Visual assessment method for risk after congenital heart disease surgery based on patient similarity

Abstract

The present invention discloses a method for visualizing the postoperative risk assessment of congenital heart disease based on patient similarity, including: constructing a congenital heart disease ontology based on multi-source heterogeneous disease knowledge; extracting multidimensional features from ultrasound report texts of historical congenital heart disease patients to obtain standard ultrasound text features; calculating the similarity distance of ultrasound text features of congenital heart disease patients based on standard ultrasound text features; on the basis of the similarity distance of ultrasound text features, integrating the preoperative and intraoperative information of congenital heart disease patients, calculating the similarity distance matrix between congenital heart disease patients, and generating a congenital heart disease patient map; based on the congenital heart disease patient map, recommending similar patient groups for target patients by a similarity algorithm, and assisting clinical personnel to use the postoperative risk factors of similar patient groups to perform postoperative risk assessment on target patients. The assessment method of the present invention can assist doctors in assessing the physiological and pathological conditions of patients and strengthen perioperative monitoring and management.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for visualizing the postoperative risk assessment of congenital heart disease based on patient similarity. Background Art

[0002] Congenital heart disease (CHD) is the main birth malformation in China and a common disease that seriously endangers children's health. According to the "China Cardiovascular Health and Disease Report 2020", there are regional differences in the detection rate of congenital heart disease in my country, which is about 2.41‰-10.36‰. According to the "China Health Statistics Yearbook 2019", in 2018, the mortality rate of congenital heart disease in urban residents in China was 0.84 / 100,000, and that in rural residents was 1.02 / 100,000. The mortality rate of congenital heart disease in men was higher than that in women. Although with the development of cardiac surgery technology and life monitoring methods, the vast majority of children with congenital heart disease can be treated, patients with congenital heart disease are prone to serious complications such as lung infection, endocarditis, and heart failure after surgery, which brings great pain and economic burden to patients and their families. The ability to predict and be prepared to deal with related complications in a timely manner after surgery is the key to determining the medical quality of a heart center.

[0003] Postoperative complications of congenital heart disease are related to many factors. First of all, it is related to the heart condition of congenital heart disease itself. Congenital heart disease includes dozens of different subtypes, the most typical of which are atrial septal defect (ASD), ventricular septal defect (VSD), patent ductus arteriosus (PDA), etc., and each subtype has different classifications. For example, ASD is divided into primary atrial septal defect and secundum atrial septal defect. Secondary atrial septal defect is further divided into four types: central defect (fossa ovalis defect), superior cavity defect (sinus venosus defect), inferior cavity defect and mixed defect. Even for the same subtype, the size of the defect varies from physiological to critical. At present, echocardiography is mostly used in the clinical evaluation of congenital heart disease, which has the advantages of being intuitive, non-invasive, safe, and repeatable. It can provide information on cardiac anatomical structure, cardiovascular function, and hemodynamics. It can also be used as the main tool for intraoperative monitoring and postoperative follow-up of congenital heart disease. At present, in most medical institutions, ultrasound doctors use unstructured free text to record the size of cardiac structures, functional measurements and preliminary diagnoses observed during the examination. Clinicians evaluate the development of congenital heart disease based on the content of the report and make corresponding clinical diagnosis and treatment decisions. Secondly, relevant information during surgery is also a major risk factor for postoperative complications. Surgery is currently the main means of treating complex congenital heart disease. However, due to differences in patient age, condition and surgical method, patients have different postoperative manifestations and risks. The length of stay in the ICU after surgery reflects the prognosis of children. The longer the stay in the ICU, the higher the risk of postoperative complications and death. Assisted ventilation is extremely important after heart surgery, especially for infants and young children. Extubation too early directly affects respiratory function and increases the reinsertion rate. If not handled in time, cardiac function is affected, and even respiratory and circulatory failure and death may occur. Extubation too late will increase the incidence of pulmonary complications and prolong hospitalization. On the other hand, the longer the hospital stay and the length of stay in the ICU, the more medical resources are consumed and the higher the medical expenses. Therefore, making full use of preoperative and intraoperative information to evaluate patients' postoperative risk factors can effectively assess patients' physiological and pathological conditions at an early stage, strengthen perioperative monitoring and management, and provide early warning for high-risk patients after surgery, which is conducive to reducing complications and mortality.

