Aortic valve function evaluation device

Abstract

The invention discloses an aortic valve function evaluation device which comprises a processor and a memory, and the memory stores programs or instructions capable of running on the processor. When the program or the instruction is executed by the processor, the following steps are realized: acquiring a first echocardiogram of a target object corresponding to a final systole and a second echocardiogram of a target object corresponding to a final diastole in a cardiac cycle, and acquiring medical record data of the target object; determining a first mask pattern corresponding to the first echocardiogram and a second mask pattern corresponding to the second echocardiogram according to the first echocardiogram, the second echocardiogram and an aortic valve segmentation model; determining an aortic valve morphological index of the target object according to the first mask pattern and the second mask pattern; determining aortic valve functional indexes of the target object according to the medical record data of the target object; and determining and outputting an aortic valve function evaluation result of the target object according to the first mask pattern, the second mask pattern, the aortic valve morphological index and the aortic valve functional index.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical treatment, and more particularly, to an aortic valve function evaluation device. BACKGROUND

[0002] Currently, aortic valve function evaluation includes a measurement stage and a grading stage. In the measurement stage, aortic valve structure in an end-systolic echocardiogram needs to be manually segmented by a human, and morphological parameters (such as calcification area, leaflet thickness, and leaflet area) of the aortic valve are manually measured. In the grading stage, the measured morphological parameters (such as calcification area, leaflet thickness, and leaflet area) are compared with reference thresholds by a human, and the function of the aortic valve is evaluated according to experience. The entire process completely relies on manual reading and threshold comparison, and it usually takes 10-15 minutes to evaluate one case, which is low in efficiency. Moreover, different doctors with different years of experience may give different function evaluation results for the same case, resulting in low consistency of aortic valve function evaluation results. In addition, the measurement stage is based on manual segmentation of aortic valve structure, and the segmentation accuracy is low, which seriously affects the accuracy of the evaluation results in the grading stage. SUMMARY

[0003] An object of embodiments of the present application is to provide a new technical solution for aortic valve function evaluation, to solve the problems of low accuracy, low efficiency, and poor consistency in evaluating aortic valve function by a human in the related art, and to improve the accuracy and efficiency of aortic valve function evaluation.

[0004] According to a first aspect of the present application, an aortic valve function evaluation device is provided, comprising a processor and a memory, the memory storing programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the following steps: obtaining a first echocardiogram corresponding to an end-systolic period and a second echocardiogram corresponding to an end-diastolic period in a cardiac cycle of a target object, and obtaining medical record data of the target object; determining a first mask image corresponding to the first echocardiogram and a second mask image corresponding to the second echocardiogram according to the first echocardiogram, the second echocardiogram, and an aortic valve segmentation model; wherein pixel values of pixel points corresponding to aortic valve tissue in the first mask image and the second mask image are first pixel values, and pixel values of pixel points corresponding to non-aortic valve membrane tissue in the first mask image and the second mask image are second pixel values; determining aortic valve morphological indexes of the target object according to the first mask image and the second mask image; determining aortic valve function indexes of the target object according to the medical record data of the target object; Based on the first mask image, the second mask image, the aortic valve morphological indicators, and the aortic valve functional indicators, the aortic valve function assessment results of the target object are determined and output.

[0005] Optionally, determining and outputting the aortic valve function assessment result of the target object based on the first mask image, the second mask image, the aortic valve morphological indices, and the aortic valve functional indices includes: The first mask image, the second mask image, the aortic valve morphology index, and the aortic valve function index are input into the aortic valve function assessment model to obtain and output the aortic valve function assessment results of the target object.

[0006] Optionally, the aortic valve morphological parameters include at least one of leaflet thickness, leaflet area, leaflet circumference, calcification area, and valve orifice area. Determining the aortic valve morphological parameters of the target object based on the first mask and the second mask includes: Based on the first mask image, determine the valve orifice area, leaflet thickness, and calcification area of ​​the aortic valve; Based on the second mask image, the leaflet area and leaflet circumference of the aortic valve are determined.

[0007] Optionally, the aortic valve functional parameters include at least one of peak aortic valve velocity, mean differential pressure, and regurgitation fraction.

[0008] Optionally, determining the aortic valve function parameters of the target subject based on the target subject's medical record data includes: Medical word segmentation is extracted from the medical record data of the target object to obtain a medical word segmentation sequence; The medical word segmentation sequence is input into the medical enhanced BERT model to obtain an N×768-dimensional embedding sequence; Aortic valve function parameters of the target object are extracted from the N×768-dimensional embedding sequence.

[0009] Optionally, the aortic valve function assessment result of the target object includes an aortic valve function assessment type, which includes one of normal, aortic stenosis, and aortic regurgitation. If the aortic valve function assessment result of the target object is aortic stenosis or aortic regurgitation, the aortic valve function assessment result also includes an aortic valve function assessment grade, which includes one of mild, moderate, and severe.

[0010] Optionally, the aortic valve segmentation model is determined through the following steps: Obtain a first training sample set; wherein, each first training sample in the first training sample set includes a sample echocardiogram and a sample aortic valve mask image corresponding to the sample echocardiogram, wherein the pixel value of the pixel point corresponding to the aortic valve tissue in the sample aortic valve mask image is a first pixel value, and the pixel value of the pixel point corresponding to the non-aortic valve tissue in the sample aortic valve mask image is a second pixel value. The aortic valve segmentation model is trained using the first training sample set to obtain the trained aortic valve segmentation model.

