Thyroid ultrasound image-based PTC (Positive Temperature Coefficient) molecular marker prediction method and equipment

Abstract

The invention discloses a thyroid ultrasound image-based PTC molecular marker prediction method and device, and the method comprises the steps: collecting serum samples of a plurality of patients, and detecting the expression level of related molecular markers; thyroid ultrasound images of a plurality of patients are collected, feature extraction is carried out on the thyroid ultrasound images, and a plurality of image feature data of each thyroid ultrasound image are obtained; training a prediction model according to the obtained expression level of the molecular marker and the obtained image feature data; the prediction model is used for predicting the expression level of the molecular marker based on the image feature data; acquiring a thyroid ultrasound image of a patient to be detected, and performing feature extraction on the thyroid ultrasound image to obtain a plurality of image feature data of the patient; and inputting the image feature data into a prediction model, and predicting the expression level of the corresponding molecular marker by the prediction model. The prediction of the expression level of the molecular marker is realized by utilizing the prediction model based on the ultrasonic image features, the prediction is accurate, and wounds and complications of tissue biopsy are avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method and device for predicting PTC molecular markers based on thyroid ultrasound images. Background Art

[0002] Thyroid cancer is the 8th most common cancer in the world. Its incidence has been on the rise over the past 20 years. The overall incidence of thyroid cancer in China is higher than the global incidence.

[0003] Papillary carcinoma of the thyroid (PTC) is the most common type of thyroid cancer, accounting for about 92% of all new cases. The occurrence and development of PTC is usually related to the gradual accumulation of various genetic changes, including gene mutations and changes in gene expression regulatory factors.

[0004] At present, the PTC molecular markers used clinically mainly include BRAF V600E gene mutation and RAS gene point mutation, RET / PTC rearrangement and TERT promoter mutation, etc., which are mainly obtained by invasive operations such as fine-needle aspiration biopsy (FNAB) or surgical resection of tissue for testing.

[0005] Due to the influence of molecular mutations on thyroid histological characteristics, PTC with different molecular mutations may show different ultrasound features. Many studies have explored the relationship between PTC molecular mutations and ultrasound features, which helps to predict the aggressiveness of PTC and patient prognosis earlier from an imaging perspective.

[0006] The existing technologies include: DNA sequence analysis, PCR-RFLP, ARMS-PCR, real-time fluorescence PCR, PCR-DHPLC, digital PCR and DNA microarray technology through laboratory examination.

[0007] DNA sequence analysis can detect PTC gene mutations through PCR amplification and DNA sequence analysis of the exon region of the PTC gene. This is the most accurate method and can detect all known and unknown mutations. For DNA sequence analysis, the conditions of general hospitals below the third level are difficult to meet, and professional gene sequencing instruments and data analysis capabilities are required.

[0008] The PCR-RFLP method uses the inactivation of restriction endonuclease recognition sites to determine whether a mutation exists by changes in the electrophoresis band. This method is applicable to known point mutations.

[0009] The ARMS-PCR method uses the principle that specific primers can only be complementary to mutant sequences or wild-type sequences, and determines whether a product is amplified by electrophoresis bands. This method is also applicable to known point mutations.

[0010] For the PCR-RFLP method and ARMS-PCR method, the nucleic acid amplification conditions in hospitals below level 3 can basically be met, but the gel imaging system and electrophoresis conditions may be slightly insufficient.

[0011] Real-time fluorescence PCR uses fluorescent probe technology to detect PCR products, which can quickly and qualitatively detect some hotspot mutations. For real-time fluorescence PCR, many hospitals below the third level are already equipped with real-time fluorescence quantitative PCR instruments, which can fully carry out real-time fluorescence PCR detection for hotspot mutations.

[0012] The PCR-DHPLC method uses denaturing high-performance liquid chromatography technology to detect PCR products and can screen for unknown mutations. The technical threshold of the PCR-DHPLC method is high and requires a dedicated DHPLC instrument, which is difficult to meet in hospitals below the third level.

[0013] Digital PCR uses the limiting dilution principle to absolutely quantify PCR products and can detect chromosome copy number variation. Digital PCR is similar to conventional PCR in principle, but the instrument is more customized and difficult to be equipped in hospitals below the third level.

[0014] DNA microarray technology: By designing mutant and wild-type probes, hybridization reactions are performed to detect whether a sample has a mutation. As a high-throughput detection technology, the cost and instrument conditions of DNA microarray technology are difficult for hospitals below the third level to meet. Summary of the invention

[0015] In order to solve the above technical problems, the present invention proposes a method and device for predicting PTC molecular markers based on thyroid ultrasound images.

