WO2021237477A1

WO2021237477A1 - Model training method and apparatus, electronic device, and medium

Info

Publication number: WO2021237477A1
Application number: PCT/CN2020/092411
Authority: WO
Inventors: 丁悦; 刘江; 马腾; 傅媛; 陈晓熠; 雷柏英; 陈仲; 肖杨
Original assignee: 广州再生医学与健康广东省实验室; 中山大学孙逸仙纪念医院
Priority date: 2020-05-26
Filing date: 2020-05-26
Publication date: 2021-12-02

Abstract

The embodiments of the present disclosure disclose a model training method and apparatus, an electronic device, and a medium. The model training method comprises: acquiring sample data, the sample data comprising an ultrasonic radio-frequency signal acquired from a human bone and a bone mass label of an acquisition object corresponding to the ultrasonic radio-frequency signal; and training a random forest model on the basis of the sample data. The method, the apparatus, the electronic device, and the medium provided in the embodiments of the present disclosure can improve the sensitivity and the specificity of a bone mass test.

Description

Model training method, device, electronic equipment and medium

Technical field

The present disclosure relates to the field of medical technology, in particular to a model training method, device, electronic equipment and medium.

Background technique

The lack of diagnostic equipment for osteoporosis is one of the important reasons for the low diagnosis and treatment rate of osteoporosis and the high incidence of osteoporotic fractures. The "gold standard" diagnostic equipment for osteoporosis is the dual-energy X-ray absorptiometry (DXA), which cannot be used in a wide range due to high cost, high testing cost, and the presence of certain radiation. It is difficult to popularize primary medical institutions. Quantitative ultrasound (QUS) is one of the most widely used osteoporosis screening equipment. It has the characteristics of low cost, low testing cost, short testing time and no radiation. It is a promising osteoporosis screening device. Check the diagnostic equipment, but because there is no recognized threshold for the diagnosis of osteoporosis, its application is limited. Speed of sound (SOS) is the main parameter of QUS, but there is still controversy about whether QUS based on SOS can be used for osteoporosis screening and diagnosis and fragility fracture risk assessment.

Summary of the invention

In order to solve the problems in the related art, the embodiments of the present disclosure provide a model training method, device, electronic device, and medium.

In the first aspect, an embodiment of the present disclosure provides a model training method.

Specifically, the model training method includes:

Acquiring sample data, where the sample data includes an ultrasound radio frequency signal collected from a human bone and a bone mass tag of the collection object corresponding to the ultrasound radio frequency signal;

Training a random forest model based on the sample data.

With reference to the first aspect, in a first implementation manner of the first aspect of the present disclosure, the training a random forest model based on the sample data includes:

Determining all possible value combinations of the parameters of the random forest model according to a predetermined step size;

By verifying all the possible value combinations, the target value combination of the parameter is determined from the all possible value combinations.

With reference to the first implementation manner of the first aspect, in the second implementation manner of the first aspect of the present disclosure, the verification of all possible value combinations is performed, and the determination is made from all possible value combinations. The target value combinations of the parameters include:

Verifying the all possible value combinations by means of cross-validation, and determining the target value combination of the parameter from the all possible value combinations.

In combination with the first aspect and the first or second implementation manner of the first aspect, in a third implementation manner of the first aspect of the present disclosure, the human bone includes a radius, and the ultrasound radio frequency signal is collected from a non-dominant human body. The distal 1/3 of the lateral radius.

With reference to the first aspect and any one of the first to third implementation manners of the first aspect, in a fourth implementation manner of the first aspect of the present disclosure, the ultrasonic radio frequency signal includes a composite type of transmitting and receiving through two sets of Raw data collected by the transducer.

With reference to the first aspect and any one of the first to fourth implementation manners of the first aspect, in the fifth implementation manner of the first aspect of the present disclosure, the bone mass label includes normal bone mass and bone mass reduction Or osteoporosis.

With reference to the first aspect and any one of the first to fifth implementation manners of the first aspect, in the sixth implementation manner of the first aspect of the present disclosure, the method further includes one or more of the following:

The value of the maximum depth of the model configured with the random forest model is between 45-55;

Configure the value of the minimum sample number of leaf nodes of the random forest model to be between 25-35;

The value of the number of base learners configured for the random forest model is between 100-150.

In the second aspect, an embodiment of the present disclosure provides a model training device.

