Automatic evaluation method and system for patent aperture of foramen ovale based on TCD foaming test signal characteristics

Abstract

The invention relates to the technical field of medical detection, and discloses an automatic evaluation method and system for the patent foramen ovale aperture based on TCD foaming test signal characteristics, and the method comprises the following steps: S1, obtaining a clinical TCD foaming test audio record; s2, identifying the number of independent microemboli, the number and duration of classical rain curtains and the number and duration of front ends of the rain curtains in the audio record; and S3, according to the number of the independent microemboli, the number and duration of the classic rain curtains, and the number and duration of the front ends of the rain curtains, determining the size of the shunting aperture. The method has the beneficial effects that four classification standards of the TCD foaming test and the PFO pore size are established for the first time by combining clinical data of transesophageal ultrasound. According to the scheme, the size of the shunting aperture of a patient can be automatically evaluated by using clinical TCD foaming test audio records, compared with the prior art, the scheme does not cause unsuitable pain of the patient, meanwhile, invasive operation is not needed, the cost is low, and the method is more suitable for market popularization.

Description

Technical Field

[0001] The present invention relates to the field of medical detection technology, and in particular to an automated evaluation method and system for patent foramen ovale diameter based on TCD foaming test signal characteristics. Background Art

[0002] Right-to-Left Shunt (RLS) is an abnormal channel connecting the arteriovenous system, including patent foramen ovale (PFO), atrial septal defect, patent ductus arteriosus, pulmonary arteriovenous fistula, etc., among which PFO is the most common, accounting for about 95% of all RLS. In recent years, more and more evidence has shown that RLS is associated with ischemic stroke. Therefore, assessing the risk of stroke caused by RLS, especially PFO, is of great significance for the prevention of ischemic stroke.

[0003] Risk factors for ischemic stroke caused by PFO include age, shunt aperture size, presence of atrial septal aneurysm and other atrial abnormalities (such as inferior vena cava valve, Chiari network, etc.). Among them, the larger the shunt aperture, the higher the risk of ischemic stroke, and the shunt aperture may be an independent risk factor for recurrence of ischemic stroke; in addition, the larger the PFO aperture, the larger the diameter of the embolus allowed to pass, the larger the diameter of the cerebral artery that can be blocked (the higher the risk of causing large artery occlusion), and the more severe the ischemic stroke caused. Therefore, judging the size of the shunt aperture is of great significance for predicting the risk and severity of PFO-related stroke.

[0004] In the prior art, the examinations for evaluating the size of the shunt aperture include transesophageal echocardiography (TEE) and intracardiac echocardiography (ICE). The TEE probe is located in the esophagus (behind the heart) and can clearly observe the anatomical structure of the atrial septum. It is the "gold standard" for diagnosing PFO and can measure the size of the PFO aperture. The maximum PFO opening diameter measured with the Valsalva maneuver is close to its actual size. However, TEE requires the probe to be inserted into the patient's esophagus. It is a semi-invasive examination that will cause discomfort and pain to the patient, poor tolerance, and may cause complications such as esophageal mucosal damage or even perforation and bleeding. There are potential safety risks and clinical applications are limited. ICE inserts a high-frequency ultrasound probe into the heart through a venous route (such as the femoral vein) for real-time imaging. Compared with TEE, it has a higher image resolution and can more accurately measure the size and length of the PFO aperture and observe complex anatomical structures. However, it requires invasive surgery through a vascular intervention route and is expensive, and is not suitable for routine clinical screening. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art that transesophageal echocardiography can cause discomfort and pain to patients when evaluating the size of the shunt aperture, and that intracardiac echocardiography requires invasive operations and is expensive, the present invention is based on the clinical data of the TCD bubble test, explores the potential of the TCD bubble test to predict the size of the PFO aperture, and establishes a four-level classification standard for PFO. A method, system and device for automatically evaluating the size of the patent foramen ovale based on the signal characteristics of the TCD bubble test are proposed, which realizes the accurate classification and statistics of the complex characteristics of the TCD bubble test and achieves the purpose of automatically evaluating the PFO aperture. The specific scheme is as follows:

[0006] An automated method for evaluating patent foramen ovale aperture based on TCD blister test signal characteristics comprises the following steps: S1, obtaining a clinical TCD blister test audio record; S2, identifying the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain front ends in the TCD blister test audio record; S3, determining the size of the shunt aperture according to the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain front ends; the independent microemboli are audio signals with typical pulse-type audio characteristics, a duration of less than 300 ms, and the number of occurrences of which can be accurately counted; the classic rain curtain is an audio signal that is connected into one piece and has a duration of ≥0.5s, and the time interval between two classic rain curtains is ≥10ms; the rain curtain front end is a rain curtain-like signal including multiple microemboli with a duration of <0.5s overlapping and fused with each other, and independent microemboli with a spacing of ≤0.3s and a density of ≥20 / second.

[0007] Furthermore, in S1, a microbubble ultrasound contrast agent was injected from the patient's cubital vein through a TCD contrast enhancement test, and the intracranial large arteries were simultaneously detected for the presence of microembolic signals to obtain a clinical TCD bubble test audio recording.

[0008] Furthermore, in S2, the TCD bubble test signals were divided into independent microemboli, classic rain curtains, and rain curtain front ends according to the density and duration of microemboli in the audio recording of the TCD bubble test.

[0009] Furthermore, in S2, the TCD blister test audio recording is identified through the attention model, and the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain fronts are obtained.

[0010] Furthermore, the attention model is a physiological time series segmentation model that integrates an attention mechanism, and includes a basic architecture of an encoder, a decoder, and an attention gate; the encoder and the decoder are used to capture local and global features in the TCD bubble test audio recording, and the attention gate is connected between each upsampling block of the decoder and the feature map of the corresponding scale of the encoder to enhance the decoder's attention to important features.

[0011] Furthermore, in S2, the audio recording of the TCD foaming test is framed, the volume and zero-crossing rate of each frame are calculated, and the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain front ends are obtained based on the adaptive threshold.

[0012] Further, in S2, the frame energy E is an indicator describing the strength of the audio signal, and the calculation formula is shown in (4), where x[n] represents the amplitude of the sampling point in the current frame, and N is the frame length;

[0013]

[0014] ZCR refers to the number of times each frame signal crosses the zero point, which can reflect the oscillation frequency and energy change of the signal. The calculation formula is shown in (5); where x[n] represents the sampling point value in the current frame, N is the length of the current frame, and II(g) is the indicator function, which takes the value of 1 when the condition is true, otherwise it takes the value of 0;

[0015]

[0016] Volume threshold T E The setting is based on the global frame energy of the signal, and the dynamic range of the volume characteristics of the background area and the segmented area is distinguished to ensure that the threshold can adapt to the amplitude change of the signal; the volume threshold is defined as shown in (6), V min is the 5th percentile of the volume series, V max is the 95th percentile of the volume sequence, α is the adjustment coefficient, which controls the threshold value; the zero-crossing rate threshold T ZCR Dynamic adjustment is performed based on the ZCR mean of the local frame. The calculation method is shown in (7).

[0017] T E =V min +α(V max -V min ) (6)

[0018]

[0019] N is the current frame number, z i is the frame zero-crossing rate, and β is the adjustment coefficient, which controls the sensitivity of the threshold to the signal.

[0020] Further, in S3, when the number of independent microemboli is 0, the number of classic rain curtains is 0, and the number of rain curtain front ends is 0, the shunt aperture is 0 mm; when the number of independent microemboli is greater than 0, and the number of classic rain curtains is 0, and the number of rain curtain front ends is 0, the shunt aperture is greater than 0 mm and less than 0.5 mm; when the number of rain curtain front ends is greater than 0, and the duration of the rain curtain front ends is less than 5 s, and the number of classic rain curtains is 0, the shunt aperture is greater than or equal to 0.5 mm and less than or equal to 1.5 mm; when the number of rain curtain front ends is greater than 0, and the duration of the rain curtain front ends is greater than 5 s, the shunt aperture is greater than 1.5 mm; or when the number of classic rain curtains is greater than 0, the shunt aperture is greater than 1.5 mm.