[0011] Optionally, the aortic valve function assessment model is determined through the following steps: Obtain a second training sample set; wherein each second training sample in the second training sample set includes a first sample mask, a second sample mask, sample aortic valve morphological indices, sample aortic valve functional indices, and sample aortic valve functional assessment type. The aortic valve function assessment model is trained using the second training sample set to obtain the trained aortic valve function assessment model.

[0012] Optionally, obtaining the second training sample set includes: Multiple sample data corresponding to multiple sample objects are obtained to obtain a sample dataset; wherein, the sample data of the sample object includes the first sample echocardiogram corresponding to the end of systole and the second sample echocardiogram corresponding to the end of diastole within the cardiac cycle, the sample medical record data of the sample object, and the sample aortic valve function assessment type of the sample object. For each sample object, based on the first sample echocardiogram, the second sample echocardiogram, and the aortic valve segmentation model of the sample object, determine the first sample mask image corresponding to the first sample echocardiogram and the second sample mask image corresponding to the second sample echocardiogram; Based on the first and second sample mask images of the sample object, the aortic valve morphological parameters of the sample object are determined. Based on the sample medical record data of the sample subjects, the sample aortic valve function indicators of the sample subjects were determined; The first sample mask, the second sample mask, the aortic valve morphology index, the aortic valve function index, and the aortic valve function assessment type of the sample object are used as a second training sample corresponding to the sample object. The second training sample set is obtained based on the multiple second training samples corresponding to the multiple sample objects.

[0013] Optionally, the aortic valve function assessment model is a Transformer-based cross-modal fusion model.

[0014] One beneficial effect of this application is that the aortic valve segmentation model can automatically segment the aortic valve mask image from echocardiography, thereby avoiding the problems of poor accuracy and low efficiency of manual segmentation and improving the accuracy and efficiency of aortic valve segmentation. Automatic measurement of aortic valve morphological parameters based on the aortic valve mask image can improve measurement efficiency. By determining the aortic valve functional parameters of the target subject based on their medical record data, and then determining and outputting the aortic valve function assessment results based on the aortic valve mask image, aortic valve morphological parameters, and aortic valve functional parameters, the aortic valve function assessment of the target subject can be realized. Furthermore, this assessment method can solve the problems of low accuracy, low efficiency, and poor consistency in related technologies that rely on manual aortic valve function assessment, thus improving the accuracy and efficiency of aortic valve function assessment. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the present application and, together with their description, serve to explain the principles of the present application.

[0016] Figure 1 This is a schematic diagram of the structure of an aortic valve function assessment device according to an embodiment of this application; Figure 2 This is a schematic flowchart of an aortic valve function assessment method provided according to an embodiment of this application. Detailed Implementation

[0017] Various exemplary embodiments of this application will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this application.

[0018] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0019] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0020] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0021] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0022] Figure 1 This is a schematic block diagram of the structure of an aortic valve function assessment device 100 provided according to an embodiment of the present disclosure.

[0023] The aortic valve function assessment device 100 can be a terminal or a server to alleviate the computing power of the terminal. Furthermore, the aortic valve function assessment device 100 can be a portable computer, tablet computer, PDA, or other terminal. The aortic valve function assessment device 100 can also be a cloud server or a network edge server.

[0024] like Figure 1 As shown, the aortic valve function assessment device 100 includes a processor 110 and a memory 120. The memory 120 stores programs or instructions that can run on the processor 110, and when the programs or instructions are executed by the processor 110, they achieve the following: Figure 2 The method for assessing aortic valve function is shown.

[0025] like Figure 2 As shown, it provides a method for assessing aortic valve function, including the following steps S210 to S250.

[0026] Step S210: Obtain a first echocardiogram of the target object at the end of systole and a second echocardiogram at the end of diastole within the cardiac cycle, and obtain the target object's medical record data.

[0027] In this embodiment, the cardiac cycle can refer to the time required for the heart to complete one full pumping cycle. For example, a cardiac cycle can be the time between the closure of the aortic valve once and the closure of the next aortic valve. One cardiac cycle involves the heart going through a cycle of diastole → systole → diastole.

[0028] End-systole corresponds to the moment when left ventricular ejection ceases. At this time, the aortic valve is about to close, the ventricular cavity is at its smallest, and the valve leaflets are in an adducting state. Measurements of valve orifice area, leaflet thickness, and calcification projection are clearest at this moment, avoiding artifacts caused by obstruction of the orifice.

[0029] End-diastole corresponds to the moment when ventricular filling ceases. At this time, the aortic valve is closed, the ventricular cavity is at its largest, and the valve leaflets are relaxed and unfolded. The valve outline is complete at this point, facilitating the calculation of leaflet area and circumference, and serving as an internationally unified reference frame for volume and diameter measurements.

[0030] Therefore, in order to improve the accuracy of the subsequently determined aortic valve morphological parameters, a first echocardiogram at end-systole and a second echocardiogram at end-diastole were obtained.

[0031] Echocardiography can be either transthoracic echocardiography (TTE) or transesophageal echocardiography (TEE), without limitation here. Transthoracic echocardiography involves placing an ultrasound probe on the surface of the chest to obtain images of the heart's structure and function. Transesophageal echocardiography involves inserting an ultrasound probe into the esophagus to obtain images of the heart's structure and function. Transesophageal echocardiography produces clearer images because the ultrasound probe is closer to the heart.

[0032] Medical record data can include both structured and unstructured medical record data. Structured medical record data can include: age, gender, ethnicity, BMI, and blood pressure.

[0033] Unstructured medical record data may include: medical reports, present medical history, past medical history (especially cardiovascular disease and surgical history), physical signs, laboratory test results (such as BNP, blood lipids) and medication use.