[0004] At present, there are two main scoring systems for the risk of congenital heart disease surgery abroad, namely the Risk Grading Score for Congenital Heart Surgery (RACHS-1) and the Aristotle score. RACHS-1 only grades the surgery and does not consider preoperative risk factors. It is suitable for predicting patient groups, but cannot be used to predict individual mortality or postoperative morbidity. The Aristotle scoring system is divided into the Aristotle Basic Score (ABC) and the Aristotle Complex Score (ACC). The ABC score scores the mortality, morbidity, and complexity of congenital heart disease surgery without considering individual differences in patients. On this basis, the ACC score incorporates more individual information including age, but the ACC can only predict 30-day mortality and the length of stay in the intensive care unit within one week after surgery, and cannot predict more postoperative risk factors.

[0005] With the deepening of medical informatization, a large amount of clinical data is constantly accumulated. It is of great significance to make full use of the large amount of data accumulated during the treatment of congenital heart disease patients for postoperative risk assessment. Some machine learning-based methods for predicting postoperative complications of congenital heart disease have been proposed, but the lack of clear clinical interpretability in actual application has affected the clinical transformation of related technologies. Summary of the invention

[0006] The present invention provides a method for visualizing the postoperative risk assessment of congenital heart disease based on patient similarity, which can assist doctors in evaluating the physiological and pathological conditions of patients and strengthen perioperative monitoring and management.

[0007] The technical solution of the present invention is as follows:

[0008] A method for visually assessing the postoperative risk of congenital heart disease based on patient similarity comprises the following steps:

[0009] (1) Based on multi-source heterogeneous disease knowledge, a congenital heart disease ontology is constructed from the anatomical structure level;

[0010] (2) Multidimensional feature extraction was performed on ultrasound report texts of historical CHD patients and mapped to standard terms in the CHD ontology to obtain standard ultrasound text features;

[0011] (3) Based on the standard ultrasound text features, the similarity distance of ultrasound text features of patients with congenital heart disease was calculated;

[0012] (4) Based on the similarity distance of ultrasound text features, the preoperative and intraoperative information of CHD patients is integrated to calculate the similarity distance between CHD patients, obtain the similarity distance matrix between CHD patient groups, and map it to the two-dimensional coordinates of Euclidean space to achieve visualization and generate a CHD patient map;

[0013] (5) Based on the congenital heart disease patient map, a similarity algorithm is used to recommend similar patient groups for the target patient. The postoperative risk factors of the similar patient groups are displayed using visualization and statistical test methods, assisting clinical personnel in using the postoperative risk factors of the similar patient groups to conduct postoperative risk assessment on the target patient.

[0014] The present invention is based on patient similarity analysis, and realizes a clinical decision support service supported by big data with natural clinical interpretability by constructing a computable congenital heart disease distance calculation method and a patient similarity visualization analysis of perioperative characteristics.

[0015] In step (1), the constructed congenital heart disease ontology establishes a hierarchical relationship between concepts based on the anatomical structure of the heart and the different pathological changes, forming a directed acyclic graph; each graph node represents a disease type, and the graph nodes are connected through the "is_a" relationship, and the child nodes have a more precise definition than the parent nodes.

[0016] Preferably, the congenital heart disease ontology also includes various annotations of the disease; the annotation content includes the International Classification of Diseases code (ICD-11) and the International Pediatric and Congenital Heart Code (IPCCC) of the disease.

[0017] Step (2) comprises:

[0018] (2-1) Correct incorrect punctuation and format in ultrasound texts, and replace synonyms and antonyms uniformly;

[0019] (2-2) The feature automatic extraction method based on rules and dictionaries extracts indicator features and diagnostic features from the preprocessed ultrasound text;

[0020] (2-3) Map the extracted indicator features and diagnostic features to the standard terms in the congenital heart disease ontology to obtain standard ultrasound text features.