[0021] A system is used to implement the above method.

[0022] According to any of the above embodiments, the present invention has at least the following beneficial effects:

[0023] The present invention combines the clinical data of transesophageal ultrasound and establishes a four-classification standard for TCD foaming test and PFO aperture size for the first time. Using the audio recording of the clinical TCD foaming test, the patient's shunt aperture size can be automatically evaluated. Compared with the existing technology, this solution will not cause discomfort and pain to the patient, and does not require invasive operations, and has a lower cost and is more suitable for market promotion.

[0024] In the present invention, the traditional AEPD algorithm can be used to process the TCD foaming test signal, and good results are achieved in MES segmentation. At the same time, in order to improve the segmentation accuracy of the complex features of the TCD foaming test, a physiological time series segmentation model integrating the attention mechanism is used for the first time to perform feature classification and segmentation on the TCD foaming test signal of PFO. U-Time is used to capture multi-scale injection information, and low-level features are combined with high-level features. The key features in the signal are focused on by combining the attention mechanism to effectively segment the signal. The model performed the TCD foaming test segmentation task on the MES and rain curtain datasets. The results show that this method has achieved good results in the segmentation of TCD foaming test features, and provides a medical standard for the four-classification of PFO pore size. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the process of the present invention.

[0026] Figure 2 (a) is an audio diagram of an independent microembolic signal of the present invention.

[0027] Figure 2 (b) is a time domain signal diagram of an independent microembolic signal of the present invention.

[0028] Figure 3(a) is the audio diagram of the classic rain curtain of the present invention.

[0029] Figure 3 (b) is the time domain signal diagram of the classic rain curtain of the present invention.

[0030] Figure 4 (a) is an audio diagram of the front end of the rain curtain of the present invention.

[0031] Figure 4 (b) is a time domain signal diagram of the front end of the rain curtain of the present invention.

[0032] Figure 5 It is a schematic diagram of the relationship between the rain curtain type and its duration and the PFO aperture of the present invention.

[0033] Figure 6 It is a network architecture diagram of the recognition model of the present invention.

[0034] Figure 7 (a) is a schematic diagram of the performance evaluation of the three segmentation models of the present invention on MES.

[0035] Figure 7 (b) is a schematic diagram of the performance evaluation of the three segmentation models of the present invention on Rain Curtain.

[0036] Figure 8 This is a diagram of the segmentation results of the present invention using different models on the MES dataset.

[0037] Fig. 9 This is a segmentation result diagram of the present invention using different models on the rain curtain dataset.

[0038] Fig.10 It is a partial enlarged view of different models of the present invention in rain curtain segmentation.

[0039] Fig.11 It is a schematic diagram of the interface of the system of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0041] The TCD contrast-enhanced Transcranial Doppler (c-TCD) test is to inject microbubble ultrasound contrast agent from the cubital vein and monitor the intracranial arteries for microembolic signals (MES) to screen for the presence of RLS. This method is safe, non-invasive, easy to operate, economical and practical, and can determine the size of the shunt volume, but it does not yet have the ability to predict the size of the shunt aperture. In the present invention, RLS is right-to-left shunt, PFO is patent foramen ovale, MES is microembolic signal, CRC is classic rain curtain, RCF is the front end of the rain curtain, TEE is transesophageal echocardiography, and ICE is intracardiac echocardiography.

[0042] The present invention proposes an automated evaluation method for the patent foramen ovale diameter based on TCD foaming test signal characteristics, the process is as follows: Figure 1 As shown, specifically including S1, obtaining the audio record of the clinical TCD blister test; S2, identifying the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain front ends in the audio record of the TCD blister test; S3, determining the size of the shunt aperture according to the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain front ends.