[0034] The present illness history is a narrative of the onset and progression of the current illness, usually presented as a continuous text. The past medical history records previous illnesses, surgeries, injuries, etc., often appearing as descriptive statements such as "hypertension for 10 years, PCI procedure performed in 2015." Physical examination findings are written descriptions written by the doctor after physical examination, such as "clear breath sounds bilaterally, apex beat located 0.5cm medial to the left midclavicular line at the 5th intercostal space." Laboratory test results (such as BNP and blood lipids) are automatically transmitted by the testing system; field names, reference values, units, and high / low arrow indicators are all stored in fixed columns of a relational database. Medication usage is entered through a drug dictionary and prescription codes, with dosage, frequency, route, start and end times strictly coded. All of this content exists in continuous text format without fixed fields or defined vocabulary; therefore, it is classified as unstructured medical record data and requires natural language processing to convert it into a computer-analyzable format.

[0035] After obtaining the first and second echocardiograms in step S210, and before executing step S220, image preprocessing can be performed on the first and second echocardiograms respectively. Image preprocessing can include at least one of resizing and grayscale normalization, which is not limited here.

[0036] Step S220: Based on the first echocardiogram, the second echocardiogram, and the aortic valve segmentation model, determine the first mask image corresponding to the first echocardiogram and the second mask image corresponding to the second echocardiogram.

[0037] In this process, the pixel values ​​of the pixels corresponding to the aortic valve tissue in the first mask image and the second mask image are the first pixel values, and the pixel values ​​of the pixels corresponding to the non-aortic valve tissue in the first mask image and the second mask image are the second pixel values.

[0038] In this embodiment, the first echocardiogram and the second echocardiogram are segmented using an aortic valve segmentation model to obtain a first mask image corresponding to the first echocardiogram and a second mask image corresponding to the second echocardiogram. The aortic valve segmentation model is used to perform pixel-level segmentation of aortic valve tissue and non-aortic valve tissue in the echocardiogram.

[0039] In one example, the aortic valve segmentation model is the nnU-Net network model.

[0040] In one example, the first pixel value is 1 and the second pixel value is 0. That is, the pixel value of the corresponding aortic valve tissue in the first mask and the second mask is 1, and the pixel value of the corresponding non-aortic valve tissue in the first mask and the second mask is 0.

[0041] In one embodiment of this application, the aortic valve segmentation model is determined through the following steps: steps S110 to S120.

[0042] Step S110: Obtain the first training sample set.

[0043] Wherein, each first training sample in the first training sample set includes a sample echocardiogram and a sample aortic valve mask image corresponding to the sample echocardiogram, wherein the pixel value of the pixel point corresponding to the aortic valve tissue in the sample aortic valve mask image is a first pixel value, and the pixel value of the pixel point corresponding to the non-aortic valve tissue in the sample aortic valve mask image is a second pixel value.

[0044] In this embodiment, multiple sample echocardiograms corresponding to multiple sample objects are first obtained. Then, for each sample echocardiogram, the expert changes the pixel value of the pixel corresponding to the aortic valve tissue in the sample echocardiogram to 1 (i.e., the first pixel value), and changes the pixel value of the pixel corresponding to the non-aortic valve tissue in the sample echocardiogram to 0 (i.e., the second pixel value), thus obtaining the sample aortic valve mask image corresponding to the sample echocardiogram, which serves as a first training sample. This allows for the acquisition of multiple first training samples corresponding to multiple sample objects, forming a first training sample set.

[0045] It should be noted that, in order to improve the segmentation effect of the aortic valve segmentation model on echocardiograms at any time during the cardiac cycle, the sample echocardiograms obtained here are not limited to end-systole and end-diastole, but can also be echocardiograms at any other time during the cardiac cycle.

[0046] It is important to note that the aortic valve consists of three leaflets: the left coronary leaflet, the right coronary leaflet, and the non-coronary leaflet. Experts should ensure the integrity of the aortic valve during annotation to improve the segmentation accuracy of the aortic valve segmentation model.

[0047] Step S120: Train the aortic valve segmentation model using the first training sample set to obtain the trained aortic valve segmentation model.

[0048] The aortic valve segmentation model is trained in the conventional way in this step, which will not be described in detail here.

[0049] Step S230: Determine the aortic valve morphological parameters of the target object based on the first mask image and the second mask image.

[0050] In one embodiment of this application, the aortic valve morphological parameters include at least one of leaflet thickness, leaflet area, leaflet circumference, calcification area, and valve orifice area.

[0051] In this embodiment, leaflet thickness refers to the physical thickness of the aortic valve tissue, reflecting the degree of tissue hyperplasia or fibrosis. Leaflet thickening is the result of fibrous tissue hyperplasia, calcium salt deposition, or rheumatic granuloma formation after the valve has been subjected to long-term blood flow impact or inflammatory stimulation. The greater the thickness, the higher the tissue rigidity, the more restricted the leaflet movement, and the smaller the opening area, thus becoming an important indicator for judging the degree of aortic valve stenosis.

[0052] The leaflet area is the projected area of ​​a single leaflet, reflecting its size and degree of deformation. A reduced leaflet area is often due to post-inflammatory adhesions or fibrous contracture, causing the leaflets to shorten and the occlusal edges to fail to open, resulting in a central gap and incomplete closure. An enlarged leaflet area, due to tissue redundancy and length, causes the free edge to droop beyond the valve annulus during closure, leading to prolapse and similarly causing occlusal failure and regurgitation. Therefore, measuring the leaflet area can, to some extent, assess whether the valve is stenotic or incomplete.