[0016] In order to achieve the above object, the technical solution of the present invention is as follows:

[0017] In one aspect, the present invention discloses a method for predicting PTC molecular markers based on thyroid ultrasound images, comprising:

[0018] Step S1: collecting serum samples from a number of patients and detecting the expression levels of relevant molecular markers, wherein the aforesaid patients include patients with papillary thyroid cancer and patients with benign thyroid diseases;

[0019] Step S2: collecting thyroid ultrasound images of several patients in step S1, and performing feature extraction on the images to obtain several image feature data of each thyroid ultrasound image;

[0020] Step S3: training a prediction model according to the expression level of the molecular marker obtained in step S1 and the image feature data obtained in step S2;

[0021] The prediction model is used to predict the expression level of molecular markers based on image feature data;

[0022] Step S4: collecting a thyroid ultrasound image of the patient to be tested, and performing feature extraction on the image to obtain a number of image feature data of the patient;

[0023] The image feature data is input into the prediction model, and the prediction model predicts the expression level of the corresponding molecular marker.

[0024] Based on the above technical solution, the following improvements can be made:

[0025] As a preferred solution, in step S1, the relevant molecular markers include: one or more of BRAF gene mutation, TERT gene amplification, RET / PTC rearrangement, and TERT promoter mutation.

[0026] As a preferred solution, in step S2, the plurality of image feature data include: one or more of nodule size features, echo features, nodule morphology features, edge features, calcification features, and blood flow richness features.

[0027] As a preferred solution, the image feature data is obtained by the following method:

[0028] For nodule size characteristics, the three-dimensional maximum diameter was used for quantitative evaluation;

[0029] For echo characteristics, quantitative evaluation was performed by comparing the echo of normal thyroid parenchyma in the same layer;

[0030] The nodule morphology characteristics are obtained by the following methods, including:

[0031] Step A1: Outline the nodule margin on the thyroid ultrasound image to obtain the nodule area;

[0032] Step A2: Measure the area of ​​the nodule and record it as A;

[0033] Step A3: With the geometric center of the nodule area as the center and the maximum axis of the nodule as the radius, draw a circle that conforms to the maximum outer contour of the nodule;

[0034] Step A4: Calculate the area of ​​the circle and record it as B;

[0035] Step A5: Calculate the dispersion coefficient c according to the following formula;

[0036] c = (BA) / B;

[0037] Step A6: quantitatively evaluate the nodule morphological characteristics using the dispersion coefficient;

[0038] The edge features are obtained by the following methods:

[0039] Step B1: Select several representative sections along the maximum axis of the nodule and obtain an image corresponding to each section;

[0040] Step B2: measuring the thickness of several different sites on the edge of the nodule on each section image;

[0041] Step B3: Calculate the ratio of the unclear edge area of ​​each site to the maximum diameter of the entire nodule, and obtain several ratios corresponding to each site;

[0042] Step B4: Calculate the average of all ratios as the edge definition index of the image of the section;

[0043] Step B5: Calculate the average edge clarity index of all slice images as the comprehensive edge clarity index of the nodule;

[0044] Step B6: quantitatively evaluating edge features using a comprehensive edge clarity index;

[0045] For calcification features, the following methods are used to obtain the features:

[0046] Step C1: adjusting the parameters of the ultrasound image to optimize the display of calcification foci;

[0047] Step C2: outlining the calcification area on the ultrasound image;

[0048] Step C3: Calculate the area of ​​the calcified region and record it as D;

[0049] Step C4: Measure the area of ​​the entire lesion and record it as E;

[0050] Step C5: Calculate the calcification area ratio f according to the following formula;

[0051] f = D / E;

[0052] Step C6: quantitatively evaluating calcification characteristics using the calcification area ratio;

[0053] A semi-quantitative method was used to quantify the blood flow richness characteristics.

[0054] As a preferred solution, in step S3, a prediction model is established using a support vector machine or a random forest algorithm.

[0055] In addition, on the other hand, the present invention also discloses a PTC molecular marker prediction device based on thyroid ultrasound images, comprising:

[0056] A collection module, used for collecting serum samples from a number of patients and detecting the expression levels of relevant molecular markers, wherein the aforesaid patients include: patients with papillary thyroid cancer and patients with benign thyroid diseases;

[0057] An acquisition module is used to acquire thyroid ultrasound images of several patients and perform feature extraction on the images to obtain several image feature data of each thyroid ultrasound image;

[0058] A training module, used to train a prediction model based on the expression levels of molecular markers obtained by the collection module and the image feature data obtained by the acquisition module;

[0059] The prediction model is used to predict the expression level of molecular markers based on image feature data;

[0060] The prediction module is used to collect the thyroid ultrasound image of the patient to be tested, perform feature extraction on the image, and obtain a number of image feature data of the patient;

[0061] The image feature data is input into the prediction model, and the prediction model predicts the expression level of the corresponding molecular marker.

