Comprehensive prediction system for pathological typing of craniopharyngeal tubuloma based on preoperative radiomics and cerebrospinal fluid genomics

Abstract

The invention discloses a construction method and a prediction system of a craniopharyngeal tubuloma pathological typing prediction model based on preoperative radiomics and cerebrospinal fluid genomics. The model construction method comprises the following steps: acquiring iconography data; semantic features and radiomics features are extracted; carrying out feature screening and model training, and constructing a Voting image scoring model; mutation of the BRAF V600E is detected through cerebrospinal fluid genomics; and performing feature fusion according to the Voting score and a cerebrospinal fluid genomics detection result, inputting a classifier to obtain a classification result of craniopharyngeal tubuloma prediction, comparing the prediction result with an actual result, and optimizing the classifier to construct a comprehensive prediction model. According to the method, an artificial intelligence model for predicting pathological typing based on preoperative radiomics of a patient with the craniopharyngeal tubuloma is combined with BRAF V600E mutation detection of cerebrospinal fluid cfDNA for guiding pathological typing prediction of the craniopharyngeal tubuloma for the first time; the prediction model has good prediction capability.

Description

Technical Field

[0001] The present invention relates to a method for constructing a craniopharyngioma pathological typing prediction model and a prediction system based on preoperative imaging genomics and cerebrospinal fluid genomics, and belongs to the field of biomedical detection technology. Background Art

[0002] Craniopharyngiomas are embryonic residual tumors and the most common congenital benign intracranial tumor. Based on their origin and pathological features, craniopharyngiomas can be divided into two subtypes: ameloblastic and papillary. The former arises from the epithelial cells of the craniopharyngeal duct or remnants of Rathke's pouch, while the latter arises from metaplasia of squamous epithelial cells remaining in the primitive oral cavity. Clinical manifestations of craniopharyngiomas often include headaches, visual impairment, and polydipsia and polyuria caused by central diabetes insipidus. Children may experience developmental delay, while adults may experience sexual dysfunction and hypothalamic syndromes (such as impaired thermoregulation and water and electrolyte balance).

[0003] Endoscopic transsphenoidal resection of the tumor is currently the treatment of choice for craniopharyngiomas. However, adhesions adjacent to the tumor invade numerous important surrounding neurovascular structures (such as the hypothalamus, optic nerves, and bilateral internal carotid arteries), making this type of surgery extremely risky and often difficult to completely resect. Furthermore, patients have a high incidence of perioperative complications, including intraoperative vascular damage leading to massive bleeding, postoperative secondary infection, cerebrospinal fluid rhinorrhea, hypothalamic dysfunction (diabetes insipidus, central hyperpyrexia, etc.), and hypopituitarism.

[0004] Recent sequencing studies have revealed a high frequency of BRAF V600E mutations in papillary craniopharyngiomas, with a mutation rate of up to 95%. This suggests that BRAF mutation inhibitors may be potential therapeutic agents for papillary craniopharyngiomas. In 2016, the first case report demonstrated that the combination of a BRAF V600E mutation inhibitor and a downstream MEK pathway inhibitor significantly reduced tumor size in recurrent papillary craniopharyngiomas. Subsequent case reports have also demonstrated the beneficial effects of this drug regimen on papillary craniopharyngiomas. Therefore, the use of drugs to control the invasion and progression of papillary craniopharyngiomas is expected to become a comprehensive treatment option for papillary craniopharyngiomas.