[0053] The leaflet circumference reflects the length of the aortic valve's marginal contour. An abnormally increased leaflet circumference may be due to leaflet curling or loosening, causing misalignment of the occlusal margin and leading to insufficiency. A shortened leaflet circumference is caused by calcification or fibrosis, which hardens and constricts the leaflets, restricting the aortic valve orifice and resulting in stenosis. Therefore, circumference is an important quantitative indicator for assessing stenosis and morphological distortion of regurgitation.

[0054] Calcification area is an indicator that quantifies the calcification load of the valve leaflets. The larger the calcification area, the harder the valve leaflets become, the more restricted their movement, and the smaller the aortic valve orifice area, thus directly affecting the degree of stenosis. Furthermore, calcification area is also a core quantitative indicator for predicting disease progression and the timing of surgery; therefore, calcification area has become a primary indicator for assessing aortic valve function.

[0055] The aortic valve orifice area directly quantifies the size of the "valve" through which blood can pass during systole. The smaller the area, the more restricted the flow. It has an inverse continuous relationship with pressure gradient and flow velocity, and is unaffected by fluctuations in heart rate and blood pressure. Therefore, the valve orifice area is the gold standard for assessing the degree of aortic stenosis.

[0056] In these embodiments, step S230, which determines the aortic valve morphological parameters of the target object based on the first mask image and the second mask image, includes steps S2301 to S2302.

[0057] Step S2301: Determine the valve orifice area, leaflet thickness, and calcification area of ​​the aortic valve based on the first mask image.

[0058] In this embodiment, since the aortic valve is about to close at end-systole, the ventricular cavity is at its smallest and the leaflets are in an adducted state, the valve orifice area, leaflet thickness, and calcification area are most clearly measured at this time. Therefore, based on the first mask image, the aortic valve orifice area, leaflet thickness, and calcification area are determined.

[0059] In one example, the aortic valve leaflet thickness is obtained by processing the first mask image through the following steps: First, a Euclidean distance transformation is performed on the connected regions belonging to the aortic valve tissue in the first mask image to obtain the shortest distance field from each pixel value to the nearest boundary. Then, based on this shortest distance field, the thickness distribution map of the leaflet along the normal direction at that pixel is determined. Finally, the average thickness distribution map is taken to obtain the average thickness value of the leaflet.

[0060] In one example, the calcification area of ​​the aortic valve is obtained by processing the first mask image through the following steps: In the first mask image, regions with pixel intensity values ​​greater than or equal to 850 HU are marked as initial calcification masks. The initial calcification mask is then logically ANDed with the first mask image of the corresponding valve leaflet to obtain a calcification mask located only within that valve leaflet. The number of pixels within the calcification mask is counted, and combined with the pixel-to-physical scale coefficient of the first mask image, the pixel count is converted into a physical area to obtain the calcification area of ​​that valve leaflet.

[0061] In one example, the aortic valve orifice area is obtained by processing the first mask image through the following steps: In the first mask image, the free edge contour of the aortic valve orifice is extracted and an orifice mask is generated. The number of pixels within the orifice mask is counted, and combined with a pixel-physical scale coefficient, the pixel count is converted into a physical area to obtain the image-based aortic valve orifice area. Further, using this aortic valve orifice area, LVOT velocity integral, and aortic valve velocity integral measured during the same cardiac cycle, a continuity equation correction is performed to obtain the final aortic valve orifice area.

[0062] Those skilled in the art should understand that the methods for measuring the aortic valve orifice area, leaflet thickness, and calcification area are well-known in the field and will not be described in detail here.

[0063] Step S2302: Determine the leaflet area and leaflet circumference of the aortic valve based on the second mask image.

[0064] In this embodiment, since the aortic valve is closed at end-diastole, the ventricular cavity is at its largest, and the valve leaflets are in a relaxed and unfolded state, it is convenient to calculate the leaflet area and leaflet circumference. Therefore, the aortic valve leaflet area and leaflet circumference can be determined based on the second mask image.

[0065] In one example, the leaflet area of ​​the aortic valve is obtained by processing the second mask image through the following steps: the connected regions corresponding to the aortic valve tissue are identified in the second mask image, the total number of pixels in the region is counted, and the pixel count is converted into physical area by combining the pixel-physical scale coefficient of the image to obtain the leaflet area of ​​the aortic valve.

[0066] In one example, the aortic valve leaflet circumference is obtained by processing the second mask image through the following steps: performing boundary tracing on the connected regions of the aortic valve in the second mask image to generate an ordered chain of boundary points, sequentially accumulating the Euclidean distance between adjacent boundary points in a chain code manner to obtain the total boundary length, and converting the total boundary length into a physical length using a pixel-physical scale coefficient to obtain the leaflet circumference.

[0067] Those skilled in the art should understand that these methods of measuring the aortic valve leaflet circumference and leaflet area are well known in the art, and therefore will not be described in detail here.

[0068] Step S240: Determine the aortic valve function parameters of the target subject based on the target subject's medical record data.

[0069] In this embodiment, the aortic valve function indicators of the target object can be obtained through natural language processing (NLP) of the unstructured medical record data of the target object.

[0070] In one embodiment of this application, the aortic valve functional parameters include at least one of peak aortic valve velocity, mean differential pressure, and regurgitation fraction.

[0071] In this embodiment, peak flow velocity directly reflects the instantaneous maximum blood flow velocity at the narrowest point of the valve orifice. A higher peak flow velocity indicates stenosis to some extent. The greater the peak flow velocity, the more severe the stenosis. A sudden drop in peak flow velocity or bidirectional turbulence may indicate regurgitation to some extent. Therefore, peak flow velocity is an important indicator for assessing the degree of aortic stenosis and regurgitation.