[0043] The research data used in this invention include audio recordings of clinical TCD blister tests of 177 PFO patients and the PFO apertures obtained by TEE exploration, all of which were provided by the Department of Neurology of the Affiliated Hospital of Chongqing Medical University and approved by the Ethics Committee of the hospital. All patients signed informed consent. The TCD acquisition equipment is the China Delikai EMS-9EA ultrasonic blood flow analyzer with a probe frequency of 1.6MHz.

[0044] Since there is no clear classification standard for the TCD blister test signal characteristics of RLS large flow, the present invention defines the following three categories based on microemboli density and duration:

[0045] ① Independent microemboli. They have typical pulse-type audio characteristics such as Figure 2 As shown in (a), the time domain signal is Figure 2 (b) shows a spindle-like shape with a duration of less than 300ms, and the number of occurrences can be accurately counted through feature detection.

[0046] ② Classic Rain Curtain CRC. Figure 3 (a) and Figure 3As shown in (b), the microembolic signals may be very dense and difficult to count, which is called CRC. In CRC, the MES signals are completely connected and the duration is ≥0.5s, and the time interval between two CRC segments is ≥10ms;

[0047] ③Rain Curtain Front RCF. RCF is Figure 4 (a) and Figure 4 As shown in (b), the density of microemboli is relatively low compared with CRC. There are rain curtain-like signals of multiple microemboli overlapping and merging with each other, but the duration is less than 0.5s. There are also some identifiable independent MES, but the interval is ≤0.3s and the density is ≥20 / second. Figure 4 (b) as shown.

[0048] Since the duration and shape of RCF and CRC are somewhat random, the difficulty of identification is greatly increased. Especially when there is clipping distortion (popping sound), a large amount of irregular noise will be generated in the frequency domain, so the present invention focuses on the analysis of time domain characteristics.

[0049] Based on the above TCD blister test signal feature classification criteria, the number of microemboli in all research data, as well as the duration and frequency of RCF and CRC were manually analyzed, and then the correlation between the TCD blister test signal features and the PFO opening diameter was established, as follows: Figure 5 shown.

[0050] The results showed that for PFO patients with only MES in the TCD foaming test signal, the aperture is usually less than 0.5 mm; if RCF is included and there is no CRC, when the total duration of RCF is <5s, the aperture is between 0.5 and 1.5 mm; when the total duration of RCF is ≥5s or CRC appears, it usually indicates the presence of a larger hole with a diameter of more than 1.5 mm. Therefore, a four-level classification standard for TCD foaming test and PFO aperture can be established, as shown in Table 1.

[0051]

[0052]

[0053] Table 1

[0054] Although the relationship between the rain curtain feature and the size of the shunt aperture is obtained above, there is a lack of methods for identifying the rain curtain feature in the audio recording of the TCD foaming test in the prior art. In the present invention, two methods, deep learning method and adaptive endpoint detection method, are used to realize the TCD foaming test feature recognition and segmentation. Then, according to the segmentation results, their number and duration are counted respectively, and then compared with the PFO aperture grading standard to realize automated pathological evaluation.

[0055] 1. Deep Learning Methods

[0056] In order to facilitate the classification and identification of TCD foaming test signals, the present invention labels the TCD foaming test audio records and establishes a feature data set. Since MES has significant differences in signal duration compared with CRC and RCF, they are constructed as two data sets respectively:

[0057] The rain curtain dataset, based on the feature classification criteria mentioned above, manually annotated the RCF and CRC that appeared in the TCD foaming test audio. To ensure the consistency of the length of the annotated data, the labeled audio was segmented into a segment of 8 seconds, and a total of 409 segments were constructed to construct the rain curtain dataset, which served as the basis for the subsequent establishment of the recognition model.