[0005] The gold standard for diagnosing papillary craniopharyngiomas relies primarily on postoperative pathological examination to determine their histological characteristics. This means that patients with papillary craniopharyngiomas must undergo surgical resection or biopsy to obtain tumor tissue for a definitive diagnosis. However, due to the close relationship between the tumor and surrounding tissues, including the hypothalamus, optic nerve, and bilateral internal carotid arteries, patients experience a high incidence of perioperative complications, including intraoperative vascular damage leading to massive bleeding, postoperative secondary infection, cerebrospinal fluid rhinorrhea, hypothalamic dysfunction (diabetes insipidus, central hyperpyrexia, etc.), and hypopituitarism. Therefore, if papillary craniopharyngiomas can be diagnosed preoperatively through minimally invasive or non-invasive means, it will help promote the promotion of medical treatment options for papillary craniopharyngiomas, reduce surgical and postoperative complications of papillary craniopharyngiomas, and further improve the quality of life of patients with papillary craniopharyngiomas after treatment. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for constructing a prediction model for the pathological classification of craniopharyngioma based on preoperative imaging genomics and cerebrospinal fluid genomics and a prediction system; the present invention comprehensively judges the tumor pathological classification (ameloblastic type or papillary type) of craniopharyngioma patients through feature extraction of preoperative magnetic resonance imaging, CT and other imaging data of patients, combined with genomic sequencing of cerebrospinal fluid cfDNA, so as to use the predicted pathological classification to guide clinical comprehensive treatment.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] The first aspect of the present invention provides a method for constructing a comprehensive prediction model for craniopharyngioma pathological typing based on preoperative imaging genomics and cerebrospinal fluid genomics, characterized by comprising the following steps:

[0009] Step 1: Acquisition and processing of CT and MRI imaging data;

[0010] Step 2: Extract semantic features and radiomics features;

[0011] Step 3: Screening radiomics feature combinations for predicting tumor pathological types and constructing radiomics scoring models;

[0012] Step 4: Construct a radiomics expert scoring model based on the combination of semantic features to predict tumor pathology type;

[0013] Step 5: Calculate the radiomics score and the radiomics expert score based on the radiomics scoring model and the radiomics expert score model, respectively, and construct a voting imaging scoring model based on the radiomics score and the radiomics expert score;

[0014] Step 6: Detection of BRAF V600E mutation by cerebrospinal fluid genomics;

[0015] Step 7: Calculate the voting score based on the voting image scoring model in step 5. Perform feature fusion on the voting score result and the cerebrospinal fluid genomic test result in step 6. Input the result into the classifier to obtain the predicted classification result of whether the craniopharyngioma is papillary craniopharyngioma. Compare the predicted classification result with the actual result, optimize the classifier, and obtain a trained comprehensive prediction model for craniopharyngioma pathological classification.

[0016] Preferably, the semantic features in step 2 include the patient's age, gender, whether there is calcification, whether there is cystic change and tumor type, and the tumor type includes intrasellar type, third ventricle type and other types except intrasellar type and third ventricle type.

[0017] Preferably, in step 2, radiomic features are extracted using Pyradiomics 3.0.1 on a 3D Slicer platform.

[0018] More preferably, the radiomics features include 852 CT imaging features, 852 T1 enhancement imaging features and 873 T1 imaging features; wherein, the 852 CT imaging features are shown in Table 1 of the specification, the 852 T1 enhancement imaging features are shown in Table 2 of the specification, and the 873 T1 imaging features are shown in Table 3 of the specification.

[0019] Preferably, the process of screening the radiomics feature combination in step 3 includes: first, screening the radiomics features from the magnetic resonance sequence using the t-test; then, normalizing the features and searching for the best feature combination using LASSO.

[0020] More preferably, the optimal feature combination includes 169 CT imaging genomics features, 195 T1 enhancement imaging genomics features and 151 T1 imaging genomics features; among them, the 169 CT imaging genomics features are shown in Table 4 of the specification, the 195 T1 enhancement imaging genomics features are shown in Table 5 of the specification, and the 151 T1 imaging genomics features are shown in Table 6 of the specification.

[0021] Preferably, in step 3, a support vector machine model is used to construct a radiomics scoring model.

[0022] Preferably, in step 4, an LDA (Linear Discriminant Analysis) model is used to calculate the radiomics expert score by combining semantic features.

[0023] Preferably, the radiomics score in step 5 includes T1 radiomics score, T1 enhanced radiomics score and CT radiomics score.