Specifically, the model training device includes:

The first acquisition module is configured to acquire sample data, the sample data including an ultrasound radio frequency signal collected from a human bone and a bone mass tag of an acquisition object corresponding to the ultrasound radio frequency signal;

The training module is configured to train a random forest model based on the sample data.

With reference to the second aspect, in the first implementation manner of the second aspect of the present disclosure, the training module includes:

The first determining submodule is configured to determine all possible value combinations of the parameters of the random forest model according to a predetermined step size;

The second determining sub-module is configured to determine the target value combination of the parameter from all the possible value combinations by verifying the all possible value combinations.

With reference to the first implementation manner of the second aspect, in the second implementation manner of the second aspect of the present disclosure, the second determination submodule is configured to verify all possible value combinations by means of cross-validation, The target value combination of the parameter is determined from all the possible value combinations.

In combination with the second aspect and the first or second implementation manner of the second aspect, in a third implementation manner of the second aspect of the present disclosure, the human bone includes a radius, and the ultrasound radio frequency signal is collected from a non-dominant human body. The distal 1/3 of the lateral radius.

With reference to the second aspect and any one of the first to third implementation manners of the second aspect, in a fourth implementation manner of the second aspect of the present disclosure, the ultrasonic radio frequency signal includes a composite type of transmitting and receiving through two sets of Raw data collected by the transducer.

With reference to the second aspect and any one of the first to fourth implementation manners of the second aspect, in the fifth implementation manner of the second aspect of the present disclosure, the bone mass label includes normal bone mass and bone mass reduction Or osteoporosis.

With reference to the second aspect and any one of the first to fifth implementation manners of the second aspect, in the sixth implementation manner of the second aspect of the present disclosure, the device further includes a configuration module configured to Do one or more of the following:

In a third aspect, an embodiment of the present disclosure provides an electronic device including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are processed by the The device executes to implement the method described in the first aspect and any one of the first to the sixth implementation manners of the first aspect.

In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium with computer instructions stored thereon. When the computer instructions are executed by a processor, the first to sixth aspects of the first aspect and the first aspect are implemented. The method described in any one of the implementation modes.

In a fifth aspect, an embodiment of the present disclosure provides an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are The processor executes to achieve:

Acquire ultrasonic radio frequency signals collected from human bones;

The ultrasound radio frequency signal is processed based on the random forest model to obtain bone mass information.

In a sixth aspect, the embodiments of the present disclosure provide a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are implemented when executed by a processor:

Acquire ultrasonic radio frequency signals collected from human bones;

According to the technical solution provided by the embodiments of the present disclosure, by acquiring sample data, the sample data includes the ultrasound radio frequency signal collected from the human bone and the bone mass tag of the collection object corresponding to the ultrasound radio frequency signal; training based on the sample data The random forest model can train a model for obtaining bone mass information, which has good sensitivity and specificity for obtaining bone mass information.

It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure.

Description of the drawings

With reference to the accompanying drawings, through the following detailed description of the non-limiting implementation manners, other features, purposes, and advantages of the present disclosure will become more apparent. In the attached picture:

Fig. 1 shows a flowchart of a model training method according to an embodiment of the present disclosure;

Fig. 2 shows a schematic diagram of acquiring an ultrasonic radio frequency signal according to an embodiment of the present disclosure;

Fig. 3 shows a flowchart of training a random forest model based on the sample data according to an embodiment of the present disclosure;

FIG. 4 shows a receiver operating characteristic curve diagram of a random forest model trained according to a model training method of an embodiment of the present disclosure applied to obtain bone mass data;

Fig. 5 shows a block diagram of a model training device according to an embodiment of the present disclosure;

Figure 6 shows a block diagram of a training module according to an embodiment of the present disclosure;

FIG. 7 shows a block diagram of an electronic device according to an embodiment of the present disclosure; and

Fig. 8 shows a schematic structural diagram of a computer system suitable for implementing a model training method according to an embodiment of the present disclosure.

Detailed ways

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts that are not related to the description of the exemplary embodiments are omitted in the drawings.

In the present disclosure, it should be understood that terms such as "including" or "having" are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof disclosed in this specification, and are not intended to exclude one The possibility of existence or addition of multiple other features, numbers, steps, behaviors, components, parts or combinations thereof.

In addition, it should be noted that the embodiments in the present disclosure and the features in the embodiments can be combined with each other if there is no conflict. Hereinafter, the present disclosure will be described in detail with reference to the drawings and in conjunction with the embodiments.