[0021] The indicator characteristics describe the cardiac anatomical structure, the size of the abnormal structure, and the hemodynamic properties, including quantitative measurement indicators of the heart and blood vessels, indicators of cardiac systolic and diastolic function, the degree of valvular regurgitation, the blood flow velocity at the valvular orifice, the size of the defect, and the hemodynamic characteristics; the diagnostic characteristics refer to the judgment of the specific disease type in the report.

[0022] The methods for extracting indicator features include:

[0023] (a) Segment the ultrasound report text into individual sentences and recognize the individual sentences one by one; when an integer or decimal is recognized in the individual sentence, if there is a unit after it, proceed to the next step; if there is no unit after it, skip it;

[0024] (b) Detect whether there are descriptive words and structural nouns before integers or decimals. If there are descriptive words and structural nouns, it is considered that there are valid indicators in the sentence and recorded as indicator features; if there are no descriptive words and structural nouns, they are ignored.

[0025] The extraction methods of diagnostic features include:

[0026] (A) Analyze the diagnostic part of historical ultrasound reports, abstract and summarize the "structural noun + abnormal description" pattern, and establish a dictionary;

[0027] (B) The ultrasound report text is segmented into individual sentences, and the individual sentences are recognized sentence by sentence; when a pattern in the dictionary is recognized in the text, it is extracted and mapped to the standard term in the congenital heart disease ontology and recorded as a diagnostic feature.

[0028] Step (3) includes:

[0029] (3-1a) Preprocessing indicator features;

[0030] (3-1b) The index features are divided into several dimensions according to the anatomical structure and pathological characteristics of the heart; for the index features of each dimension, the Camberra distance of the index features of the two patients is calculated according to formula (a):

[0031]

[0032] Among them, x i and i Respectively represent the measurement values ​​of the same indicator reported by two patients in ultrasound reports; n represents the number of indicators in this dimension;

[0033] (3-1c) dist indicator (x, y) is normalized, and the indicator characteristics of all dimensions are averaged to obtain the indicator characteristic distance;

[0034] (3-2a) The diagnostic feature corresponds to the diagnostic node in the congenital heart disease ontology. The distance between two diagnostic nodes is calculated according to formula (b):

[0035]

[0036] Among them, c i and c j Represent two diagnostic nodes respectively; c LCS Represents c i and c j The minimum common ancestor of the diagnostic node c; depth(c) represents the depth of the diagnostic node c, that is, the number of nodes in the shortest path from the diagnostic node c to the root node;

[0037] (3-2b) Since a patient may have multiple diagnoses, the diagnostic feature distance between patients is calculated as the mean of the maximum distances of bidirectional single diagnoses, calculated using formula (c):

[0038]

[0039] Among them, A and B represent the diagnosis vectors of two patients respectively, A=[a 1 , a 2 , ..., a m ],B=[b 1 , b 2 , ..., b n ], dist diag (a i , b j ) is the distance between diagnosis nodes;

[0040] (3-3) Similarity distance dist of ultrasonic text features us is the indicator feature distance dist indicator and diagnostic feature distance dist Diag The weighted sum of is calculated by formula (d):

[0041] dist us =dist indicator × 1 +dist Diag × 2 (d);

[0042] Among them, w 1 、w 2 is a constant.

[0043] Preferably, w 1 and w 2 The values ​​are 0.5 and 1 respectively.

[0044] Preferably, in step (3-1a), preprocessing the indicator features includes:

[0045] Statistical means of indicators describing the defect or effusion type of disease;

[0046] If a defect type or effusion type disease is diagnosed in a certain ultrasound report, and the corresponding indicator is empty, the indicator will be assigned the mentioned mean value; if there is no defect type or effusion type disease, the corresponding indicator will be assigned 0.