[0024] The second aspect of the present invention provides a method for predicting craniopharyngioma pathological typing based on preoperative imaging genomics and cerebrospinal fluid genomics, comprising:

[0025] Obtain CT and MRI imaging data of patients with craniopharyngioma to be examined;

[0026] Extract radiomics feature combinations and semantic feature combinations for predicting tumor pathological types;

[0027] The Voting score was calculated according to the Voting image scoring model;

[0028] Obtain the results of cerebrospinal fluid genomic testing for BRAF V600E mutation;

[0029] The Voting score was combined with cerebrospinal fluid genomics testing for feature fusion and input into the classifier to obtain the classification result of whether the patient had papillary craniopharyngioma.

[0030] A third aspect of the present invention provides a craniopharyngioma pathological classification prediction system based on preoperative imaging genomics and cerebrospinal fluid genomics, the system comprising:

[0031] A data acquisition unit is used to obtain CT and MRI imaging data of the patient with craniopharyngioma to be tested and the results of BRAF V600E mutation obtained by cerebrospinal fluid genomic testing;

[0032] A radiomics feature extraction unit, used to extract radiomics feature combinations for predicting tumor pathological types;

[0033] A semantic feature extraction unit, used to extract semantic feature combinations for predicting tumor pathological types;

[0034] Voting scoring unit, used to calculate voting scores based on the voting image scoring model;

[0035] The classification unit is used to fuse the Voting score with the cerebrospinal fluid genomics test, input the feature into the classifier, and obtain the classification result of whether the patient has papillary craniopharyngioma.

[0036] A fourth aspect of the present invention provides a device for predicting craniopharyngioma pathological typing based on preoperative imaging genomics and cerebrospinal fluid genomics, comprising: a memory and a processor;

[0037] The memory is used to store program instructions;

[0038] The processor is used to call program instructions. When the program instructions are executed, the prediction method as described above is implemented or the comprehensive prediction model for craniopharyngioma pathological typing constructed using the construction method as described above is used to implement the craniopharyngioma pathological typing prediction method.

[0039] A fifth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, it implements the prediction method as described above or implements the craniopharyngioma pathological classification prediction method using the comprehensive prediction model for craniopharyngioma pathological classification constructed using the construction method as described above.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) This paper first constructed an artificial intelligence model based on preoperative imaging genomics to predict pathological classification of craniopharyngioma patients;

[0042] (2) This paper verifies for the first time the reliability of BRAF V600E mutation detection in cerebrospinal fluid cfDNA for predicting the pathological classification of craniopharyngioma;

[0043] (3) This invention is the first to combine an artificial intelligence model based on preoperative imaging genomics to predict pathological classification of craniopharyngioma patients with BRAF V600E mutation detection in cerebrospinal fluid cfDNA to guide the prediction of craniopharyngioma pathological classification;

[0044] (4) The prediction model constructed by the present invention has good prediction ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the comprehensive model prediction process constructed for the present invention.

[0046] Figure 2 Figure 2 is a coronal MR-enhanced image of the pituitary gland of an exemplary patient.

[0047] Figure 3 Figure 2 is a sagittal MR-enhanced image of the pituitary gland of an exemplary patient.

[0048] Figure 4 This is an exemplary transverse CT image of the sella turcica of patient A.

[0049] Figure 5 Figure 2 is an exemplary patient's postoperative pathology report and molecular pathology results. DETAILED DESCRIPTION

[0050] To make the present invention more clearly understood, preferred embodiments are described in detail below with reference to the accompanying drawings.

[0051] Example

[0052] This example provides a method for constructing a prediction model for craniopharyngioma pathological classification based on preoperative imaging genomics and cerebrospinal fluid genomics:

[0053] (1) Preoperative radiomics prediction

[0054] (1) Acquisition and processing of CT and MRI imaging data

[0055] 1) CT images were acquired on a high-resolution 64-slice spiral CT scanner (Discovery CT750 HD; GE Medical Systems) using standard cephalographic imaging mode. Each patient underwent a preoperative cranial CT scan. CT scan parameters were as follows: tube voltage 120 kV, tube current auto-adjustable range 150–350 mA, helical pitch 0.984, field of view 25 cm, matrix size 512 × 512, and slice thickness 1 mm.