The inventor found through a large number of literature searches that the current SOS-based QUS diagnosis of osteoporosis has low efficiency, with an average sensitivity of 52.22% and an average specificity of 58.27%. The area under the receiver operating curve is the highest. operating characteristic curve, AUC) is 0.696, which is not yet of diagnostic value and can only be used for the initial screening of female osteoporosis.

Ultrasound radio frequency signal (RF signal) is a standard form of unprocessed raw data in ultrasound imaging. The method of extracting and processing RF signals has been widely used in the characterization of medical ultrasound imaging tissues, such as breast, prostate and liver ultrasound, which can effectively improve the diagnostic capabilities of ultrasound. There is no precedent for the application of RF signals in the field of orthopedics, and the application value of RF signals in the field of orthopedics needs to be further explored.

Random forest (Random forest) is a supervised machine learning algorithm. The inventor proposes that random forest is used to train multi-feature sample data of ultrasound radio frequency signals of human bone quantitative ultrasound. Porosity can achieve better prediction accuracy.

The methods, devices, electronic equipment and media provided by the embodiments of the present disclosure explore the method of quantitative ultrasound of human bones, and for the first time use machine learning random forest algorithm to train the ultrasound radio frequency signals of quantitative ultrasound of human bones, establish a bone mass detection model, and improve bones. The diagnostic efficiency of quantitative ultrasound provides a strong basis for the application of bone quantitative ultrasound in the diagnosis of osteoporosis in primary medical institutions.

Fig. 1 shows a flowchart of a model training method according to an embodiment of the present disclosure. As shown in Figure 1, the model training method includes the following steps S110 and S120:

In step S110, sample data is obtained, and the sample data includes the ultrasound radio frequency signal collected from the human bone and the bone mass tag of the collection object corresponding to the ultrasound radio frequency signal;

In step S120, a random forest model is trained based on the sample data.

According to an embodiment of the present disclosure, the ultrasonic radio frequency signal includes raw data collected by two sets of transmitting and receiving composite transducers.

The pulse wave emitted by the ultrasonic transducer propagates in the human tissue, and then a reflected wave is generated. The probe converts the reflected wave into an electrical signal for output, and this signal is an ultrasonic radio frequency signal. The ultrasonic radio frequency signal contains all the amplitude, frequency and phase information, that is, a large amount of information about the interaction between the sound field and the tissue, and the characteristics of the microstructure. At present, the application of ultrasound radio frequency signals is usually to convert them into ultrasound imaging, such as A-mode ultrasound and B-mode ultrasound. Among them, the most widely used B-mode ultrasound only uses the amplitude information in the original data, and the gray value imaging is obtained after a series of filtering, mediation and compression processing. Therefore, in the current ultrasound technology, a large amount of information in the raw data of the ultrasound radio frequency signal is ignored or lost in the imaging stage.

According to the technical solution provided by the embodiments of the present disclosure, the raw data of the ultrasonic radio frequency signal is collected by two sets of transmitting and receiving composite transducers, which can obtain a richer bone condition, which is beneficial to improve the sensitivity and specificity of detection.

According to an embodiment of the present disclosure, the human bones include radius, tibia or root bone, etc. Preferably, the radius can be selected. The ultrasound radio frequency signal can be collected at the distal 1/3 of the radius on the non-dominant side of the human body, as shown in FIG. 2.

According to the technical solutions provided by the embodiments of the present disclosure, the radius is an easily exposed site, and is more sensitive to bone metabolism, and is also one of the important osteoporotic fracture sites. It is a basic condition for wide-scale promotion as a diagnostic site. The 1/3 of the distal radius is used to assess the risk of fragility fractures, especially hip fractures, with better results. Since the non-dominant side is less used, it is less affected by other factors, which is beneficial to reflect the actual situation of the human bones as a whole. Of course, in the case of fractures on the non-dominant side, the radius on the dominant side can also be used for measurement.

According to an embodiment of the present disclosure, the bone mass label includes normal bone mass, reduced bone mass, or osteoporosis. The label of reduced bone mass is helpful for reminding the risk of bone mass. It is also possible to combine normal bone mass and bone mass reduction into non-osteoporosis and treat them as one type. The combined label is more simplified and can directly distinguish between osteoporosis and non-osteoporosis.