[0062] As a preferred solution, in the collection module, the relevant molecular markers include: one or more of BRAF gene mutation, TERT gene amplification, RET / PTC rearrangement, and TERT promoter mutation.

[0063] As a preferred solution, in the acquisition module, the plurality of image feature data include: one or more of nodule size features, echo features, nodule morphology features, edge features, calcification features, and blood flow richness features.

[0064] As a preferred solution, image feature data is obtained through the following modules:

[0065] Nodule size feature assessment module, used to quantify nodule size features using three-dimensional maximum diameter;

[0066] An echo characteristic evaluation module is used to quantitatively evaluate the echo characteristics by comparing the echo of normal thyroid parenchyma in the same layer;

[0067] Nodule morphological characteristics assessment module, including:

[0068] A nodule region acquisition unit is used to outline the nodule margin on the thyroid ultrasound image and acquire the nodule region;

[0069] Nodule area measurement unit, used to measure the area of ​​the nodule area, recorded as A;

[0070] A circle drawing unit is used to draw a circle that conforms to the maximum outer contour of the nodule, with the geometric center of the nodule area as the center and the maximum axis of the nodule as the radius;

[0071] The circle area calculation unit is used to calculate the area of ​​the circle, which is recorded as B;

[0072] The discrete coefficient calculation unit is used to calculate the discrete coefficient c according to the following formula;

[0073] c = (BA) / B;

[0074] A nodule morphological feature evaluation unit, used for quantitatively evaluating nodule morphological features using a dispersion coefficient;

[0075] Edge feature evaluation module, including:

[0076] A section image acquisition unit, used for selecting a number of representative sections along the maximum axis of the nodule to obtain an image corresponding to each section;

[0077] A site thickness measurement unit is used to measure the thickness of several different sites on the edge of the nodule on each section image;

[0078] A ratio calculation unit is used to calculate the ratio of the unclear edge area of ​​each site to the maximum diameter of the entire nodule, and obtain a number of ratios corresponding to each site;

[0079] A first averaging unit is used to calculate the average of all ratios as an edge definition index of the image of the section;

[0080] The second averaging unit is used to calculate the average of the edge clarity indexes of all the slice images as the comprehensive edge clarity index of the nodule;

[0081] An edge feature evaluation unit, used for quantitatively evaluating edge features using a comprehensive edge clarity index;

[0082] Calcification feature assessment module, including:

[0083] A parameter adjustment unit, used to adjust the parameters of the ultrasound image and optimize the display of calcification foci;

[0084] A calcification area delineation unit, used for delineating a calcification area on an ultrasound image;

[0085] A calcification area calculation unit is used to calculate the area of ​​the calcification area, which is recorded as D;

[0086] The lesion area measurement unit is used to measure the area of ​​the entire lesion area, which is recorded as E;

[0087] The calcification area ratio calculation unit is used to calculate the calcification area ratio f according to the following formula;

[0088] f = D / E;

[0089] A calcification feature evaluation unit, used for quantitatively evaluating calcification features using a calcification area ratio;

[0090] The blood flow richness characteristic evaluation module is used to quantitatively evaluate the blood flow richness characteristics using a semi-quantitative method.

[0091] As a preferred solution, the training module uses a support vector machine or a random forest algorithm to establish a prediction model.

[0092] The present invention discloses a method and device for predicting PTC molecular markers based on thyroid ultrasound images, which has the following beneficial effects:

[0093] First, the present invention establishes a correlation prediction model between ultrasound image features and the expression levels of molecular markers, and uses the prediction model to predict the expression levels of molecular markers, and the prediction is accurate.

[0094] Second, the present invention is based on ultrasound image features and is a non-invasive test, which avoids the trauma and complications of tissue biopsy. It also has low equipment requirements and is compatible with conventional ultrasound equipment in current medical institutions at all levels, making it easy to popularize in hospitals below the tertiary level. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0096] Figure 1 This is a flow chart of a method for predicting PTC molecular markers provided in an embodiment of the present invention.

[0097] Figure 2 A schematic diagram of a support vector machine provided in an embodiment of the present invention.

[0098] Figure 3 A schematic diagram of a random forest provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0099] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0100] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0101] Using ordinal numbers “first,” “second,” “third,” etc. to describe common objects merely indicates that different instances of similar objects are involved and is not intended to imply that the objects so described must have a given order in time, space, order, or in any other manner.