[0047] Preferably, in step (3-1b), the indicator characteristics are divided into 10 dimensions, including:

[0048] ① Left ventricular type: left ventricular internal diameter, ejection fraction, interventricular septum thickness, left ventricular posterior wall thickness;

[0049] ②Left atrial type: left atrial diameter;

[0050] ③ Defect category: ventricular septal defect, atrial septal defect, patent ductus arteriosus, main pulmonary artery window defect size, blood flow velocity, pressure difference;

[0051] ④ Effusion: no echo dark areas in the apex, heart base, left ventricular anterior wall, etc.;

[0052] ⑤ Semilunar valves: pulmonary valve, aortic valve inner diameter, blood flow velocity, pressure difference;

[0053] ⑥ Atrioventricular valves: mitral and tricuspid valve inner diameters, blood flow velocity, and pressure difference;

[0054] ⑦ Aorta: inner diameter of aorta and its branches, blood flow velocity, pressure difference, aortic riding rate;

[0055] ⑧Coronary artery: inner diameter of coronary artery and its branches;

[0056] ⑨ Pulmonary artery: pulmonary artery pressure, inner diameter of left and right pulmonary arteries, flow velocity, and pressure difference;

[0057] ⑩Ventricular outflow tract category: inner diameter, flow velocity, and pressure difference of left and right ventricular outflow tracts.

[0058] In step (4), the similarity distance dist between patients with congenital heart disease p is the similarity distance dist of ultrasonic text features us , preoperative information distance dist pre , intraoperative information distance dist intra The mean of is calculated according to formula (e):

[0059]

[0060] Preoperative information distance dist pre is the Euclidean distance of preoperative information; the intraoperative information distance dist intra is the Euclidean distance of intraoperative information.

[0061] The preoperative information of the congenital heart disease patients includes gender, age, height, weight, and blood oxygen saturation of the right upper limb; the intraoperative information includes operation time, aortic clamping time, and extracorporeal circulation time.

[0062] Preferably, in step (4), the congenital heart disease patient map has the functions of zooming, panning, querying, searching, hiding and displaying specific diseases.

[0063] The congenital heart disease map has a patient search function: select a case in the congenital heart disease patient map or input new ultrasound report, preoperative and intraoperative information, and the congenital heart disease patient map can recommend a group of similar patient groups.

[0064] The congenital heart disease patient map has at least two adjustment parameters k and a threshold value, which are respectively used to adjust and screen out the first k most similar cases and cases whose similarity distance is within the threshold range.

[0065] In step (5), a variety of statistical charts are used to intuitively describe a variety of postoperative risk factors; the postoperative risk factors include length of hospital stay, length of assisted ventilation, length of stay in the intensive care unit, whether there are complications, type of complications, and survival.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] 1) Compared with RACHS-1 and Aristotle score, the present invention not only makes full use of the patient's preoperative and intraoperative information, especially free text data such as ultrasound reports, to describe the patient's condition more accurately from multiple dimensions, but also can recommend similar patients to assist clinicians in making decisions.

[0068] 2) Compared with artificial intelligence technology based on machine learning, the patient similarity algorithm we proposed has better interpretability, and at the same time uses hypothesis testing methods to statistically demonstrate its significance, making it easier to be clinically accepted.

[0069] 3) The visualized postoperative risk assessment method we proposed can intuitively describe multiple postoperative risk factors of patients, effectively assess the physiological and pathological conditions of patients at an early stage, strengthen perioperative monitoring and management, and provide early warning for high-risk patients after surgery. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 A flowchart of a method for visually assessing the postoperative risk of congenital heart disease based on patient similarity;

[0071] Figure 2 Schematic diagram of the process of automatic extraction of ultrasound text features. DETAILED DESCRIPTION

[0072] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0073] The present invention constructs a congenital heart disease ontology from the anatomical structure level through congenital heart disease clinical guidelines, encyclopedia knowledge, scientific literature, expert experience and other knowledge; proposes a rule- and dictionary-based feature automatic extraction method to extract multidimensional features of the heart description from ultrasound reports and map them into the ontology; proposes an ultrasound text similarity distance algorithm based on ultrasound text features; integrates the patient's preoperative and intraoperative information, calculates the patient's similarity distance, and transforms it into a congenital heart disease patient map; uses a similarity algorithm to recommend similar patient groups, evaluates the current patient's postoperative risk, and performs visual comparative statistical analysis.