According to the technical solution provided by the embodiments of the present disclosure, through the bone mass label including normal bone mass, reduced bone mass, or osteoporosis, it is possible to train a model for obtaining bone mass information, and the model can be used to obtain bone mass information. The sensitivity and specificity.

Fig. 3 shows a flowchart of training a random forest model based on the sample data according to an embodiment of the present disclosure. As shown in Figure 3, the model training method includes the following steps S121 and S122:

In step S121, all possible value combinations of the parameters of the random forest model are determined according to a predetermined step size;

In step S122, the target value combination of the parameter is determined from all the possible value combinations by verifying the all possible value combinations.

According to the technical solution provided by the embodiment of the present disclosure, all possible value combinations of the parameters of the random forest model are determined according to a predetermined step; In the combination, the target value combination of the parameters is determined, and the parameters are adjusted in the manner of grid search as described above, which can effectively improve the classification accuracy and avoid the problem of overfitting.

According to an embodiment of the present disclosure, the step of verifying all possible value combinations and determining the target value combination of the parameter from all possible value combinations includes verifying all possible value combinations by means of cross-validation. Determine the target value combination of the parameter from all possible value combinations.

For example, three-fold cross-validation can be used to divide the sample data into three different subsets, two of which are used as the training set and the other as the test set each time, and the model is trained and tested for three times.

According to the technical solution provided by the embodiments of the present disclosure, the all possible value combinations are verified by means of cross-validation, and the target value combination of the parameter is determined from the all possible value combinations, which can effectively improve the classification accuracy And avoid the problem of overfitting.

According to the embodiments of the present disclosure, one or more of the following is further included:

According to the technical solutions provided by the embodiments of the present disclosure, after training, it is found that by configuring the above parameters, the best classification effect can be achieved.

The following describes with specific examples.

The research objects of the embodiments of the present disclosure are derived from 32 subjects who planned to undergo DXA examination at Sun Yat-sen Memorial Hospital of Sun Yat-sen University in Guangzhou, Guangdong Province between January 2020 and March 25 to April 25, 2020. Among them, 26 were females, all were postmenopausal females, and 6 were males, with an average age of 61.50±9.09 years old. The subjects did not have the following conditions: (1) Malnutrition, using the Mini Nutritional Assessment-short form (MNA-SF), and the score ≤11 is considered malnutrition; (2) Severe cardiopulmonary, liver and kidney failure And malignant tumors; (3) Combined diseases that affect bone density around joints, such as rheumatoid arthritis, ankylosing spondylitis, etc. Subjects who have a history of fractures, suffer from related diseases that affect bone metabolism, take related drugs that affect bone metabolism, or are receiving anti-osteoporosis treatments do not need to be specifically excluded. All subjects completed radial QUS and DXA measurements on the same day, and no intervention was given in the interval. All subjects signed an informed consent form. Patients can withdraw from the study if they actively request to withdraw from the study or if the patient has serious adverse reactions. This study was approved by the Ethics Committee of Sun Yat-sen Memorial Hospital of Sun Yat-sen University.

The diagnosis of osteoporosis is based on the diagnostic criteria recommended by the World Health Organization. A person who meets one of the following three criteria can be diagnosed as osteoporosis: (1) Fragile fracture of the hip or vertebral body; (2) Axial bone measured by DXA Bone mineral density (lumbar spine 1-4, femoral neck or total hip) or distal radius 1/3 bone mineral density T value ≤ -2.5; (3) Bone mineral density measurement accords with low bone mass (-2.5<T value<-1.0) + Fragile fractures of the proximal humerus, pelvis or distal forearm. The classification standard for determining bone mineral density by DXA: T value ≥-1.0 is normal bone mass; -2.5<T value<-1.0 is bone mass reduction; T value ≤-2.5 is osteoporosis; T value ≤-2.5+ fragility fracture is Severe osteoporosis. According to the above criteria, this part of the study divided the subjects into three groups: normal bone mass group (12 cases), bone mass reduction group (10 cases) and osteoporosis group (10 cases).

Before the examination, it is necessary to clean and wipe the radial side of the forearm of the subject to be measured in order to apply the ultrasound couplant later. Air is a poor conductor of ultrasound, and it is difficult for the ultrasound probe to completely fit the human body surface. Therefore, a liquid conductive medium, that is, couplant, is required between the ultrasound probe and the body surface.