[0056] MRI images were acquired on a 3.0 Tesla scanner (Discovery MR 750W; GE Medical Systems) using a 24-channel head matrix coil. Each patient underwent preoperative multiparametric MRI, including conventional sequences (coronal and sagittal T1-weighted imaging and contrast-enhanced coronal and sagittal T1-weighted imaging). MRI sequence parameters were as follows: T1-weighted imaging: repetition time / echo time 400 / 12 milliseconds, field of view 22 cm, matrix size 256 × 224, bandwidth 62.5 kHz, echo train length 30, and slice thickness 2 mm. Contrast-enhanced imaging was performed immediately after the standard dose (0.1 mmol / kg) of gadopentetate dimethylamine, administered via a dorsal or cubital intravenous injection at a flow rate of approximately 3–4 ml / second.

[0057] (2) Semantic feature and radiomics feature extraction

[0058] Semantic features include patient age, gender, presence of calcification, cystic changes, and tumor classification (intrasellar type, other types, and third ventricle type). Calcification, cystic changes, and tumor classification were independently determined by three experienced radiologists. Tumor contours were manually outlined on the 3D Slicer platform to construct regions of interest (ROIs). Subsequently, radiomic features were extracted using Pyradiomics 3.0.1 after enabling resampling and filtering. 852 CT radiomic features, 852 T1 enhancement radiomic features, and 873 T1 radiomic features were used for subsequent model scoring (see Tables 1-3 for specific features):

[0059] Table 1 List of all CT radiomics feature names

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068] Table 2 List of all T1-enhanced radiomics feature names

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077] Table 3 List of all T1 radiomics feature names

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] (3) Feature selection and model training

[0087] A two-stage process was used to select features in the training cohort. First, radiomic features were screened from MRI and CT sequences using the t-test. Then, after normalization, the LASSO was used to search for the optimal feature combination, which included 169 CT radiomic features, 195 T1-enhanced radiomic features, and 151 T1-enhanced radiomic features (see Tables 4-6 for details). Ten-fold cross-validation was used to identify the optimal hyperparameter alpha (α). An LDA (Linear Discriminant Analysis) model was used to calculate radiomic expert scores based on combinations of semantic features to predict tumor pathology. A support vector machine model was used to calculate radiomic scores for each sequence (e.g., T1 radiomic score, T1-enhanced radiomic score, and CT radiomic score) using combinations of selected features to predict tumor pathology. Four scores were obtained for each patient: an expert radiomic score, a CT radiomic score, a T1 radiomic score, and a T1-enhanced radiomic score. A voting method was used to calculate a summary radiomic score.

[0088] Table 4 List of screened CT radiomics feature names

[0089]

[0090]

[0091] Table 5 List of screened T1-enhanced radiomics feature names

[0092]

[0093]

[0094] Table 6 List of screened T1 radiomics feature names

[0095]

[0096]

[0097]

[0098] (II) CSF cfDNA BRAF V600E genotype sequencing

[0099] (1) Sample collection and processing

[0100] 1) Collection of a cerebrospinal fluid sample by lumbar puncture. First, the patient should assume a lateral or sitting position, keeping their back straight and their knees as close to their chest as possible to expand the intervertebral space. Alternatively, the patient can bend forward while sitting to facilitate the procedure. Under aseptic conditions, the physician will mark the target puncture site (usually the interspinous space between the 3rd and 4th or 4th and 5th lumbar vertebrae), perform routine local skin disinfection, and then cover the puncture with a sterile drape. A local anesthetic (such as 2% lidocaine) is then applied to ensure a painless puncture site. Following local anesthesia, a needle (usually 22 or 25 gauge) is slowly and vertically inserted into the marked site, gradually advancing until the needle passes through the dura mater and enters the subarachnoid space. This is achieved by a sudden loss of resistance and a "drop" sensation. Once the needle is positioned correctly, the core of the needle is removed and observation is made for any cerebrospinal fluid dripping. If cerebrospinal fluid is leaking, 10 ml of cerebrospinal fluid is collected using a Streck cfDNA storage tube. After sampling is complete, the needle is removed and the puncture site is covered with sterile gauze. Gently apply pressure for several minutes before applying adhesive tape. The patient should rest in bed for several hours to avoid hypotensive headaches caused by cerebrospinal fluid loss. Strict aseptic procedures must be followed throughout the entire process to ensure sample quality and patient safety.