[0102] In addition, the expression of “comprising” an element is an “open” expression, which merely means that corresponding components or steps exist, and should not be interpreted as excluding additional components or steps.

[0103] In order to achieve the purpose of the present invention, in some embodiments of a method and device for predicting PTC molecular markers based on thyroid ultrasound images, as Figure 1 As shown, the PTC molecular marker prediction method includes:

[0104] Step S1: collecting serum samples from a number of patients and detecting the expression levels of relevant molecular markers, wherein the aforesaid patients include patients with papillary thyroid cancer and patients with benign thyroid diseases;

[0105] Step S2: collecting thyroid ultrasound images of several patients in step S1, and performing feature extraction on the images to obtain several image feature data of each thyroid ultrasound image;

[0106] Step S3: training a prediction model according to the expression level of the molecular marker obtained in step S1 and the image feature data obtained in step S2;

[0107] The prediction model is used to predict the expression level of molecular markers based on image feature data;

[0108] Step S4: collecting a thyroid ultrasound image of the patient to be tested, and performing feature extraction on the image to obtain a number of image feature data of the patient;

[0109] The image feature data is input into the prediction model, and the prediction model predicts the expression level of the corresponding molecular marker.

[0110] Furthermore, in step S1, the relevant molecular markers include: BRAF gene mutation, TERT gene amplification, RET / PTC rearrangement, and TERT promoter mutation.

[0111] BRAF V600E gene mutation: This is the most common mutation in PTC, occurring in approximately 29-83% of PTC patients. This mutation leads to overactivation of the MAPK signaling pathway, promoting cancer cell proliferation. BRAF V600E mutation is associated with the aggressive clinical pathological features of PTC.

[0112] RAS gene point mutation: RAS gene constitutes the upstream of RAS / MAPK signaling pathway, and mutation activates downstream MEK and ERK to promote tumorigenesis. RAS mutation can be implanted in about 10-20% of PTC. Among them, KRAS mutation is relatively rare, while NRAS and HRAS mutations are more common.

[0113] RET / PTC rearrangement: This gene rearrangement can lead to coding region fusion, activate the RET protein kinase domain, enhance downstream signaling pathway conduction, and stimulate cell proliferation. The detection rate of RET / PTC rearrangement accounts for about 20-40% of PTC.

[0114] TERT promoter mutation: This will increase the expression of telomerase reverse transcriptase, promote telomerase activity, and maintain cell proliferation. TERT mutation is mutually exclusive with BRAF and RAS mutations and can be implanted in 10-20% of PTC.

[0115] Detection of the above-mentioned related molecular markers is helpful for risk classification and targeted treatment, but traditional detection methods require obtaining samples through tissue biopsy, and the conditions in general county-level hospitals are difficult to meet, and professional gene sequencing equipment and data analysis capabilities are required.

[0116] The present invention predicts the levels of different molecular markers by using the ultrasonic image features through the correlation between different molecular markers of PTC and ultrasonic image features.

[0117] Furthermore, in step S2, the plurality of image feature data include: nodule size feature, echo feature, nodule morphology feature, edge feature, calcification feature, and blood flow richness feature.

[0118] It is worth noting that image feature data can be obtained by:

[0119] For nodule size characteristics, the three-dimensional maximum diameter was used for quantitative evaluation;

[0120] When the maximum diameter of a nodule is ≤5 mm, it is a micronodule;

[0121] When 5mm<the maximum diameter of the nodule≤10mm, it is a small nodule;

[0122] When 10mm<the maximum diameter of the nodule≤20mm, it is a medium nodule;

[0123] When the maximum diameter of a nodule is greater than 20 mm, it is considered a large nodule.

[0124] For echo characteristics, quantitative evaluation is performed by comparing the echo of normal thyroid parenchyma in the same layer. According to the intensity classification, levels 1-3 represent low echo, medium echo, and high echo, respectively. Mixed echo contains more than two echoes.

[0125] The nodule morphology characteristics are obtained by the following methods, including:

[0126] Step A1: Outline the nodule margin on the thyroid ultrasound image to obtain the nodule area;

[0127] Step A2: Measure the area of ​​the nodule and record it as A;

[0128] Step A3: With the geometric center of the nodule area as the center and the maximum axis of the nodule as the radius, draw a circle that conforms to the maximum outer contour of the nodule;

[0129] Step A4: Calculate the area of ​​the circle and record it as B;

[0130] Step A5: Calculate the dispersion coefficient c according to the following formula;

[0131] c = (BA) / B;

[0132] Step A6: Use the dispersion coefficient to quantitatively evaluate the nodule morphological characteristics.