[0058] MES dataset,Based on the MES features, we extracted 120 high-quality MES signals from the original data. Then, we randomly inserted the above MES signals into the 8-second TCD baseline audio and ensured that there was no overlap between them, thus generating a MES dataset containing 1,200 audio clips.

[0059] Inspired by U-Time and Attention U-Net, a physiological time series architecture integrating attention mechanism is designed in this paper, such as Figure 6 As shown in the figure, named Attention U-Time (AUT), it consists of three modules: encoder, decoder and attention gate (AG). First, its encoder-decoder structure is used to effectively capture the local and global features of the target signal. Secondly, AG is integrated in the architecture and connected between each upsampling block of the decoder and the feature map of the corresponding scale of the encoder to enhance the decoder's attention to important features, thereby effectively improving the signal segmentation accuracy.

[0060] ①U-Time as the basic architecture. U-Time uses an encoder-decoder architecture and a temporal convolutional network to segment 1D physiological time series. The encoder can gradually extract the hierarchical features of the TCD foaming test signal, and can capture the long-range dependencies in the TCD foaming test signal by increasing the kernel size and padding settings of the convolution layer. During the downsampling process of the encoder, the feature map reduces the temporal dimension layer by layer, but retains the high-dimensional semantic information to ensure that feature information is captured at multiple scales. The decoder reconstructs these features in reverse and restores the temporal resolution through upsampling operations. Unlike the encoder, the decoder combines the upsampled feature maps in each block for convolution processing. With this design, the model can restore the resolution of the TCD foaming test signal layer by layer and reconstruct the temporal features from coarse to fine, thereby more accurately retaining important details in the signal.

[0061] ② Attention mechanism. This paper introduces the attention mechanism into the standard encoder-decoder architecture of U-Time to enhance the model's attention to specific features and its ability to filter interference information in complex backgrounds, thereby improving the segmentation accuracy of the target area of ​​the physiological time series. The entire processing process is based on AG, which screens the key features transmitted by the decoder through jump links, and helps the decoder focus on the segmentation target of the model when reconstructing features. This module does not significantly increase the computational overhead, and can effectively improve the accuracy and robustness of the model. Its detailed working process is shown in Formula (1)-Formula (3).

[0062]

[0063] α=δ(W ψ (ReLU(q g +q x ))+b ψ ) (2)

[0064] x′=α·x (3)

[0065] In the formula, g is the global feature output by the encoder, x is the local feature output by the decoder, and W g and W x is the convolution kernel weight, b g and b x is the bias term; first, formula (1) linearly maps the global and local features of the input, and then formula (2) converts q g and q x Add and apply the nonlinear activation function ReLU, and then generate the attention weight α through a single-channel convolution layer. Finally, formula (3) multiplies the generated attention weight α with the input feature x to obtain the weighted feature x′, thereby completing the selective filtering of the feature.

[0066] 2. Adaptive endpoint detection method

[0067] Endpoint detection (EPD) is a traditional segmentation algorithm that finds the start and end endpoints of audio features. In the time domain, threshold judgment is often made based on the sudden change of volume and zero crossing rate (ZCR). Since the volume of each characteristic signal of the TCD bubble test is much higher than the background sound of blood flow and the frequency is also very different, the application of EPD can obtain a better segmentation effect. However, the echo intensity and frequency of the MES group are also related to the width of the temporal window of the skull and its flow velocity, which makes the optimal segmentation threshold of different TCD bubble test samples vary greatly. This paper introduces adaptive threshold calculation to improve the robustness of the EPD algorithm, which is called AdaptiveEPD (AEPD):

[0068] ① Calculation of frame energy. Frame energy E is an indicator that describes the strength of the audio signal. The calculation is shown in formula (4). Where x[n] represents the amplitude of the sampling point in the current frame, and N is the frame length.