[0101] 2) After sample collection, deliver the sample to the laboratory as soon as possible. A qualified technician will centrifuge the sample at 950 rpm for 10 minutes at 4°C to pellet cells and cell debris. Carefully transfer the supernatant to a sterile centrifuge tube, avoiding disturbing the pellet. If fine particles are still present in the supernatant, a second high-speed centrifugation can be performed to further remove any remaining cells or debris. The processed supernatant should be quickly frozen and stored at -80°C to maximize the integrity of the cfDNA and prevent degradation or contamination that could interfere with subsequent genetic testing.

[0102] (2) Magnetic bead method for extracting cerebrospinal fluid cfDNA

[0103] Store the opened bottles of magnetic beads and proteinase K in the kit at 4°C. Before use, thoroughly mix the magnetic beads on a shaker for at least 15 seconds. Subsequently, add 6-8 mL of CSF supernatant to a 50 mL centrifuge tube according to the order and ratios listed in Table 1. Incubate at room temperature for 25 minutes on a constant temperature shaker at 100 rpm or vortex for 15 seconds every 3 minutes. After the incubation period, centrifuge the mixture to remove the adhering liquid from the tube. Place the tube on a magnetic rack. Once the solution has clarified and the magnetic beads have accumulated on the tube walls, carefully remove the liquid with a pipette and remove the tube. Add 500 μL of Wash Buffer 1 to the tube, vortex for 15 seconds to resuspend the magnetic beads, and centrifuge briefly to remove the adhering liquid from the tube. Carefully transfer the mixture to a 1.5 mL centrifuge tube with a pipette, transfer it to a magnetic rack, and let it sit for approximately 30-60 seconds. After the mixture has clarified, transfer the supernatant (free of cfDNA) to the large centrifuge tube from the previous step. Rinse the tube walls and cap, transfer any remaining magnetic beads back to the 1.5 mL centrifuge tube, and discard the large centrifuge tube. Repeat this rinse cycle three times. Next, place the mixture on a 1.5 mL magnetic rack for approximately 30-60 seconds. Once the mixture has clarified, discard the supernatant. Add 500 μL of Wash Buffer 1 for another wash, vortex for 15 seconds, centrifuge briefly to remove any adhering liquid from the wall, transfer the tube to a 1.5 mL magnetic rack, and allow the beads to fully adhere to the beads before discarding the supernatant. Next, add 1000 μL of 80% ethanol for another wash, vortex for 15 seconds, centrifuge briefly to remove any adhering liquid from the wall, transfer the tube to a 1.5 mL magnetic rack, and allow the beads to fully adhere to the beads before discarding the supernatant. Repeat the ethanol wash step once more, then place the tube on the magnetic rack and air dry until the ethanol has completely evaporated (approximately 5-8 minutes). Add 50 μL of elution buffer to the centrifuge tube and incubate in a constant-temperature shaking heater at 56°C, 1150 rpm, with shaking for 5 minutes. Transfer to a magnetic stand and let it rest for 2 minutes. After the magnetic beads are completely adsorbed, carefully transfer the supernatant nucleic acid solution to a new centrifuge tube. After extraction, the concentration and quality of the cfDNA can be measured using a spectrophotometer (such as NanoDrop) or a fluorescent dye method (such as Qubit). Store the extract at -20°C or -80°C for further analysis.

[0104] Table 7 Sample addition information

[0105] Sample volume (mL) Lysis buffer (μL) Proteinase K (μL) Binding solution (μL) Magnetic beads (μL) 1 500 20 900 15 2 1000 40 1800 30 4 2000 80 3600 60

[0106] (3) BRAF V600E mutation detection

[0107] 1) Prepare the reaction mixture (7.5 μL Solution A + 7.5 μL Solution B = 15 μL) and the reaction system (15 μL reaction solution + 15 μL template) according to the manufacturer's instructions. Vortex the reagents thoroughly. First, add 30 μL of the reaction system to the chip, followed by 160 μL of the generated oil. Place a new row of 8-tube strips in the droplet generator, then insert the chip. Turn on the drop marker—the channel will display "Selected"—and start the run. After droplet generation, replace the rubber cap and perform amplification in a standard PCR instrument (Slan 10-12). After amplification, transfer the product to the analyzer.