[0072] Mean pressure gradient is the time-averaged instantaneous pressure drop across the valve orifice during systole. A larger mean pressure gradient corresponds to a narrower valve orifice area, and a smaller mean pressure gradient corresponds to a wider valve orifice area. A sudden decrease in mean pressure gradient or the appearance of a bidirectional pressure gradient can, to some extent, reflect regurgitation. Therefore, mean pressure gradient can serve as an important indicator for assessing the degree of aortic stenosis and regurgitation.

[0073] Regurgitation fraction is the proportion of regurgitated blood flow to stroke volume, directly quantifying the percentage of blood returning to the left ventricle after systolic ejection. A regurgitation fraction within the range of 0–20% is considered physiological; a regurgitation fraction greater than or equal to 20% indicates aortic regurgitation; and a regurgitation fraction greater than or equal to 50% corresponds to severe regurgitation. A higher regurgitation fraction indicates a larger volume of regurgitated blood and a heavier load on the left ventricle, thus it is an important indicator for assessing the degree of regurgitation.

[0074] In one embodiment of this application, step S240, which determines the aortic valve function index of the target object based on the target object's medical record data, includes steps S2401 to S2403.

[0075] Step S2401: Extract medical word segments from the medical record data of the target object to obtain a medical word segmentation sequence.

[0076] In this embodiment, cardiovascular terms such as "jet flow," "reflux fraction," and "mean pressure difference" are added to the general medical dictionary. Then, a token segmentation algorithm is used to perform word-by-word maximum matching and rule merging on the medical record data to obtain a token sequence that retains the complete medical semantics, namely the "medical word segmentation sequence."

[0077] For example, if a medical record states "chest tightness and shortness of breath have lasted for three days", a word segmentation algorithm can be used to segment it into chest tightness / shortness of breath / three days. Then, stop words such as "and" and "already" are filtered out to obtain a medical word segmentation sequence that includes chest tightness, shortness of breath, and three days.

[0078] Step S2402: Input the medical word segmentation sequence into the medical enhanced BERT model to obtain an N×768-dimensional embedding sequence.

[0079] In this embodiment, the medical-enhanced BERT model can specifically be a cardiovascular medical semantic-enhanced BERT model.

[0080] The training process of the BERT model for semantic enhancement in cardiovascular medicine is as follows: First, at least 1 million Chinese cardiovascular reports are collected. After de-identification and anonymization, characters are unified, and HTML tags are removed while retaining numerical values ​​and units. Then, approximately 5 million sentences are obtained by a sliding window segmenting the text into 256-512 characters. For each sentence, 15% of the characters are randomly masked (80% replaced with a mask, 10% randomly replaced, and 10% left unchanged), generating "masked sentence - original sentence" paired samples. Starting with the existing BERT-Base weights, only the masked language model task (Masked LM task: randomly masking some words in the input text, training the model to guess the original words based on the context) is performed: the token ID sequence of the masked sentence is input, and the probability distribution of the original tokens at the masked positions is output. The trained model is the BERT model for semantic enhancement in cardiovascular medicine.

[0081] The medical-enhanced BERT model can map a token sequence of arbitrary length to an N×768-dimensional embedding sequence, where N is the total number of tokens and 768 is the hidden unit dimension.

[0082] It should be noted that the training process of the semantic enhancement BERT model for cardiovascular medicine is well known in the field and will not be elaborated here.

[0083] Step S2403: Extract aortic valve function indicators of the target object from the N×768-dimensional embedded sequence.

[0084] In this embodiment, the N×768-dimensional embedded sequence is first averaged to obtain a 1×768 vector, and then the peak flow velocity, average pressure difference, and backflow fraction are obtained by regression through a three-layer fully connected regression head.

[0085] By automatically processing medical record data, aortic valve function indicators can be obtained. This process requires no manual intervention and achieves end-to-end automatic extraction.

[0086] Step S250: Based on the first mask image, the second mask image, the aortic valve morphology indicators, and the aortic valve function indicators, determine and output the aortic valve function assessment results of the target object.

[0087] In one embodiment of this application, the aortic valve function assessment result of the target object includes the aortic valve function assessment type, which includes one of normal, aortic stenosis, and aortic regurgitation. If the aortic valve function assessment result of the target object is aortic stenosis or aortic regurgitation, the aortic valve function assessment result also includes the aortic valve function assessment grade, which includes one of mild, moderate, and severe.

[0088] When determining the aortic valve function assessment results for the target subject, the first and second mask images are only used to confirm that the measurement coordinates are consistent. The aortic valve morphological parameters and aortic valve functional parameters are the main factors affecting the aortic valve function assessment results.

[0089] In one example, aortic valve morphological parameters include valve orifice area and calcification area, while aortic valve functional parameters include peak flow velocity, mean pressure gradient, and regurgitation fraction.

[0090] If the valve orifice area is greater than or equal to a first valve orifice area threshold, the peak flow velocity is less than a first peak flow velocity threshold, the mean differential pressure is less than a first differential pressure threshold, and the regurgitation fraction is less than a first regurgitation fraction threshold, then the aortic valve function assessment result for the target subject is determined to be normal. The first valve orifice area threshold can be, for example, 2.0. The first peak flow rate threshold can be, for example, 2.5. The first differential pressure threshold can be, for example, 20, and the first reflux fraction threshold can be, for example, 20%.