In this part of the study, DXA was tested using Lunar model DXA equipment produced by GE in the United States. The testing sites included lumbar spine 1-4, femoral neck and total hip. The measurement was carried out by a fixed professional and technical personnel daily for quality control inspection and according to routine The operation is blinded. Before the measurement, no information about the radius QUS is provided except the basic information of the subject, and the coefficient of variation of the precision of the DXA equipment is less than 1%.

The source data of the radius QUS-RF signal is collected by the OSTEOKJ7000+ ultrasonic bone densitometer produced by Nanjing Kejin Industrial Co., Ltd. The ultrasonic working frequency is 1MHz, the error is ±15%, the ultrasonic velocity error is ≤±0.4%, and the accuracy is ≤0.15%. The ultrasonic probe is a linear array, including two sets of transmitter-receiver composite transducers, which can output 4 sets of data. RF signal source data is saved in .txt format. Each .txt file records valid waveform data, including four channels of data, 1024 data per channel, and the range of each data is 0-255, with data1, data2, and data3 , Data4 as a separator. Each subject will collect at least 200 time points of source data. Part of the RF signal source data of a subject at a certain time point is shown below:

Table 1 Part of the original output data of the RF signal

The measurement site of the radius QUS is the non-dominant forearm. If there is a history of non-dominant forearm fracture, the other side is selected. The subject is placed on a smooth and flat surface such as a desk to fully expose the radius and the distal 1/3 of the radius. The position is the midpoint of the line from the end of the middle finger to the olecranon when the forearm is straightened. The placement of the ultrasound probe during measurement is shown in Figure 2. Apply a proper amount of ultrasound coupling agent to the skin on the radial side. Keep the ultrasound probe parallel to the longitudinal axis of the radius and perpendicular to the bone surface. Repeat the measurement 3 times. For each measurement, the measurement site needs to be re-determined with a soft ruler, and the SOS value results and corresponding T are recorded and stored. Value, extract and store the raw data of the measured ultrasonic radio frequency signal. All measurements are operated by 2 fixed professional technicians, and the QUS equipment is calibrated with the standard module provided by the manufacturer before the daily measurement. The QUS measurement of radius is blinded, and the DXA measurement result of the subject is not provided before the measurement.

According to the embodiment of the present disclosure, the stored SOS value result and the corresponding T value are used as a reference method, and the standard recommended by the World Health Organization is used as the basis for dividing the bone mass. This method is described below as "radius QUS-SOS T value". The raw data of the measured ultrasonic radio frequency signal is extracted and stored for use in the method of the embodiment of the present disclosure to train the bone mass detection model, and the method of applying the model for diagnosis is described as "radius QUS-RF signal" in the following.

According to the embodiment of the present disclosure, the source data of the first 200 time points will be extracted from each of the top 25 subjects, a total of 5000 data will be used as the training set, and the 1400 data of the last 7 subjects will be used as the test set. Delete the first negative value in all RF signal source data, and subtract the reference value from all remaining data for correction. The number of sample units in the training set is N (N=5000), and M feature variables (M=4092). Each data value collected is considered a feature variable, and training is performed using the method as shown in Figure 3.

The optimal parameters after grid search and triple cross-validation are: the maximum depth of the model (max_depth) is between 45-55, more preferably, it can be set to 50; the minimum number of samples for leaf nodes (min_samples_leaf) is selected The value is between 25-35, more preferably, it can be set to 30; the value of the number of base learners (n_estimators) is between 100-150, and more preferably, it can be set to 100; the other parameters are Defaults:

RandomForestClassifier(bootstrap=True,class_weight=None,criterion='gini',max_depth=50,max_feature='auto',max_leaf_nodes=None,min_impurity_decrease=0.0,min_samples_split=2,min_weight_fraction_leaf=0.0,n_n_jobs=100,n_n_jobs= oob_score=False, random_state=None, verbose=0, warm_start=False)

The basic information of the subjects is shown in Table 2.