[0069]

[0070] ② Calculation of ZCR. ZCR refers to the number of times each frame signal crosses the zero point, which can reflect the oscillation frequency and energy change of the signal. The calculation is shown in formula (5). Among them, x[n] represents the sampling point value in the current frame, N is the length of the current frame, and II(g) is the indicator function, which takes the value of 1 when the condition is true, otherwise it is 0.

[0071]

[0072] ③Adaptive threshold algorithm. Volume threshold T E The setting is based on the global frame energy of the signal, distinguishing the dynamic range of the volume characteristics of the background area and the segmented area, ensuring that the threshold can adapt to the amplitude change of the signal. The volume threshold is defined as shown in formula (6), V min is the 5th percentile of the volume series, V max is the 95th percentile of the volume sequence, α is the adjustment coefficient, which controls the threshold value; the zero-crossing rate threshold T ZCR Dynamic adjustment is performed based on the ZCR mean of the local frame. The calculation method is shown in formula (7), where N is the current frame number, z i is the frame zero-crossing rate, and β is the adjustment coefficient, which controls the sensitivity of the threshold to the signal.

[0073] T E =V min +α(V max -V min ) (6)

[0074]

[0075] By processing the TCD foaming test audio in frames (a short time window of 30 milliseconds), the volume and zero-crossing rate of each frame are calculated, and the signal is classified according to the adaptive threshold. The typical feature of MES is an instantaneous high-energy spike signal, and its ZCR is also significantly higher than the background noise. AEPD first screens out the segments that may contain MES by setting the volume threshold, and then further judges the suspicious segments by the zero-crossing rate threshold. When both thresholds are met, these segments are marked as MES. In contrast, the rain curtain signal presents continuous high energy, and the ZCR remains in a relatively stable range. After the rain curtain signal is analyzed by AEPD, the suspicious CRC and RCF segments are screened by the volume threshold and the zero-crossing rate threshold, and the CRC and RCF are judged by analyzing the continuity of the volume and ZCR between adjacent frames, using the characteristics of the rain curtain signal with a certain duration.

[0076] 3. Performance evaluation

[0077] The evaluation process is mainly achieved by quantifying the difference between the sample segmentation results and the expert annotation results. Six evaluation indicators are used in this paper: loss function (LOSS), accuracy (ACC), true positive rate (TPR), true negative rate (TNR), positive predictive value (Positive Predictive Value, PPV) and F-measure (F1). Among them, the LOSS indicator is calculated by combining the binary cross entropy loss function (Binary Cross Entropy Loss, BCELoss) and the Dice loss function. BCELoss is used to measure the difference between the predicted probability and the true label, and DiceLoss is used to optimize the shape similarity between the predicted and true labels. The mathematical expression is shown in formula (8). Where λ is the weight parameter, N is the number of samples, and y is the weight parameter. i is the true label, p i is the probability value predicted by the model.

[0078]

[0079] The mathematical expressions of the last five indicators are shown in formula (9). Among them, TP is the number of true positive samples; TN is the number of true negative samples; FP is the number of false positive samples; and FN is the number of false negative samples.

[0080]

[0081] 4. Experiments and Results

[0082] The AUT deep learning network was constructed using the PyTorch framework, and then trained on the MES and rain curtain datasets to build the corresponding segmentation models. The training GPU model was NVIDIA GeForce RTX 3090, the batch size was 32, and the total training cycle was 150 times. In order to better compare the model performance, the U-Time model was also trained in the same way. The segmentation capabilities of the three models of AUT, U-Time, and AEPD were then evaluated. The results are shown in Tables 2, 3, and Figure 7 (a) and Figure 7 (b) It should be noted that AEPD is a traditional algorithm and does not require model training.