[0091] If the valve orifice area is less than a first valve orifice area threshold, the peak flow velocity is greater than or equal to a first peak flow velocity threshold, and the mean differential pressure is greater than or equal to a first differential pressure threshold, then aortic stenosis is preliminarily determined. The first valve orifice area threshold can be, for example, 2.0. At this point, it is necessary to further determine the degree of aortic stenosis, that is, to perform the following steps: when the valve orifice area is within the first valve orifice area range (e.g., 1.5~2.0... And the peak flow velocity is within the first velocity range (e.g., 2.5~3.0). If the mean pressure gradient is within the first pressure gradient range (e.g., 20-30) and the ratio of calcified area to aortic valve tissue area is within the first ratio range (e.g., 5-15%), then the aortic valve function assessment result is determined to be mild aortic stenosis. If the valve orifice area is within the second valve orifice area range (e.g., 1.0-1.5), then... And the peak flow velocity is in the second velocity range (e.g., 3.0~4.0). If the mean pressure difference is within the second pressure difference range (e.g., 30-40) and the ratio of calcified area to aortic valve tissue area is within the second ratio range (e.g., 15-30%), then the aortic valve function assessment result of the target subject is determined to be moderate aortic stenosis.

[0092] If the valve orifice area is less than the second valve orifice area threshold (e.g., 1.0) ), or, the peak flow rate is greater than or equal to the second peak flow rate threshold (e.g., 4.0). If the mean differential pressure is greater than or equal to the second differential pressure threshold (e.g., 40%), or the ratio of calcified area to aortic valve tissue area is greater than or equal to the calcification percentage threshold (e.g., 30%), then the aortic valve function assessment result of the target subject is determined to be severe aortic stenosis.

[0093] In one example, aortic valve functional parameters include regurgitation fraction, and aortic valve morphological parameters include leaflet thickness, leaflet area, and leaflet circumference.

[0094] If the regurgitation fraction is greater than or equal to the first regurgitation fraction threshold (e.g., 20%), aortic regurgitation is preliminarily determined. At this point, further determination of the regurgitation degree is needed: If the regurgitation fraction is within the first regurgitation fraction range (e.g., 20-30%) and the leaflet thickness is less than the leaflet thickness threshold, the leaflet area is less than the leaflet area threshold, and the leaflet circumference is less than the leaflet circumference threshold, then the target subject's aortic valve function assessment result is determined to be mild aortic regurgitation. If the regurgitation fraction is within the second regurgitation fraction range (e.g., 30-50%) and the leaflet area is greater than or equal to the leaflet area threshold, then the target subject's aortic valve function assessment result is determined to be moderate aortic regurgitation. If the regurgitation fraction is greater than or equal to the second regurgitation fraction threshold (e.g., 50%), or the leaflet thickness is less than the leaflet thickness threshold, then the target subject's aortic valve function assessment result is determined to be severe aortic regurgitation.

[0095] In one embodiment of this application, the aortic valve function assessment device 100 further includes a display ( Figure 1 (Not shown in the image), the display is used to show the aortic valve function assessment results of the target object.

[0096] In one embodiment of this application, step S250 involves determining and outputting the aortic valve function assessment result of the target object based on the first mask image, the second mask image, the aortic valve morphological indicators, and the aortic valve functional indicators, including step S2501.

[0097] Step S2501: Input the first mask image, the second mask image, the aortic valve morphological indicators, and the aortic valve functional indicators into the aortic valve function assessment model to obtain and output the aortic valve function assessment results of the target object.

[0098] In one embodiment of this application, the aortic valve function assessment model is a Transformer-based cross-modal fusion model.

[0099] In this embodiment, the multi-head self-attention and cross-attention mechanisms of Transformer are used to enable deep interaction and fusion of the first mask image, the second mask image, aortic valve morphological indicators, and aortic valve functional indicators, thereby outputting accurate aortic valve function assessment results.

[0100] In one example, the input modalities of the Transformer-based cross-modal fusion model include image and text modalities. The image modal includes a first mask, a second mask, and aortic valve morphological parameters. The text modal includes medical record data and aortic valve functional parameters. If the attention weight of the image modal in the valve orifice region is greater than or equal to a first image attention threshold (e.g., 0.6), and the attention weight of the text modal on keywords such as "stenosis" and "regurgitation" is greater than or equal to a first text attention threshold (e.g., 0.5), and the probability of the Transformer classification head outputting the "normal" category is greater than or equal to a first probability threshold (e.g., 0.8), then the aortic valve functional assessment result of the target object is determined to be normal.

[0101] If the attention weight of the image modality is focused on the leaflet thickening or calcification area, and the attention weight of the text modality is focused on the description related to "stenosis", and the probability of the Transformer classification head outputting the "stenosis" category is greater than or equal to the second probability threshold (e.g., 0.7), then aortic stenosis is preliminarily determined. At this point, further assessment of the degree of stenosis is required: If the Transformer classification head outputs a "mild stenosis" probability within the first stenosis range (e.g., 0.6–0.8), and the image attention weight is distributed in the valve orifice margin region, then the aortic valve function assessment result for the target subject is determined to be mild stenosis; if the Transformer classification head outputs a "moderate stenosis" probability within the first stenosis range (e.g., 0.6–0.8), and the image attention weight is concentrated in the leaflet calcification region, then the aortic valve function assessment result for the target subject is determined to be moderate stenosis; if the Transformer classification head outputs a "severe stenosis" category probability greater than or equal to the third probability threshold (e.g., 0.85), or if the image modal attention is significantly enhanced in the valve orifice insufficiency region, or if the text modal attention weight on the word "severe stenosis" is greater than or equal to the second text attention threshold (e.g., 0.7), then the aortic valve function assessment result for the target subject is determined to be severe stenosis.

[0102] In one embodiment of this application, the aortic valve function assessment model in step S2501 is determined through the following steps: steps S310 to S320.