[0133] The larger the dispersion coefficient value, the more irregular the nodule; conversely, the smaller the dispersion coefficient value, the more regular the nodule and the smoother the edge. Generally, the dispersion coefficient of PTC is greater than 20%.

[0134] The edge features are obtained by the following methods:

[0135] Step B1: Select three representative sections along the maximum axis of the nodule and obtain images corresponding to each section;

[0136] Step B2: Measure the thickness of the nodule edge at 10 different locations on each section image;

[0137] Step B3: Calculate the ratio of the unclear edge area of ​​each site to the maximum diameter of the entire nodule, and obtain 10 ratios corresponding to each site;

[0138] Step B4: Calculate the average of the 10 ratios as the edge definition index of the image of the section;

[0139] Step B5: Calculate the average of the edge clarity indexes of the images of the three sections as the comprehensive edge clarity index of the nodule;

[0140] Step B6: Quantitatively evaluate edge features using a comprehensive edge clarity index.

[0141] The comprehensive edge clarity index is between 0 and 1. The smaller the value (i.e. the closer it is to 0), the clearer the edge.

[0142] The PTC index is usually greater than 0.5. Generally speaking, the index of malignant nodules is usually greater than 0.5, while the index of benign nodules is usually less than 0.3.

[0143] For calcification features, the following methods are used to obtain the features:

[0144] Step C1: adjusting parameters on the color Doppler ultrasound image to optimize the display of calcification foci;

[0145] Step C2: delineate the calcification area on the ultrasound image by manual tracing or semi-automatic segmentation;

[0146] Step C3: Calculate the area of ​​the calcified region and record it as D;

[0147] Step C4: Measure the area of ​​the entire lesion and record it as E;

[0148] Step C5: Calculate the calcification area ratio f according to the following formula;

[0149] f = D / E;

[0150] Step C6: Quantitatively evaluate the calcification characteristics using the calcification area ratio.

[0151] PTC can be seen with microcalcifications with a track of <1 mm and an area of ​​approximately <10%.

[0152] A semi-quantitative method was used to quantify the blood flow richness characteristics.

[0153] Grades 0-III represent no blood flow signal, low, moderate, and high blood flow, respectively.

[0154] Level 0: No blood flow signal. No blood flow signal is shown in or around the lesion in color Doppler and power Doppler modes.

[0155] Grade I: Low blood flow. Only color Doppler or power Doppler can show a few blood flow spots or small blood vessel blood flow signals.

[0156] Level II: Moderate blood flow. Both color Doppler and power Doppler can clearly show the lesion and the surrounding areas with more punctate or tubular blood flow signals.

[0157] Level III: High blood flow. A large amount of blood flow signals are visible, the lesion and the surrounding area are filled with blood, the blood flow velocity is fast, and the spectrum is multi-peaked or broad.

[0158] Step S3 uses a machine learning algorithm to establish a model. Specifically, a support vector machine or a random forest algorithm is used to establish a prediction model.

[0159] The corresponding ultrasound image features (i.e., image feature data) and the expression levels of molecular markers are regarded as the x and y variables of the data set, respectively. If there are N collected image data, the data set can be expressed as {x i ,y i}, i=1, 2, ..., N.

[0160] like Figure 2 As shown, the purpose of the support vector machine is to find a hyperplane to segment the samples, and the principle of segmentation is to maximize the interval;

[0161]

[0162] sty i (ωx i +b)≥1, i=1,2,...,N

[0163] Where: ω is the normal vector of the hyperplane;

[0164] b is the intercept.

[0165] like Figure 3 As shown in the figure, random forest is an integrated learning method. The principle of the integrated method is to improve the overall performance by combining multiple models and using a group of models instead of a single model for prediction.

[0166] The basis of random forest is decision trees, which do not rely on a single tree, but get predictions from each tree and select the majority of the results based on the vote to decide the final output. Each tree in the forest is constructed from a sub-dataset obtained by sampling with replacement of the original data set, and at each node of the tree, a subset of features is randomly selected to determine the best split. This randomness ensures that each tree is different, thereby reducing the risk of overfitting. The decision of each tree can be regarded as a vote. When the final decision needs to be made, the random forest will take a majority vote from all the constructed trees and select the most common result.

[0167] The present invention divides the collected ultrasound image features (i.e., image feature data) and the corresponding molecular marker level data into a training set and a test set. The training set data is used to train the support vector machine or random forest model, and the test set data is used to verify the generalization ability of the trained model.

[0168] Step S4 is an application step, which predicts the levels of several molecular markers based on the thyroid ultrasound image of the patient to be tested.

[0169] By utilizing the method of the present invention, complicated gene detection methods can be avoided, which is beneficial for patients and county-level hospitals to make preliminary diagnoses, thereby saving medical resources and providing patients with more guaranteed treatment.