[0074] The process of the visual assessment method for postoperative risk of congenital heart disease based on patient similarity is as follows Figure 1 As shown, the following steps are included:

[0075] (1) Construction of mental illness ontology

[0076] Ontology is a formalized expression of concepts and their relationships in a specific field. We consulted heterogeneous disease knowledge from multiple sources, such as clinical guidelines for congenital heart disease, encyclopedic knowledge, scientific literature, and expert experience, and constructed a congenital heart disease ontology from the anatomical structure level, which includes 436 types of congenital heart disease, each with a unique code. Congenital heart disease types are connected by the "is_a" relationship, and subtypes have more precise definitions than parent types. The congenital heart disease ontology also provides a variety of annotations for disease types, including Chinese and English terms, ICD-11, and IPCCC codes.

[0077] (2) Ultrasonic text feature extraction

[0078] We proposed a rule-based and dictionary-based feature automatic extraction method to extract two types of features from ultrasound text, namely indicator features and diagnostic features. Before feature extraction, the text is preprocessed to correct incorrect punctuation and format in the text, and replace common synonyms and antonyms. For example, "left atrioventricular valve" and "mitral valve" both refer to the same anatomical structure and are uniformly replaced with "mitral valve".

[0079] The indicator features describe the cardiac anatomical structure, the size of the abnormal structure, and the hemodynamic properties (peak flow rate, pressure, pressure difference). In the text, they are usually expressed as a sentence structure of "structural noun + descriptive word (diameter, width, echo interruption, flow rate, etc.) + integer / decimal + unit (CM, M / S, mmHg)". For example, "atrial septum echo interruption Φ0.35CM" means that there is a defect in the atrial septum, and the size of the defect is 0.35CM. We analyzed and counted 370,000 ultrasound reports and established a dictionary containing 83 commonly used indicators. The specific extraction process is as follows: Figure 2As shown in the figure, the whole text is divided into single sentences. When an integer or decimal is identified in a single sentence, if there is a unit (CM, MM, M / S, CM / S, mmHg) after it, the next step is entered. If there is no unit, it means that the number is not an ultrasound measurement indicator, and it may be an irrelevant number such as a paragraph number, so it is skipped; detect whether there are descriptive words and structural nouns before the number. If there are, it is considered that there is a valid indicator in the sentence and recorded as an indicator feature. If there are no descriptive words and structural nouns, for example, "The echo interruption of the atrial septum is Φ0.4CM, located at the secundum foramen, and the distance between the atrial septum and the superior vena cava on the subcostal section is 1.4CM", the two numbers "0.4" and "1.4" in the sentence are followed by the unit "CM", "0.4" is preceded by the structural noun "atrial septum" and the descriptive word "echo interruption", and "1.4" is preceded by the structural noun "superior vena cava" and no descriptive word, so "0.4" is a valid indicator and recorded, and "1.4" is an irrelevant measurement value and ignored.

[0080] Ultrasound reports often have different descriptions for the same type of disease, such as "ventricular septal defect" and "interrupted ventricular septal echo", "patent ductus arteriosus" and "a tube connecting the descending aorta and the pulmonary artery". We analyzed the diagnosis section of 370,000 ultrasound reports, abstracted and summarized the pattern of "structural noun + abnormal description", and established a dictionary. When a pattern in the dictionary is identified in the text, it is extracted and mapped to a standard term in the CHD ontology. For example, "moderate tricuspid regurgitation" is mapped to "tricuspid regurgitation, moderate".

[0081] (3) Ultrasonic text feature distance calculation

[0082] Before calculating the feature distance, in order to reduce errors, the extracted text features are preprocessed: the mean of the indicators describing defect-type or effusion-type diseases (ventricular septal defect size, echo-free dark area size, etc.) is statistically calculated. If the ultrasound report diagnosis contains defect-type or effusion-type diseases (ventricular septal defect, pericardial effusion, etc.), and the corresponding indicators are empty, the indicators are assigned the mean value; if these diseases do not exist, the corresponding indicators are assigned 0.