Table 2 Basic information of subjects

Note: a-ANOVA (one-way analysis of variance); b-Fisher exact test; c-Kruskal-Wallis test

According to the embodiments of the present disclosure, when the radius SOS-T value is used alone to diagnose osteoporosis, according to the diagnostic threshold recommended by the World Health Organization, a T value ≤ -2.5 is diagnosed as osteoporosis, and -2.5<T value<-1.0 is The bone mass is decreased, and the T value ≥-1.0 is the normal bone mass. According to the data in Table 3 and Table 4 below, the sensitivity of the radius QUS-SOS T value to diagnose osteoporosis is 60.00% (95% CI: 27.37%-86.31%), and the specificity is 81.82% (95% CI: 58.99%- 94.01%), positive predictive value 60.00% (95% CI: 27.37%-86.31%), negative predictive value 81.82% (95% CI: 58.99%-94.01%), positive likelihood ratio 3.3 (95% CI: 1.19- 9.16), the negative likelihood ratio is 0.49 (95% CI: 0.22-1.07), and the Youden index is 0.418. However, there was no statistically significant difference between the QUS-SOS T value of radius in the diagnosis of bone mass or osteoporosis and the results of DXA measurement, and the consistency was poor (P>0.05).

Table 3 Comparison of radius QUS-SOS T value diagnosis of bone mass and DXA

Table 4 Comparison of QUS-SOS T value diagnosis of radius and DXA for osteoporosis

According to the embodiment of the present disclosure, the diagnostic model performs classification diagnosis on the radius QUS-RF signal in the test set and compares it with the DXA result. The detailed results are shown in Table 5 and Table 6. The QUS-RF signal of radius in the diagnosis of bone mass or osteoporosis is correlated with the DXA measurement results (P<0.05), and the diagnosis of bone mass or osteoporosis has good repeatability (Kappa values, respectively) 0.60, 0.72, Cohen's kappa).

Table 5 Comparison of radius QUS-RF signal in diagnosis of bone mass and DXA

Table 6 Comparison of radius QUS-RF signal in diagnosis of osteoporosis and DXA

Draw the ROC curve of the QUS-RF signal osteoporosis diagnosis model of the radius, as shown in Figure 4, the area under the curve AUC is 0.91 (95% CI: 0.89-0.94), which is much larger than the retrieved QUS-SOS T value diagnosis method The average value is 0.696. In addition, the model used to detect bone loss and normal bone mass AUC were 0.66 and 0.74, respectively.

After statistical analysis, the sensitivity of the method in the embodiments of the present disclosure is 62.50% (95% CI: 55.36%-69.15%), the specificity is 99.42% (95% CI: 98.75%-99.74%), and the positive predictive value is 94.70% ( 95% CI: 88.98%-97.66%), negative predictive value 94.09% (95% CI: 92.61%-95.29%), positive likelihood ratio 107.14 (95% CI: 50.79-226.00), negative likelihood ratio 0.3772 (95 %CI: 0.3154-0.4511), Youden Index 0.6192. The comparison of the diagnostic efficacy of radius QUS-RF signal and radius QUS-SOS T value in diagnosing osteoporosis can be seen in Table 7.

Table 7 Comparison of diagnostic efficacy of radius QUS-RF signal and radius QUS-SOS T value in diagnosing osteoporosis

Among them, the sensitivity description describes the proportion of all positive cases identified in all positive cases, which can be used to measure the missed diagnosis rate; the specificity description describes the proportion of negative cases identified in all negative cases, which can be used to measure the misdiagnosis rate; the positive predictive value description The proportion of positive samples that are correctly predicted to all positive samples; negative predictive value describes the proportion of negative samples that are correctly predicted to all negative samples; the positive likelihood ratio describes the probability of correctly judging positive as the multiple of the probability of incorrectly judging positive; Negative likelihood ratio means that the probability of false negative judgment is a multiple of the probability of correct negative judgment; Youden index = sensitivity + specificity -1, which means the total ability of the screening method to find real patients and non-patients.

It can be seen from Table 6 that the diagnostic efficiency of the radius QUS-RF signal and the random forest model for diagnosing osteoporosis in the embodiments of the present disclosure is better than the radius QUS-SOS T value method, especially the specificity reaches 99.42%, indicating a misdiagnosis The rate is extremely low.

Fig. 5 shows a block diagram of a model training device according to an embodiment of the present disclosure. Wherein, the device can be implemented as part or all of the electronic device through software, hardware or a combination of the two.

As shown in FIG. 5, the model training device 500 includes a first acquisition module 510 and a training module 520.

The first acquisition module 510 is configured to acquire sample data, the sample data including an ultrasound radio frequency signal collected from a human bone and a bone mass tag of an acquisition object corresponding to the ultrasound radio frequency signal;

The training module 520 is configured to train a random forest model based on the sample data.