[0170] In summary, the present invention has the following beneficial effects:

[0171] First, non-invasive: The present invention is based only on ultrasonic image features and is a non-invasive test, thus avoiding the trauma and complications of tissue biopsy.

[0172] Second, the results are real-time: the results are obtained in one step, which significantly shortens the cycle of sending biopsy tissue for examination.

[0173] Third, the detection range is wide: the prediction model established by the present invention covers mutation detection of multiple key genes.

[0174] Fourth, low cost: This prediction relies only on conventional ultrasound examination, without the need for additional body or imaging agents, and is cost-effective.

[0175] Fifth, it is easy to operate: after collecting and processing ultrasound images, the prediction can be automatically output, without the need for professional laboratories and manual reading, making it easy to popularize.

[0176] Sixth, high repeatability: There is no human error in the interpretation of ultrasound images, and the prediction results are reliable, accurate and highly repeatable.

[0177] The embodiment of the present invention also discloses a PTC molecular marker prediction device based on thyroid ultrasound images, comprising:

[0178] A collection module, used for collecting serum samples from a number of patients and detecting the expression levels of relevant molecular markers, wherein the aforesaid patients include: patients with papillary thyroid cancer and patients with benign thyroid diseases;

[0179] An acquisition module is used to acquire thyroid ultrasound images of several patients and perform feature extraction on the images to obtain several image feature data of each thyroid ultrasound image;

[0180] A training module, used to train a prediction model based on the expression levels of molecular markers obtained by the collection module and the image feature data obtained by the acquisition module;

[0181] The prediction model is used to predict the expression level of molecular markers based on image feature data;

[0182] The prediction module is used to collect the thyroid ultrasound image of the patient to be tested, perform feature extraction on the image, and obtain a number of image feature data of the patient;

[0183] The image feature data is input into the prediction model, and the prediction model predicts the expression level of the corresponding molecular marker.

[0184] Furthermore, in the collection module, the relevant molecular markers include: one or more of BRAF gene mutation, TERT gene amplification, RET / PTC rearrangement, and TERT promoter mutation.

[0185] Furthermore, in the acquisition module, the plurality of image feature data include: one or more of nodule size features, echo features, nodule morphology features, edge features, calcification features, and blood flow richness features.

[0186] Furthermore, image feature data is obtained through the following modules:

[0187] Nodule size feature assessment module, used to quantify nodule size features using three-dimensional maximum diameter;

[0188] An echo characteristic evaluation module is used to quantitatively evaluate the echo characteristics by comparing the echo of normal thyroid parenchyma in the same layer;

[0189] Nodule morphological characteristics assessment module, including:

[0190] A nodule region acquisition unit is used to outline the nodule margin on the thyroid ultrasound image and acquire the nodule region;

[0191] Nodule area measurement unit, used to measure the area of ​​the nodule area, recorded as A;

[0192] A circle drawing unit is used to draw a circle that conforms to the maximum outer contour of the nodule, with the geometric center of the nodule area as the center and the maximum axis of the nodule as the radius;

[0193] The circle area calculation unit is used to calculate the area of ​​the circle, which is recorded as B;

[0194] The discrete coefficient calculation unit is used to calculate the discrete coefficient c according to the following formula;

[0195] c = (BA) / B;

[0196] A nodule morphological feature evaluation unit, used for quantitatively evaluating nodule morphological features using a dispersion coefficient;

[0197] Edge feature evaluation module, including:

[0198] A section image acquisition unit, used for selecting a number of representative sections along the maximum axis of the nodule to obtain an image corresponding to each section;

[0199] A site thickness measurement unit is used to measure the thickness of several different sites on the edge of the nodule on each section image;

[0200] A ratio calculation unit is used to calculate the ratio of the unclear edge area of ​​each site to the maximum diameter of the entire nodule, and obtain a number of ratios corresponding to each site;

[0201] A first averaging unit is used to calculate the average of all ratios as an edge definition index of the image of the section;

[0202] The second averaging unit is used to calculate the average of the edge clarity indexes of all the slice images as the comprehensive edge clarity index of the nodule;

[0203] An edge feature evaluation unit, used for quantitatively evaluating edge features using a comprehensive edge clarity index;

[0204] Calcification feature assessment module, including:

[0205] A parameter adjustment unit, used to adjust the parameters of the ultrasound image and optimize the display of calcification foci;

[0206] A calcification area delineation unit, used for delineating a calcification area on an ultrasound image;

[0207] A calcification area calculation unit is used to calculate the area of ​​the calcification area, which is recorded as D;

[0208] The lesion area measurement unit is used to measure the area of ​​the entire lesion area, which is recorded as E;

[0209] The calcification area ratio calculation unit is used to calculate the calcification area ratio f according to the following formula;

[0210] f = D / E;

[0211] A calcification feature evaluation unit, used for quantitatively evaluating calcification features using a calcification area ratio;

[0212] The blood flow richness characteristic evaluation module is used to quantitatively evaluate the blood flow richness characteristics using a semi-quantitative method.