[0083] Ultrasound text feature distance calculation is divided into two parts: indicator feature distance calculation and diagnostic feature distance calculation.

[0084] Regarding the characteristics of indicators, when there are different types of abnormalities in the heart, doctors often use indicators of different types in different parts to describe the abnormalities. For example, when there is a defect-type abnormality in the heart (ventricular septal defect, atrial septal defect, patent ductus arteriosus, aortopulmonary window), doctors often measure the size of the defect, blood flow rate, and pressure difference; when the atrioventricular valve of the heart has abnormalities such as stenosis, regurgitation, and dysplasia, doctors often measure the diameter of the atrioventricular valve, blood flow rate, and pressure difference. Therefore, we divide the heart into 10 dimensions based on its anatomical structure and pathological characteristics, namely (1) left ventricular type: left ventricular internal diameter, ejection fraction, ventricular septum thickness, and left ventricular posterior wall thickness; (2) left atrium type: left atrial diameter; (3) defect type: size of ventricular septal defect, atrial septal defect, patent ductus arteriosus, and aortopulmonary window defect, blood flow rate, and pressure difference; (4) effusion type: echo-free dark areas in the apex, base of the heart, and left ventricular anterior wall; (5) semilunar valve type: pulmonary Aortic valve, aortic valve inner diameter, blood flow velocity, pressure difference; (6) Atrioventricular valve: mitral and tricuspid valve inner diameter, blood flow velocity, pressure difference; (7) Aorta: aorta and branch inner diameter, blood flow velocity, pressure difference, aortic riding rate; (8) Coronary artery: coronary artery and branch inner diameter; (9) Pulmonary artery: pulmonary artery pressure, left and right pulmonary artery inner diameter, flow velocity, pressure difference; (10) Ventricular outflow tract: left and right ventricular outflow tract inner diameter, flow velocity, pressure difference. Each dimension contains several indicators. For each dimension feature, calculate its Camberra distance:

[0085]

[0086] Among them, x i and i They represent the same indicator in two ultrasound reports, and n represents the number of indicators in this dimension. The characteristic distance calculated for each dimension is normalized, and the mean of the 10 dimensions is calculated to obtain the indicator characteristic distance.

[0087] For the diagnosis feature, each diagnosis corresponds to a node in the congenital heart disease ontology, and the calculation of a single diagnosis distance uses the depth of the corresponding node in the congenital heart disease ontology:

[0088]

[0089] Among them, c i and c j Represent two diagnostic nodes, c LCS Represents c i and c j depth(c) represents the depth of node c, that is, the number of nodes in the shortest path from the node to the root node.

[0090] Since a patient may have multiple diagnoses, the inter-patient diagnostic feature distance was calculated as the mean of the maximum distances of the two-way single diagnoses.

[0091]

[0092] Among them, A and B represent the diagnosis vectors of two patients respectively, A=[a 1 , a 2 , ..., a m ],B=[b 1 , b 2 , ..., b n ], dist diag (a i , b j ) is the distance between diagnosis nodes.

[0093] The final similarity distance of ultrasound text features is the weighted sum of the indicator feature distance and the diagnostic feature distance, as shown in the following formula:

[0094] dist us =dist indicator × 1 +dist Diag × 2

[0095] Based on our multiple experiments, w 1 and w 2 The best empirical effect is achieved when the values ​​are 0.5 and 1 respectively.

[0096] (4) Patient similarity distance calculation and congenital heart disease patient map generation

[0097] In order to make full use of the patient's medical data, we integrated the patient's preoperative information (gender, age, height, weight, right upper limb blood oxygen saturation) and intraoperative information (operation time, aortic clamping time, extracorporeal circulation time) into the algorithm and calculated their Euclidean distances dist pre 、dist intra .