[0083]

[0084] Table 2

[0085]

[0086] Table 3

[0087] From the above evaluation results, it can be seen that the AUT model has shown higher accuracy in most indicators. AEPD has the lowest computational overhead, but its accuracy is slightly lower. However, on Rain Curtain, the global accuracy of AEPD, U-Time, and AUT are 0.5703 / 0.6023 / 0.9731, respectively, indicating that the introduction of the attention mechanism significantly improves the model performance. Especially on MES, the F1 of U-Time and AUT are 0.4717 / 0.9485, and 0.1667 / 0.6864 on Rain Curtain, indicating that AUT achieves a balance between precision and recall, and shows high robustness and reliability. The MES segmentation results are shown in Figure 2. Figure 8 As shown in the figure, AUT and U-Time can complete signal segmentation very accurately, while the AEPD model has some deficiencies. However, MES usually only needs to perform quantity statistics, and AEPD can also complete recognition and counting tasks well.

[0088] The segmentation results for RCF and CRC are as follows Fig. 9 and Fig.10 As shown in the figure, due to the complexity of the target features, the segmentation performance of AEPD and U-Time near the interface is relatively rough. With the help of the attention block, the AUT model can reconstruct the target features more accurately and effectively maintain consistency with the true value. This is very important for providing more accurate rain curtain duration statistics.

[0089] In order to assist doctors in evaluating patients' conditions conveniently and efficiently, the present invention designs a "foaming test auxiliary evaluation system" based on the AUT model and the four-level classification standard of PFO aperture. The software interface is as follows Fig.11As shown, it has the functions of TCD foaming test audio file loading, feature segmentation, event log, automatic evaluation, etc. Among them, the upper figure shows the analyzed TCD foaming test audio signal, divides the occurrence areas of RCF and CRC, and marks the MES detection location and number. In addition, each RCF and CRC area can be selected, and the corresponding audio and spectrogram are enlarged and displayed below, which is convenient for doctors to confirm the segmentation results. The event log on the right records in detail the time point and duration of various rain curtains, as well as the automatic RLS and PFO grading evaluation conclusions.

[0090] The clinical data of patients who had both TCD foam test and TEE examination were analyzed using the foam test auxiliary evaluation system. In the data set, there were 18 people with no PFO aperture (class 0); 27 people with PFO aperture 0-0.5mm (class 1); 12 people with PFO aperture 0.5-1.5mm (class 2); and 33 people with PFO aperture greater than 1.5mm (class 3). The foam test auxiliary evaluation system was used to automatically grade and evaluate the patient data, with an accuracy of nearly 90%.

[0091] In this invention, we deeply analyzed the characteristics of TCD foaming test signals, proposed the discrimination methods of MES, RCF and CRC, and established the four-classification standard of TCD foaming test and PFO aperture size for the first time in combination with the clinical data of transesophageal ultrasound. We tried to process TCD foaming test signals with the traditional AEPD algorithm, and achieved good results in MES segmentation. At the same time, in order to improve the segmentation accuracy of complex features of TCD foaming test, a physiological time series segmentation model integrating attention mechanism was used for the first time to perform feature classification and segmentation on TCD foaming test signals of PFO. The model makes full use of the advantages of U-Time and attention mechanism, uses U-Time to capture multi-scale information, and combines low-level features with high-level features, and combines attention mechanism to focus on the key features in the signal, effectively segmenting the signal. The model performed TCD foaming test segmentation tasks on MES and rain curtain datasets. The results show that this method has achieved good results in TCD foaming test feature segmentation, and provides a medical standard for the four-classification of PFO aperture size.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automated evaluation method for the patent foramen ovale diameter based on TCD foaming test signal characteristics, characterized in that: The following steps are involved: S1, obtain audio recording of clinical TCD foaming test; S2, identifying the number of independent microemboli in the audio recording of the TCD blister test, the number and duration of the classic rain curtain, and the number and duration of the front end of the rain curtain; S3, determining the size of the shunt aperture according to the number of independent microemboli, the number and duration of the classic rain curtains, and the number and duration of the front ends of the rain curtains; The independent microemboli are audio signals with typical pulse-type audio characteristics, a duration of less than 300 ms, and the number of occurrences of which can be accurately counted; The classic rain curtain is an audio signal that is connected into one piece and has a duration of ≥0.5s, and the time interval between two classic rain curtains is ≥10ms; The front end of the rain curtain is a rain curtain-like signal including a plurality of microemboli with a duration of less than 0.5 s that overlap and fuse with each other, and independent microemboli with a spacing of ≤ 0.3 s and a density of ≥ 20 per second.

2. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 1, characterized in that: In S1, a TCD contrast-enhanced test was performed by injecting microbubble ultrasound contrast agent from the patient's cubital vein, and the intracranial large arteries were simultaneously tested for the presence of microembolic signals to obtain a clinical TCD bubble test audio recording.

3. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 1, characterized in that: In S2, the audio signals in the TCD bubble test were divided into independent microemboli, classic rain curtains, and rain curtain front ends according to the density and duration of microemboli in the audio recording of the TCD bubble test.

4. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 3, characterized in that: In S2, the TCD blister test audio recordings were identified using the attention model, and the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain fronts were obtained.

5. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 4, characterized in that: The attention model is a physiological time series segmentation model that integrates the attention mechanism, and includes a basic architecture of an encoder, a decoder, and an attention gate; the encoder and the decoder are used to capture local and global features in the TCD foaming test audio recording, and the attention gate is connected between each upsampling block of the decoder and the feature map of the corresponding scale of the encoder to enhance the decoder's attention to important features.

6. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 3, characterized in that: In S2, the audio recording of the TCD foaming test is framed, the volume and zero-crossing rate of each frame are calculated, and the number of independent microemboli, the number and duration of classic rain curtains, and the number and duration of rain curtain fronts are obtained based on the adaptive threshold.

7. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 6, characterized in that: In S2, frame energy E is an indicator describing the strength of the audio signal. The calculation formula is shown in (4), where x[n] represents the amplitude of the sampling point in the current frame, and N is the frame length; ZCR refers to the number of times each frame signal crosses the zero point, which can reflect the oscillation frequency and energy change of the signal. The calculation formula is shown in (5); where x[n] represents the sampling point value in the current frame, N is the length of the current frame, and II(g) is the indicator function, which takes the value of 1 when the condition is true, otherwise it takes the value of 0; Volume threshold T E The setting is based on the global frame energy of the signal, and the dynamic range of the volume characteristics of the background area and the segmented area is distinguished to ensure that the threshold can adapt to the amplitude change of the signal; the volume threshold is defined as shown in (6), V min is the 5th percentile of the volume series, V max is the 95th percentile of the volume sequence, α is the adjustment coefficient, which controls the threshold value; the zero-crossing rate threshold T ZCR Dynamic adjustment is performed based on the ZCR mean of the local frame. The calculation method is shown in (7). T E =V min +α(V max -V min ) (6) N is the current frame number, z i is the frame zero-crossing rate, and β is the adjustment coefficient, which controls the sensitivity of the threshold to the signal.

8. The method for automatically evaluating the patent foramen ovale diameter based on TCD foaming test signal characteristics according to claim 1, characterized in that: In S3, when the number of independent microemboli is 0, the number of classic rain curtains is 0, and the number of rain curtain front ends is 0, the diversion aperture is 0 mm; When the number of independent microemboli is greater than 0, the number of classic rain curtains is 0, and the number of front ends of the rain curtains is 0, the diversion aperture is greater than 0 mm and less than 0.5 mm; When the number of rain curtain front ends is greater than 0, and the duration of the rain curtain front end is less than 5s, and the number of classic rain curtains is 0, the diversion aperture is greater than or equal to 0.5mm and less than or equal to 1.5mm; When the number of rain curtain front ends is greater than 0 and the duration of the rain curtain front end is greater than 5s, the diversion aperture is greater than 1.5mm; or when the number of classic rain curtains is greater than 0, the diversion aperture is greater than 1.5mm.

9. A system, characterized in that: Used to implement the method according to any one of claims 1 to 8.

Images