[0213] Furthermore, the training module uses a support vector machine or a random forest algorithm to establish a prediction model.

[0214] In this embodiment, the specific content of the PTC molecular marker prediction device based on thyroid ultrasound images is similar to the content of the PTC molecular marker prediction method based on thyroid ultrasound images disclosed in the above embodiment, and will not be repeated here.

[0215] The present invention discloses a method and device for predicting PTC molecular markers based on thyroid ultrasound images, which has the following beneficial effects:

[0216] First, the present invention establishes a correlation prediction model between ultrasound image features and the expression levels of molecular markers, and uses the prediction model to predict the expression levels of molecular markers, and the prediction is accurate.

[0217] Second, the present invention is based on ultrasound image features and is a non-invasive test, which avoids the trauma and complications of tissue biopsy. It also has low equipment requirements and is compatible with conventional ultrasound equipment in current medical institutions at all levels, making it easy to popularize in hospitals below the tertiary level.

[0218] It should be understood that the various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the method and apparatus of the present invention, or some aspects or portions of the method and apparatus of the present invention, can take the form of program codes (i.e., instructions) embedded in a tangible medium, such as a floppy disk, a CD-ROM, a hard disk drive, or any other machine-readable storage medium, wherein when the program is loaded into a machine such as a computer and executed by the machine, the machine becomes a device for practicing the present invention.

[0219] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for illustrating the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which shall fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for predicting PTC molecular markers based on thyroid ultrasound images, characterized in that: include: Step S1: collecting serum samples from a number of patients and detecting the expression levels of relevant molecular markers, wherein the aforesaid patients include patients with papillary thyroid cancer and patients with benign thyroid diseases; Step S2: collecting thyroid ultrasound images of several patients in step S1, and performing feature extraction on the images to obtain several image feature data of each thyroid ultrasound image; Step S3: training a prediction model according to the expression level of the molecular marker obtained in step S1 and the image feature data obtained in step S2; The prediction model is used to predict the expression level of molecular markers based on image feature data; Step S4: collecting a thyroid ultrasound image of the patient to be tested, and performing feature extraction on the image to obtain a number of image feature data of the patient; The image feature data is input into the prediction model, and the prediction model predicts the expression level of the corresponding molecular marker.

2. The method for predicting PTC molecular markers according to claim 1, characterized in that: In step S1, the relevant molecular markers include: one or more of BRAF gene mutation, TERT gene amplification, RET / PTC rearrangement, and TERT promoter mutation.

3. The method for predicting PTC molecular markers according to claim 1, characterized in that: In step S2, the plurality of image feature data include: one or more of nodule size features, echo features, nodule morphology features, edge features, calcification features, and blood flow richness features.

4. The method for predicting PTC molecular markers according to claim 3, characterized in that: The image feature data is obtained by the following method: For nodule size characteristics, the three-dimensional maximum diameter was used for quantitative evaluation; For echo characteristics, quantitative evaluation was performed by comparing the echo of normal thyroid parenchyma in the same layer; The nodule morphology characteristics are obtained by the following methods, including: Step A1: Outline the nodule margin on the thyroid ultrasound image to obtain the nodule area; Step A2: Measure the area of ​​the nodule and record it as A; Step A3: With the geometric center of the nodule area as the center and the maximum axis of the nodule as the radius, draw a circle that conforms to the maximum outer contour of the nodule; Step A4: Calculate the area of ​​the circle and record it as B; Step A5: Calculate the dispersion coefficient c according to the following formula; c = (BA) / B; Step A6: quantitatively evaluate the nodule morphological characteristics using the dispersion coefficient; The edge features are obtained by the following methods: Step B1: Select several representative sections along the maximum axis of the nodule and obtain an image corresponding to each section; Step B2: measuring the thickness of several different sites on the edge of the nodule on each section image; Step B3: Calculate the ratio of the unclear edge area of ​​each site to the maximum diameter of the entire nodule, and obtain several ratios corresponding to each site; Step B4: Calculate the average of all ratios as the edge definition index of the image of the section; Step B5: Calculate the average edge clarity index of all slice images as the comprehensive edge clarity index of the nodule; Step B6: quantitatively evaluating edge features using a comprehensive edge clarity index; For calcification features, the following methods are used to obtain the features: Step C1: adjusting the parameters of the ultrasound image to optimize the display of calcification foci; Step C2: outlining the calcification area on the ultrasound image; Step C3: Calculate the area of ​​the calcified region and record it as D; Step C4: Measure the area of ​​the entire lesion and record it as E; Step C5: Calculate the calcification area ratio f according to the following formula; f = D / E; Step C6: quantitatively evaluating calcification characteristics using the calcification area ratio; A semi-quantitative method was used to quantitatively evaluate the blood flow richness characteristics.

