascular permeability, perfusion status, and cell density through dynamic contrast-enhanced imaging and diffusion-weighted imaging, providing new objective evidence for early identification. Therefore, this study aims to integrate DCE-MRI quantitative parameters with serum HE4 and CA125 levels to construct an early diagnostic model for ovarian cancer and systematically evaluate its diagnostic efficacy and clinical application value, in order to provide a reliable reference for early screening and precision treatment. Serum HE4 and CA125 are important tumor markers in ovarian cancer, and their expression levels are closely related to the presence and progression of the tumor. In this application, the concentrations of these markers in the ovarian cancer patient group were significantly higher than those in the benign lesion control group. Furthermore, within the ovarian cancer group, the expression levels in FIGO stage III / IV patients were significantly higher than those in FIGO stage I / II patients. This gradient-like change can be explained by the biological behavior of the tumor. Malignant ovarian epithelial tumor cells typically exhibit overexpression of HE4 and CA125 genes and hypersecretion of proteins, leading to serum concentrations significantly exceeding those in benign lesions when malignant tumors are present. More importantly, with the progression of FIGO staging, the tumor burden increases significantly, the tumor volume increases, and the blood supply becomes richer, directly resulting in the release of more markers into the bloodstream. Advanced lesions are often accompanied by extensive peritoneal implantation metastases; these metastatic lesions also possess secretory functions, and increased tumor invasiveness leads to altered vascular permeability, further promoting the leakage of markers into the circulatory system. Therefore, serum HE4 and CA125 levels can not only effectively distinguish between benign and malignant ovarian tumors, but their concentration changes also objectively reflect the degree of tumor progression and tumor burden in the body, becoming powerful serological indicators for assessing disease severity. The changing patterns of dynamic contrast-enhanced MRI parameters and ADC values profoundly reflect the pathophysiological changes during the development of ovarian cancer. This study found that compared with the benign control group, the ovarian cancer group had significantly increased Ktrans, Kep, and Ve values, while the ADC value was significantly decreased; further, within the cancer group, this trend was more pronounced in the advanced stage group than in the early stage group.The underlying mechanism lies in the fact that the growth and progression of malignant tumors are highly dependent on angiogenesis. These new blood vessels have abnormal structures, incomplete walls, and increased permeability, leading to a significant increase in the transport rate and leakage of contrast agents within and outside the blood vessels, manifested as an increase in Ktrans and Kep. Simultaneously, tumor cell proliferation increases their density and atypia, relatively compressing the extracellular space, all contributing to an increase in Ve values. On the other hand, increased cell density, increased nucleocytoplasmic ratio, and crowding of intracellular organelles collectively restrict the free diffusion of water molecules, resulting in a decrease in ADC values. As tumors progress to advanced stages, angiogenesis becomes more abundant, cell proliferation becomes denser, and invasiveness becomes stronger, making these microenvironmental characteristics increasingly pronounced, ultimately manifesting as quantitative changes in imaging parameters that are positively correlated with tumor malignancy and burden.
[0054] The multimodal combined model constructed in this application demonstrated excellent efficacy in the identification of early-stage ovarian cancer, with a near-perfect AUC and high levels of sensitivity and specificity, showing significant potential for clinical translation. The advantage of the combined model lies in its integration of functional quantitative parameters from DCE-MRI and serum tumor markers. CA125 and HE4 enable the simultaneous reflection of local lesion microcirculation status, cell density changes, and systemic tumor biological characteristics. DCE-MRI parameters can quantitatively reflect the microvascular permeability, perfusion status, and cell diffusion restriction at the lesion site, while serum indicators provide systematic evidence related to tumor invasiveness, stage, and burden. These two types of information complement each other, reducing the risk of misdiagnosis and missed diagnosis caused by single tests, and laying a reliable foundation for the future development of a safe, non-invasive, and highly accurate early ovarian cancer auxiliary decision-making tool.
[0055] In conclusion, the DCE-MRI quantitative parameter combined with serum HE4 and CA125 model constructed in this study showed high sensitivity and specificity in the early diagnosis of ovarian cancer, suggesting that it has good clinical application prospects as a non-invasive and objective auxiliary tool.