According to the technical solution provided by the embodiments of the present disclosure, the first acquisition module is configured to acquire sample data, the sample data including the ultrasound radio frequency signal collected from the human bone and the bone mass of the collection object corresponding to the ultrasound radio frequency signal Label; training module, configured to train a random forest model based on the sample data, capable of training a model for acquiring bone mass information, which has good sensitivity and specificity for acquiring bone mass information.

FIG. 6 shows a block diagram of a training module 600 according to an embodiment of the present disclosure.

As shown in FIG. 6, the training module 600 includes a first determining sub-module 610 and a second determining sub-module 620.

The first determining submodule 610 is configured to determine all possible value combinations of the parameters of the random forest model according to a predetermined step size;

The second determining submodule 620 is configured to determine the target value combination of the parameter from all the possible value combinations by verifying the all possible value combinations.

According to the technical solution provided by the embodiments of the present disclosure, the first determining sub-module is configured to determine all possible value combinations of the parameters of the random forest model according to a predetermined step size; the second determining sub-module is configured to pass Verifying all possible value combinations and determining the target value combination of the parameter from all possible value combinations can effectively improve the classification accuracy and avoid the problem of overfitting.

According to an embodiment of the present disclosure, the second determining submodule is configured to verify all possible value combinations by means of cross-validation, and determine the target value combination of the parameter from the all possible value combinations .

According to the technical solution provided by the embodiments of the present disclosure, the second determining sub-module is configured to verify all possible value combinations in a cross-validation manner, and determine the target of the parameter from the all possible value combinations The combination of values can effectively improve the classification accuracy and avoid the problem of overfitting.

According to an embodiment of the present disclosure, the human bone includes a radius, and the ultrasound radio frequency signal is collected at the distal 1/3 of the radius on the non-dominant side of the human body.

According to the technical solutions provided by the embodiments of the present disclosure, the radius is an easily exposed site, and is more sensitive to bone metabolism, and is also one of the important osteoporotic fracture sites. It is a basic condition for wide-scale promotion as a diagnostic site. The 1/3 of the distal radius is used to assess the risk of fragility fractures, especially hip fractures, with better results.

According to an embodiment of the present disclosure, the bone mass label includes normal bone mass, reduced bone mass, or osteoporosis.

According to an embodiment of the present disclosure, the device further includes a configuration module configured to perform one or more of the following:

The present disclosure also discloses an electronic device, and FIG. 7 shows a structural block diagram of the electronic device according to an embodiment of the present disclosure.

As shown in FIG. 7, the electronic device 700 includes a memory 701 and a processor 702, wherein the memory 701 is used to store a program that supports the electronic device to execute the model training method or code generation method in any of the above embodiments, so The processor 702 is configured to execute a program stored in the memory 701.

According to an embodiment of the present disclosure, the memory 701 is configured to store one or more computer instructions, where the one or more computer instructions are executed by the processor 702 to implement the following steps:

Training a random forest model based on the sample data.

According to an embodiment of the present disclosure, training a random forest model based on the sample data includes:

According to an embodiment of the present disclosure, the step of verifying all possible value combinations and determining the target value combination of the parameter from the all possible value combinations includes:

According to an embodiment of the present disclosure, the human bone includes a radius, and the ultrasonic radio frequency signal is collected at the distal 1/3 of the radius on the non-dominant side of the human body; the ultrasonic radio frequency signal is collected by two sets of transmitting and receiving composite transducers And/or, the bone mass label includes normal bone mass, reduced bone mass, or osteoporosis.

According to the embodiment of the present disclosure, the processor 702 is further configured to implement one or more of the following:

Acquire ultrasonic radio frequency signals collected from human bones;

Fig. 8 shows a schematic structural diagram of a computer system suitable for implementing the model training method according to an embodiment of the present disclosure.

As shown in FIG. 8, the computer system 800 includes a processing unit 801, which can execute the above-mentioned implementation according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 808 to a random access memory (RAM) 803 Various treatments in the example. In the RAM 803, various programs and data required for the operation of the system 800 are also stored. The processing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input/output (I/O) interface 805 is also connected to the bus 804.

The following components are connected to the I/O interface 805: an input part 806 including a keyboard, a mouse, etc.; an output part 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and speakers, etc.; a storage part 808 including a hard disk, etc. ; And a communication section 809 including a network interface card such as a LAN card, a modem, and the like. The communication section 809 performs communication processing via a network such as the Internet. The driver 810 is also connected to the I/O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed, so that the computer program read therefrom is installed into the storage section 808 as needed. Wherein, the processing unit 801 may be implemented as a processing unit such as CPU, GPU, TPU, FPGA, and NPU.