5. The method for predicting PTC molecular markers according to claim 1, characterized in that: In step S3, a prediction model is established by using a support vector machine or a random forest algorithm.

6. A PTC molecular marker prediction device based on thyroid ultrasound images, characterized in that: include: A collection module, used for collecting serum samples from a number of patients and detecting the expression levels of relevant molecular markers, wherein the aforesaid patients include: patients with papillary thyroid cancer and patients with benign thyroid diseases; An acquisition module is used to acquire thyroid ultrasound images of several patients and perform feature extraction on the images to obtain several image feature data of each thyroid ultrasound image; A training module, used to train a prediction model based on the expression levels of molecular markers obtained by the collection module and the image feature data obtained by the acquisition module; The prediction model is used to predict the expression level of molecular markers based on image feature data; The prediction module is used to collect the thyroid ultrasound image of the patient to be tested, perform feature extraction on the image, and obtain a number of image feature data of the patient; The image feature data is input into the prediction model, and the prediction model predicts the expression level of the corresponding molecular marker.

7. The PTC molecular marker prediction device according to claim 6, characterized in that: In the collection module, the relevant molecular markers include: one or more of BRAF gene mutation, TERT gene amplification, RET / PTC rearrangement, and TERT promoter mutation.

8. The PTC molecular marker prediction device according to claim 6, characterized in that: In the acquisition module, the plurality of image feature data include: one or more of nodule size features, echo features, nodule morphology features, edge features, calcification features, and blood flow richness features.

9. The PTC molecular marker prediction device according to claim 8, characterized in that: The image feature data is obtained through the following modules: Nodule size feature assessment module, used to quantify nodule size features using three-dimensional maximum diameter; An echo characteristic evaluation module is used to quantitatively evaluate the echo characteristics by comparing the echo of normal thyroid parenchyma in the same layer; Nodule morphological characteristics assessment module, including: A nodule region acquisition unit is used to outline the nodule margin on the thyroid ultrasound image and acquire the nodule region; Nodule area measurement unit, used to measure the area of ​​the nodule area, recorded as A; A circle drawing unit is used to draw a circle that conforms to the maximum outer contour of the nodule, with the geometric center of the nodule area as the center and the maximum axis of the nodule as the radius; The circle area calculation unit is used to calculate the area of ​​the circle, which is recorded as B; The discrete coefficient calculation unit is used to calculate the discrete coefficient c according to the following formula; c = (BA) / B; A nodule morphological feature evaluation unit, used for quantitatively evaluating nodule morphological features using a dispersion coefficient; Edge feature evaluation module, including: A section image acquisition unit, used for selecting a number of representative sections along the maximum axis of the nodule to obtain an image corresponding to each section; A site thickness measurement unit is used to measure the thickness of several different sites on the edge of the nodule on each section image; A ratio calculation unit is used to calculate the ratio of the unclear edge area of ​​each site to the maximum diameter of the entire nodule, and obtain a number of ratios corresponding to each site; A first averaging unit is used to calculate the average of all ratios as an edge definition index of the image of the section; The second averaging unit is used to calculate the average of the edge clarity indexes of all the slice images as the comprehensive edge clarity index of the nodule; An edge feature evaluation unit, used for quantitatively evaluating edge features using a comprehensive edge clarity index; Calcification feature assessment module, including: A parameter adjustment unit, used to adjust the parameters of the ultrasound image and optimize the display of calcification foci; A calcification area delineation unit, used for delineating a calcification area on an ultrasound image; A calcification area calculation unit is used to calculate the area of ​​the calcification area, which is recorded as D; The lesion area measurement unit is used to measure the area of ​​the entire lesion area, which is recorded as E; The calcification area ratio calculation unit is used to calculate the calcification area ratio f according to the following formula; f = D / E; A calcification feature evaluation unit, used for quantitatively evaluating calcification features using a calcification area ratio; The blood flow richness characteristic evaluation module is used to quantitatively evaluate the blood flow richness characteristics using a semi-quantitative method.

10. The PTC molecular marker prediction device according to claim 6, characterized in that: The training module uses a support vector machine or a random forest algorithm to establish a prediction model.

Images