[0108] 2) Place the 8-tube strip containing the sample to be tested into the microdroplet assay chip adapter. Place the microdroplet assay chip into the adapter and press the chip to ensure it punctures the cap of the 8-tube strip. Add assay oil to the corresponding wells on the chip. Finally, set up the program: Chip reader > Create experiment > Name > Chip ID (unique) > Select Project > Channel Selected > Save > Run. Determine the BRAF V600E mutation based on the instrument's results.

[0109] (III) Comprehensive model predicts the pathological subtype of craniopharyngioma in patients (process as follows Figure 1 shown)

[0110] The Voting score results were combined with the cerebrospinal fluid genomics test results for feature fusion and input into the classifier to obtain a predicted classification result of whether the craniopharyngioma is papillary craniopharyngioma. The predicted classification result was compared with the actual result, and the classifier was optimized to obtain a trained comprehensive prediction model for craniopharyngioma pathological classification. Based on the range of the model score (Voting value), the comprehensive model was divided into three categories:

[0111] (1) Model score Voting < 0.5: When the prediction score is less than 0.5, it indicates that the tumor may be ameloblastic craniopharyngioma (ACP), and simple surgery is recommended;

[0112] (2) 0.5≤Model score Voting≤0.8: When the prediction score is between 0.5 and 0.8, further testing of cfDNA in the cerebrospinal fluid is required to determine whether there is a BRAF V600E mutation: If the cfDNA BRAF V600E in the cerebrospinal fluid is negative, it indicates that the patient may be an ameloblastic craniopharyngioma (ACP), and simple surgical treatment is recommended; if the cfDNA BRAF V600E in the cerebrospinal fluid is positive, it indicates that the patient may be a papillary craniopharyngioma (PCP), and a comprehensive treatment plan is recommended.

[0113] (3) Model score Voting>0.8: When the prediction score is greater than 0.8, it indicates that the tumor may be papillary craniopharyngioma (PCP), and it is recommended to directly adopt a comprehensive treatment plan.

[0114] (IV) Comprehensive model verification analysis

[0115] A total of 20 patient samples from Huashan Hospital affiliated to Fudan University were collected as test set samples. Postoperative pathological examination results showed that there were 10 cases of papillary craniopharyngioma and 10 cases of ameloblastic craniopharyngioma. The prediction process diagram is shown below. Figure 1 The prediction results are shown in Table 8, and the prediction accuracy is 100%.

[0116] Table 8 Prediction results of patient samples

[0117]

[0118]

[0119] An exemplary pathological image of one of the patients is shown in Figures 2-4 The postoperative pathology report and molecular pathology results of this exemplary patient are shown in Figure 5 As shown in the figure, the patient developed binocular blurred vision without apparent cause in December 2023 and subsequently sought treatment at a local hospital. Pituitary enhancement MRI revealed a suprasellar mass, suggesting a possible craniopharyngioma. Preoperative sellar CT scan, plain MRI scan of the sellar region, and enhanced DICOM data were obtained. Based on the radiomics data, craniopharyngioma subtype prediction was performed, with a Voting result of 0.948091883, suggesting a high likelihood of a papillary craniopharyngioma. The patient underwent surgery on July 29, 2024. Postoperative pathology revealed: a papillary craniopharyngioma (sellar region), locally closely adherent to the brain; and a papillary craniopharyngioma (third ventricle), locally closely adherent to the brain. Both subtypes were BRAF V600E mutant.

[0120] The above description is only a preferred embodiment of the present invention and does not constitute any formal or substantial limitation to the present invention. It should be noted that ordinary technicians in this technical field can make several improvements and supplements without departing from the present invention, and these improvements and supplements should also be regarded as the scope of protection of the present invention.