[0056] The key design focus of this invention is to comprehensively utilize multimodal data to construct an early diagnostic model for ovarian cancer, and to improve the accuracy and reliability of diagnosis through accurate feature extraction and model evaluation. Its advantages lie in the fusion of multimodal data, the precision of feature extraction, the scientific rigor of model evaluation, and the practicality of the system, providing effective support for the early diagnosis of ovarian cancer. By integrating pelvic MRI imaging data with serum sample data, it overcomes the limitations of single diagnostic methods. The combination of imaging features and serum biomarker levels reflects the characteristics of ovarian lesions from different dimensions, significantly improving the accuracy of early diagnosis and effectively distinguishing early ovarian cancer lesions from benign ovarian tumors. This diagnostic system is highly practical; its modules work collaboratively to quickly output diagnostic results, and the included multiple diagnostic indicators provide clinicians with intuitive and reliable decision-making support.
[0057] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the technical scope of the present invention. Therefore, any minor modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.
Claims
1. A method for constructing an early diagnostic model for ovarian cancer based on multimodal data, characterized in that: Includes the following steps: S1. Obtain a sample dataset, which includes pelvic MRI images and serum samples from patients with pathologically confirmed ovarian cancer and patients with benign ovarian tumors. S2. Extract DCE-MRI quantitative parameters and ADC values from the pelvic MRI image data; S3. Detect the levels of serum human epididymal protein 4 and carbohydrate antigen 125 from the serum sample; S4. Using the DCE-MRI quantitative parameters, ADC value, serum human epididymal protein 4 level and carbohydrate antigen 125 level as input features, a multivariate logistic regression model for early diagnosis of ovarian cancer is constructed.
2. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 1, characterized in that: In step S2, the DCE-MRI quantitative parameters are obtained by fitting dynamic contrast-enhanced MRI data to an extended Tofts pharmacokinetic model; the ADC value is obtained by acquiring and calculating diffusion-weighted imaging sequences.
3. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 2, characterized in that: The extraction of DCE-MRI quantitative parameters and ADC values in step S2 specifically involves: delineating the region of interest or three-dimensional volumetric region of interest from the pelvic MRI image data; delineating a two-dimensional region of interest at the level where the solid component of the lesion is most prominent on the diffusion-weighted imaging image for ADC value measurement; and delineating the lesion boundary layer by layer on the conventional sequence enhanced T1WI image of the pelvic MRI image data to construct a three-dimensional volumetric region of interest for DCE-MRI quantitative parameter measurement, while excluding necrotic, cystic, and hemorrhage areas.
4. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 1, characterized in that: The quantitative parameters of DCE-MRI in step S2 include volume transport constant, reflux rate constant, and extracellular extravascular space fraction.
5. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 1, characterized in that: The serum levels of human epididymal protein 4 and carbohydrate antigen 125 were quantitatively analyzed by enzyme-linked immunosorbent assay (ELISA).
6. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 1, characterized in that: The ovarian cancer early diagnosis model is used to distinguish early ovarian cancer lesions from benign ovarian tumors; the ovarian cancer early diagnosis model is a multivariate logistic regression model, and its output diagnostic results include the probability value of the subject having ovarian cancer.
7. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 1, characterized in that: Step S4 further includes: calculating the area under the curve, sensitivity, and specificity through receiver operating characteristic (ROC) curve analysis to evaluate the early diagnostic efficacy of the diagnostic model.
8. The method for constructing an early diagnostic model for ovarian cancer based on multimodal data according to claim 1, characterized in that, Step S4 further includes: performing statistical analysis on the input features and selecting features with differences as model variables.
9. An early diagnostic system for ovarian cancer constructed using the method described in any one of claims 1-8, characterized in that: include: The image data acquisition module is used to acquire pelvic MRI image data of the subject under test; The serum data acquisition module is used to acquire the serum levels of human epididymal protein 4 and carbohydrate antigen 125 in the subjects to be tested. The feature extraction module is used to extract DCE-MRI quantitative parameters and ADC values from the pelvic MRI image data; The model processing module contains a diagnostic model trained by multivariate logistic regression based on the quantitative parameters of DCE-MRI, ADC value, serum human epididymal protein 4 level, and carbohydrate antigen 125 level, which is used to output the early diagnosis results of ovarian cancer for the test subjects.
10. The early diagnostic system for ovarian cancer according to claim 9, characterized in that: The diagnostic results include at least one of the area under the receiver operating characteristic (ROC) curve, sensitivity, and specificity; the extraction of DCE-MRI quantitative parameters in the feature extraction module is based on fitting DCE-MRI data to an extended Tofts pharmacokinetic model.