[0103] Step S310: Obtain the second training sample set.

[0104] Each second training sample in the second training sample set includes a first sample mask, a second sample mask, sample aortic valve morphological indicators, sample aortic valve functional indicators, and sample aortic valve function assessment type.

[0105] In this embodiment, the first sample mask is obtained by segmenting the first sample echocardiogram at end-systole, and the second sample mask is obtained by segmenting the second sample echocardiogram at end-diastole. The aortic function assessment type of the sample includes one of normal, aortic stenosis, and aortic regurgitation. If the aortic function assessment type of the second training sample is aortic stenosis or aortic regurgitation, the second training sample also includes the aortic valve function assessment level, which includes one of mild, moderate, and severe.

[0106] In one embodiment of this application, step S310, which obtains a second training sample set, includes steps S3101 to S3106.

[0107] Step S3101: Obtain multiple sample data corresponding to multiple sample objects to obtain a sample dataset.

[0108] The sample data of the sample object includes a first sample echocardiogram of the sample object at the end of systole and a second sample echocardiogram of the sample object at the end of diastole during the cardiac cycle, the sample medical record data of the sample object, and the sample aortic valve function assessment type of the sample object.

[0109] In this embodiment, multiple sample data corresponding to multiple sample objects can be collected from multiple hospitals. One sample object corresponds to one sample data.

[0110] It should be noted that the first sample echocardiogram, the second sample echocardiogram, and the sample medical record data of the sample object in this step correspond to the first sample echocardiogram, the second sample echocardiogram, and the medical record data in step S210. Therefore, the meaning of the first sample echocardiogram, the second sample echocardiogram, and the sample medical record data in this step will not be explained in detail here.

[0111] Step S3102: For each sample object, based on the first sample echocardiogram, the second sample echocardiogram, and the aortic valve segmentation model of the sample object, determine the first sample mask corresponding to the first sample echocardiogram and the second sample mask corresponding to the second sample echocardiogram.

[0112] In the first sample mask and the second sample mask, the pixel value of the pixel corresponding to the aortic valve tissue is the first pixel value, and the pixel value of the pixel corresponding to the non-aortic valve tissue in the first sample mask and the second sample mask is the second pixel value.

[0113] In this embodiment, the first sample echocardiogram and the second sample echocardiogram are segmented using the aortic valve segmentation model to obtain the first sample mask map corresponding to the first sample echocardiogram and the second sample mask map corresponding to the second sample echocardiogram.

[0114] The first pixel has a value of 1, and the second pixel has a value of 0.

[0115] Step S3103: Determine the aortic valve morphological parameters of the sample object based on the first sample mask and the second sample mask of the sample object.

[0116] In one embodiment of this application, the morphological parameters of the sample aortic valve include at least one of the following: leaflet thickness, leaflet area, leaflet circumference, calcification area, and valve orifice area.

[0117] Since these morphological indicators have already been mentioned above, they will not be elaborated upon here.

[0118] In these embodiments, step S3103, which determines the aortic valve morphological parameters of the sample object based on the first sample mask and the second sample mask of the sample object, includes steps S3103.1 to S3103.2.

[0119] Step S3103.1: Determine the valve orifice area, leaflet thickness, and calcification area of ​​the aortic valve of the sample based on the first sample mask image.

[0120] This step is basically the same as step S2301 above, and will not be described again here.

[0121] Step S3103.2: Determine the leaflet area and leaflet circumference of the sample aortic valve based on the second sample mask.

[0122] This step is basically the same as step S2302 above, and will not be described again here.

[0123] Step S3104: Determine the aortic valve function indicators of the sample object based on the sample medical record data of the sample object.

[0124] This step is basically the same as step S240 above, and will not be described again here.

[0125] In one embodiment of this application, the aortic valve functional parameters of the sample include at least one of peak aortic valve velocity, mean differential pressure, and regurgitation fraction.

[0126] Since these functional indicators have already been mentioned above, they will not be elaborated upon here.

[0127] In one embodiment of this application, step S3104, which determines the aortic valve function index of the sample object based on the sample medical record data of the sample object, includes: steps S3104.1 to S3104.3.

[0128] Step S3104.1: Extract medical word segments from the sample medical record data of the sample object to obtain the sample medical word segment sequence.

[0129] This step is basically the same as step S2401 above, and will not be described again here.

[0130] Step S3104.2: Input the sample medical word segmentation sequence into the medical enhanced BERT model to obtain an N×768-dimensional sample embedding sequence.

[0131] This step is basically the same as step S2402 above, and will not be described again here.

[0132] Step S3104.3: Extract the aortic valve function indicators of the sample object from the N×768-dimensional sample embedding sequence.

[0133] This step is basically the same as step S2403 above, and will not be described again here.

[0134] Step S3105: The first sample mask, the second sample mask, the aortic valve morphology index, the aortic valve function index, and the aortic valve function assessment type of the sample object are used as a second training sample corresponding to the sample object.

[0135] Step S3106: Obtain the second training sample set based on the multiple second training samples corresponding to the multiple sample objects.

[0136] Step S320: Train the aortic valve function assessment model using the second training sample set to obtain the trained aortic valve function assessment model.

[0137] The model training method used in this step is the same as the conventional model training method, and will not be described further here.

[0138] The aortic valve segmentation model can automatically segment the aortic valve mask from echocardiography, thus avoiding the problems of poor accuracy and low efficiency of manual segmentation and improving the accuracy and efficiency of aortic valve segmentation. Automatic measurement of aortic valve morphological parameters based on the aortic valve mask can improve measurement efficiency. By determining the aortic valve functional parameters of the target subject based on their medical record data, and then determining and outputting the aortic valve function assessment results based on the aortic valve mask, aortic valve morphological parameters, and aortic valve functional parameters, aortic valve function assessment of the target subject can be achieved. Furthermore, this assessment method solves the problems of low accuracy, low efficiency, and poor consistency in manual aortic valve function assessment in related technologies, improving the accuracy and efficiency of aortic valve function assessment.