In particular, according to an embodiment of the present disclosure, the method described above may be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a readable medium thereof, and the computer program includes program code for executing the above-mentioned method. In such an embodiment, the computer program may be downloaded and installed from the network through the communication part 809, and/or installed from the removable medium 811.

The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions, and operations of the system, method, and computer program product according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of the code, and the module, program segment, or part of the code contains one or more functions for realizing the specified logical function. Executable instructions. It should also be noted that, in some alternative implementations, the functions marked in the block may also occur in a different order from the order marked in the drawings. For example, two blocks shown one after the other can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations Or it can be realized by a combination of dedicated hardware and computer instructions.

The units or modules involved in the embodiments described in the present disclosure can be implemented in the form of software, and can also be implemented in the form of programmable hardware. The described units or modules may also be provided in the processor, and the names of these units or modules do not constitute a limitation on the units or modules themselves in certain circumstances.

As another aspect, the present disclosure also provides a computer-readable storage medium. The computer-readable storage medium may be the computer-readable storage medium contained in the electronic device or computer system in the above-mentioned embodiment; or it may exist alone. , A computer-readable storage medium that is not installed in the device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the methods described in the present disclosure.

The above description is only a preferred embodiment of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover the technical solutions based on the above technical features without departing from the inventive concept. Or other technical solutions formed by any combination of its equivalent features. For example, the above-mentioned features and the technical features disclosed in the present disclosure (but not limited to) having similar functions are replaced with each other to form a technical solution.

Claims

A model training method is characterized in that it includes:

Acquiring sample data, where the sample data includes an ultrasound radio frequency signal collected from a human bone and a bone mass tag of the collection object corresponding to the ultrasound radio frequency signal;

Training a random forest model based on the sample data.
The method according to claim 1, wherein the training a random forest model based on the sample data comprises:

Determining all possible value combinations of the parameters of the random forest model according to a predetermined step size;

By verifying all the possible value combinations, the target value combination of the parameter is determined from the all possible value combinations.
The method according to claim 2, wherein the determining the target value combination of the parameter from the all possible value combinations by verifying the all possible value combinations comprises:

Verifying the all possible value combinations by means of cross-validation, and determining the target value combination of the parameter from the all possible value combinations.
The method according to any one of claims 1 to 3, characterized in that:

The human bone includes a radius, and the ultrasound radio frequency signal is collected at the distal 1/3 of the radius on the non-dominant side of the human body;

The ultrasonic radio frequency signal includes the original data collected by the two sets of transmitting and receiving composite transducers; and/or

The bone mass label includes normal bone mass, reduced bone mass, or osteoporosis.
The method according to any one of claims 1 to 3, further comprising one or more of the following:

The value of the maximum depth of the model configured with the random forest model is between 45-55;

Configure the value of the minimum sample number of leaf nodes of the random forest model to be between 25-35;

The value of the number of base learners configured for the random forest model is between 100-150.
A model training device is characterized in that it comprises:

The first acquisition module is configured to acquire sample data, the sample data including an ultrasound radio frequency signal collected from a human bone and a bone mass tag of an acquisition object corresponding to the ultrasound radio frequency signal;

The training module is configured to train a random forest model based on the sample data.
An electronic device, characterized by comprising a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to realize:

Acquiring sample data, where the sample data includes an ultrasound radio frequency signal collected from a human bone and a bone mass tag of the collection object corresponding to the ultrasound radio frequency signal;

Training a random forest model based on the sample data.
A readable storage medium having computer instructions stored thereon, characterized in that the computer instructions are implemented when executed by a processor:

Acquiring sample data, where the sample data includes an ultrasound radio frequency signal collected from a human bone and a bone mass tag of the collection object corresponding to the ultrasound radio frequency signal;

Training a random forest model based on the sample data.
An electronic device, characterized by comprising a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to realize:

Acquire ultrasonic radio frequency signals collected from human bones;

The ultrasound radio frequency signal is processed based on the random forest model to obtain bone mass information.
A readable storage medium having computer instructions stored thereon, characterized in that the computer instructions are implemented when executed by a processor:

Acquire ultrasonic radio frequency signals collected from human bones;

The ultrasound radio frequency signal is processed based on the random forest model to obtain bone mass information.