Claims

1. A method for constructing a comprehensive prediction model for craniopharyngioma pathological typing based on preoperative imaging genomics and cerebrospinal fluid genomics, characterized by: The following steps are involved: Step 1: Acquisition and processing of CT and MRI imaging data; Step 2: Extract semantic features and radiomics features; Step 3: Screening radiomic feature combinations for predicting tumor pathological types and constructing radiomics scoring models; Step 4: Construct a radiomics expert scoring model based on the combination of semantic features to predict tumor pathology type; Step 5: Calculate the radiomics score and the radiomics expert score based on the radiomics scoring model and the radiomics expert score model, respectively, and construct a voting imaging scoring model based on the radiomics score and the radiomics expert score; Step 6: Detection of BRAF V600E mutation by cerebrospinal fluid genomics; Step 7: Calculate the voting score based on the voting image scoring model in step 5. Perform feature fusion on the voting score result and the cerebrospinal fluid genomic test result in step 6. Input the result into the classifier to obtain the predicted classification result of whether the craniopharyngioma is papillary craniopharyngioma. Compare the predicted classification result with the actual result, optimize the classifier, and obtain a trained comprehensive prediction model for craniopharyngioma pathological classification.

2. The construction method according to claim 1, wherein The semantic features in step 2 include the patient's age, gender, whether there is calcification, whether there is cystic change, and tumor type. The tumor type includes intrasellar type, third ventricle type, and other types except intrasellar type and third ventricle type.

3. The construction method according to claim 1, wherein In step 2, radiomic features were extracted using Pyradiomics 3.0.1 on the 3D Slicer platform.

4. The construction method according to claim 1, wherein The process of screening the radiomics feature combination in step 3 includes: first, screening the radiomics features from the magnetic resonance sequence using the t-test; then, after normalizing the features, searching for the best feature combination using LASSO.

5. The construction method according to claim 1, wherein: In step 3, a support vector machine model is used to construct a radiomics scoring model; And / or, in step 4, a linear discriminant analysis model is used to calculate the radiomics expert score by combining semantic features.

6. The construction method according to claim 1, wherein: The radiomics score in step 5 includes T1 radiomics score, T1 enhanced radiomics score and CT radiomics score.

7. A method for predicting craniopharyngioma pathological typing based on preoperative radiomics and cerebrospinal fluid genomics, comprising: Obtain CT and MRI imaging data of patients with craniopharyngioma to be examined; Extract radiomics feature combinations and semantic feature combinations for predicting tumor pathological types; The Voting score was calculated according to the Voting image scoring model; Obtain the results of cerebrospinal fluid genomic testing for BRAF V600E mutation; The Voting score was combined with cerebrospinal fluid genomics testing for feature fusion and input into the classifier to obtain the classification result of whether the patient had papillary craniopharyngioma.

8. A craniopharyngioma pathological classification prediction system based on preoperative imaging genomics and cerebrospinal fluid genomics, characterized by: The system comprises: A data acquisition unit is used to obtain CT and MRI imaging data of the patient with craniopharyngioma to be tested and the results of BRAF V600E mutation obtained by cerebrospinal fluid genomic testing; A radiomics feature extraction unit, used to extract radiomics feature combinations for predicting tumor pathological types; A semantic feature extraction unit, used to extract semantic feature combinations for predicting tumor pathological types; Voting scoring unit, used to calculate voting scores based on the voting image scoring model; The classification unit is used to fuse the Voting score with the cerebrospinal fluid genomics test, input the feature into the classifier, and obtain the classification result of whether the patient has papillary craniopharyngioma.

9. A device for predicting craniopharyngioma pathological typing based on preoperative imaging genomics and cerebrospinal fluid genomics, comprising: memory and processor; The memory is used to store program instructions; The processor is used to call program instructions. When the program instructions are executed, the prediction method described in claim 6 is implemented or the comprehensive prediction model for craniopharyngioma pathological classification constructed using the construction method described in any one of claims 1 to 6 is used to implement the craniopharyngioma pathological classification prediction method.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the prediction method according to claim 7 is implemented or the craniopharyngioma pathological typing comprehensive prediction model constructed by the construction method according to any one of claims 1 to 6 is used to implement the craniopharyngioma pathological typing prediction method.

Images