[0139] This application may be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this application.

[0140] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0141] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0142] The computer program instructions used to perform the operations of this application may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing the status information of the computer-readable program instructions. These electronic circuits can execute the computer-readable program instructions to implement various aspects of this application.

[0143] Various aspects of this application are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0144] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0145] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0146] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be well known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.

[0147] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical applications, or technological improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of this application is defined by the appended claims.

Claims

1. An aortic valve function assessment device, characterized in that, It includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to perform the following steps: Acquire a first echocardiogram of the target object at the end of systole and a second echocardiogram at the end of diastole within the cardiac cycle, and acquire the target object's medical record data; Based on the first echocardiogram, the second echocardiogram, and the aortic valve segmentation model, determine the first mask image corresponding to the first echocardiogram and the second mask image corresponding to the second echocardiogram; Based on the first mask image and the second mask image, the aortic valve morphological parameters of the target object are determined; Based on the medical record data of the target subjects, the aortic valve function parameters of the target subjects were determined; Based on the first mask image, the second mask image, the aortic valve morphological indicators, and the aortic valve functional indicators, the aortic valve function assessment results of the target object are determined and output.

2. The apparatus according to claim 1, characterized in that, The step of determining and outputting the aortic valve function assessment result of the target object based on the first mask image, the second mask image, the aortic valve morphological indicators, and the aortic valve functional indicators includes: The first mask image, the second mask image, the aortic valve morphology index, and the aortic valve function index are input into the aortic valve function assessment model to obtain and output the aortic valve function assessment results of the target object.

3. The apparatus according to claim 1, characterized in that, The aortic valve morphological parameters include at least one of leaflet thickness, leaflet area, leaflet circumference, calcification area, and orifice area. Determining the aortic valve morphological parameters of the target object based on the first mask and the second mask includes: Based on the first mask image, determine the valve orifice area, leaflet thickness, and calcification area of ​​the aortic valve; Based on the second mask image, the leaflet area and leaflet circumference of the aortic valve are determined.

4. The apparatus according to claim 1, characterized in that, The aortic valve functional parameters include at least one of the following: peak aortic velocity, mean differential pressure, and regurgitation fraction.

5. The apparatus according to claim 1, characterized in that, The step of determining the aortic valve function parameters of the target subject based on the target subject's medical record data includes: Medical word segmentation is extracted from the medical record data of the target object to obtain a medical word segmentation sequence; The medical word segmentation sequence is input into the medical enhanced BERT model to obtain an N×768-dimensional embedding sequence; Aortic valve function parameters of the target object are extracted from the N×768-dimensional embedding sequence.

6. The apparatus according to claim 1, characterized in that, The aortic valve function assessment result of the target subject includes the aortic valve function assessment type, which includes one of normal, aortic stenosis, and aortic regurgitation. If the aortic valve function assessment result of the target subject is aortic stenosis or aortic regurgitation, the aortic valve function assessment result also includes the aortic valve function assessment grade, which includes one of mild, moderate, and severe.

7. The apparatus according to claim 1, characterized in that, The aortic valve segmentation model was determined through the following steps: Obtain a first training sample set; wherein, each first training sample in the first training sample set includes a sample echocardiogram and a sample aortic valve mask image corresponding to the sample echocardiogram, wherein the pixel value of the pixel point corresponding to the aortic valve tissue in the sample aortic valve mask image is a first pixel value, and the pixel value of the pixel point corresponding to the non-aortic valve tissue in the sample aortic valve mask image is a second pixel value. The aortic valve segmentation model is trained using the first training sample set to obtain the trained aortic valve segmentation model.

8. The apparatus according to claim 2, characterized in that, The aortic valve function assessment model was determined through the following steps: Obtain a second training sample set; wherein each second training sample in the second training sample set includes a first sample mask, a second sample mask, sample aortic valve morphological indices, sample aortic valve functional indices, and sample aortic valve functional assessment type. The aortic valve function assessment model is trained using the second training sample set to obtain the trained aortic valve function assessment model.

9. The apparatus according to claim 8, characterized in that, The process of obtaining the second training sample set includes: Multiple sample data corresponding to multiple sample objects are obtained to obtain a sample dataset; wherein, the sample data of the sample object includes the first sample echocardiogram corresponding to the end of systole and the second sample echocardiogram corresponding to the end of diastole within the cardiac cycle, the sample medical record data of the sample object, and the sample aortic valve function assessment type of the sample object. For each sample object, based on the first sample echocardiogram, the second sample echocardiogram, and the aortic valve segmentation model of the sample object, determine the first sample mask image corresponding to the first sample echocardiogram and the second sample mask image corresponding to the second sample echocardiogram; Based on the first and second sample mask images of the sample object, the aortic valve morphological parameters of the sample object are determined. Based on the sample medical record data of the sample subjects, the sample aortic valve function indicators of the sample subjects were determined; The first sample mask, the second sample mask, the aortic valve morphology index, the aortic valve function index, and the aortic valve function assessment type of the sample object are used as a second training sample corresponding to the sample object. The second training sample set is obtained based on the multiple second training samples corresponding to the multiple sample objects.

10. The apparatus according to claim 2, characterized in that, The aortic valve function assessment model is a Transformer-based cross-modal fusion model.

Images