[0098] The final patient similarity distance is calculated as the mean of ultrasound text feature distance, preoperative information distance, and intraoperative information distance, that is:

[0099]

[0100] We calculated the distances between 4778 CHD cases and obtained a distance matrix of CHD patients. We used the t-SNE dimensionality reduction method to transform this matrix into a two-dimensional map of CHD patients. The map provides the functions of zooming, panning, querying, searching, hiding, and displaying specific diseases. Limiting the query conditions can filter out all cases that meet the conditions. Selecting or entering a new case in the map can search for similar groups. At the same time, the map also integrates rich physiological and pathological information to help doctors discover the relationship between different types and conditions of CHD.

[0101] (5) Postoperative risk assessment of patients

[0102] Based on the similarity distance algorithm mentioned above, we designed an interactive interface to recommend similar patients to assist doctors in making decisions. We set two parameters, k and threshold. By adjusting these two parameters, we can screen out the first k most similar cases or cases with similarity distances within a certain range. In order to illustrate its statistical significance, the Mann-Whitney U test and chi-square test were used to perform significance tests on postoperative risk factors of similar and non-similar groups, and the postoperative risk factors of similar groups were used to assess the postoperative risk of the current patient.

[0103] The present invention uses a variety of statistical charts to intuitively describe a variety of postoperative risk factors, including hospital stay, assisted ventilation, ICU stay, whether there are complications, complication types, and survival. In practical applications, ultrasound reports and other information are entered before or just after surgery. The congenital heart disease patient map recommends a group of similar patients to doctors, gives postoperative risk factors for similar groups, and assesses the current patient's postoperative risk. It can effectively assess the patient's physiological and pathological conditions at an early stage, strengthen perioperative monitoring and management, and provide early warnings for high-risk patients after surgery.

[0104] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity, characterized in that: The following steps are involved: (1) Based on multi-source heterogeneous disease knowledge, a congenital heart disease ontology is constructed from the anatomical structure level; (2) Multidimensional feature extraction was performed on ultrasound report texts of historical CHD patients and mapped to standard terms in the CHD ontology to obtain standard ultrasound text features; (3) Based on standard ultrasound text features, calculate the similarity distance of ultrasound text features of patients with congenital heart disease; including: (3-1a) Preprocessing indicator features; (3-1b) The index features are divided into several dimensions according to the anatomical structure and pathological characteristics of the heart; for the index features of each dimension, the Camberra distance of the index features of the two patients is calculated according to formula (a): Among them, x i and i Respectively represent the measurement values ​​of the same indicator reported by two patients in ultrasound reports; n represents the number of indicators in this dimension; (3-1c) dist indicator (x, y) is normalized, and the indicator characteristics of all dimensions are averaged to obtain the indicator characteristic distance; (3-2a) The diagnostic feature corresponds to the diagnostic node in the congenital heart disease ontology. The distance between two diagnostic nodes is calculated according to formula (b): Among them, c i and c j Represent two diagnostic nodes respectively; c LCS Represents c i and c j The minimum common ancestor of the diagnostic node c; depth(c) represents the depth of the diagnostic node c, that is, the number of nodes in the shortest path from the diagnostic node c to the root node; (3-2b) Since a patient may have multiple diagnoses, the diagnostic feature distance between patients is calculated as the mean of the maximum distances of bidirectional single diagnoses, calculated using formula (c): Among them, A and B represent the diagnosis vectors of two patients respectively, A=[a1,a2,…,a m ],B=[b1,b2,…,b n ], dist diag (a i ,b j ) is the distance between diagnosis nodes; (3-3) Similarity distance dist of ultrasonic text features us is the indicator feature distance dist indicator and diagnostic feature distance dist Diag The weighted sum of is calculated by formula (d): dist us =dist indicator ×w1+dist Diag ×w2 (d); Among them, w1 and w2 are constants; (4) Based on the similarity distance of ultrasound text features, the preoperative and intraoperative information of CHD patients is integrated to calculate the similarity distance between CHD patients, obtain the similarity distance matrix between CHD patient groups, and map it to the two-dimensional coordinates of Euclidean space to achieve visualization and generate a CHD patient map; the similarity distance between CHD patients is dist p is the similarity distance dist of ultrasonic text features us , preoperative information distance dist pre , intraoperative information distance dist intra The mean of is calculated according to formula (e): (5) Based on the congenital heart disease patient map, a similarity algorithm is used to recommend similar patient groups for the target patient. The postoperative risk factors of the similar patient groups are displayed using visualization and statistical test methods, assisting clinical personnel in using the postoperative risk factors of the similar patient groups to conduct postoperative risk assessment on the target patient.

2. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 1, characterized in that: Step (2) comprises: (2-1) Correct incorrect punctuation and format in ultrasound texts, and replace synonyms and antonyms uniformly; (2-2) The feature automatic extraction method based on rules and dictionaries extracts indicator features and diagnostic features from the preprocessed ultrasound text; (2-3) Map the extracted indicator features and diagnostic features to the standard terms in the congenital heart disease ontology to obtain standard ultrasound text features.

3. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 2, characterized in that: The methods for extracting indicator features include: (a) Segment the ultrasound report text into individual sentences and recognize the individual sentences one by one; when an integer or decimal is recognized in the individual sentence, if there is a unit after it, proceed to the next step; if there is no unit after it, skip it; (b) Detect whether there are descriptive words and structural nouns before integers or decimals. If there are descriptive words and structural nouns, it is considered that there are valid indicators in the sentence and recorded as indicator features; if there are no descriptive words and structural nouns, they are ignored.

4. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 2, characterized in that: The extraction methods of diagnostic features include: (A) Analyze the diagnostic part of historical ultrasound reports, abstract and summarize the "structural noun + abnormal description" pattern, and establish a dictionary; (B) The ultrasound report text is segmented into individual sentences, and the individual sentences are recognized sentence by sentence; when a pattern in the dictionary is recognized in the text, it is extracted and mapped to the standard term in the congenital heart disease ontology and recorded as a diagnostic feature.

5. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 1, characterized in that: In step (3-1a), preprocessing the indicator features includes: Statistical means of indicators describing the defect or effusion type of disease; If a defect type or effusion type disease is diagnosed in a certain ultrasound report, and the corresponding indicator is empty, the indicator will be assigned the mentioned mean value; if there is no defect type or effusion type disease, the corresponding indicator will be assigned 0.

6. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 1, characterized in that: In step (3-1b), the indicator characteristics are divided into 10 dimensions, including: ① Left ventricular type: left ventricular internal diameter, ejection fraction, interventricular septum thickness, left ventricular posterior wall thickness; ②Left atrial type: left atrial diameter; ③ Defect category: ventricular septal defect, atrial septal defect, patent ductus arteriosus, main pulmonary artery window defect size, blood flow velocity, pressure difference; ④ Effusion: no echo dark area at the apex, bottom of the heart, and left ventricular anterior wall; ⑤ Semilunar valves: pulmonary valve, aortic valve inner diameter, blood flow velocity, pressure difference; ⑥ Atrioventricular valves: mitral and tricuspid valve inner diameters, blood flow velocity, and pressure difference; ⑦ Aorta: inner diameter of aorta and its branches, blood flow velocity, pressure difference, aortic riding rate; ⑧Coronary artery: inner diameter of coronary artery and its branches; ⑨ Pulmonary artery: pulmonary artery pressure, inner diameter of left and right pulmonary arteries, flow velocity, and pressure difference; ⑩Ventricular outflow tract category: inner diameter, flow velocity, and pressure difference of left and right ventricular outflow tracts.

7. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 1, characterized in that: The congenital heart disease map has a patient search function: select a case in the congenital heart disease patient map or input new ultrasound report, preoperative and intraoperative information, and the congenital heart disease patient map can recommend a group of similar patient groups.

8. The method for visualizing the risk assessment of postoperative congenital heart disease based on patient similarity according to claim 1, characterized in that: In step (5), a variety of statistical charts are used to intuitively describe a variety of postoperative risk factors; the postoperative risk factors include length of hospital stay, length of assisted ventilation, length of stay in the intensive care unit, whether there are complications, type of complications, and survival.

Images