Construction method of central vertigo diagnosis model and diagnosis system

Abstract

The invention provides a construction method of a central vertigo diagnosis model and a diagnosis system. The construction method of the central vertigo diagnosis model comprises the following steps: S1, acquiring data of dizziness patients meeting a holding and discharging standard, and randomly dividing the data into training group data and verification group data according to a certain proportion; s2, screening risk factors related to central vertigo by adopting a minimum absolute contraction and selection operator regression method based on the training group data; s3, on the basis of the risk factors related to the central vertigo, constructing different models by using multiple machine learning algorithms and training the models; and S4, based on the verification group data, performing performance verification on each constructed model, and selecting the model with the best comprehensive performance as the central vertigo diagnosis model. According to the method, the central vertigo can be simply, conveniently, quickly and accurately identified, and early diagnosis is provided in clinical scenes such as emergency departments and primary hospitals.

Description

Technical Field

[0001] The present invention relates to the technical field of healthcare informatics, and particularly to the technical field of clinical data analysis and processing. Background Art

[0002] Vertigo is a common symptom among neurological patients and seriously affects the quality of life of patients. Its prevalence in the general population is 15% to 30%, and in people over 85 years old, the prevalence is close to 50%. [1-3] Since the causes of vertigo are diverse, including vestibular, neurological, psychogenic factors, and systemic diseases of various systems in the body may also be accompanied by vertigo symptoms, diagnosing vertigo faces great challenges. Benign paroxysmal positional vertigo (BPPV), acute unilateral vestibulopathy, vestibular migraine, and Meniere's disease are the most common types of benign peripheral vertigo. [4] It is worth noting that approximately 25% of acute vertigo is associated with life-threatening central nervous system diseases, among which stroke accounts for 4% - 15%. [5, 6] In the early stage of some central vertigo, due to the non-specific symptoms, it is often misdiagnosed as peripheral vertigo, which may lead to delayed treatment and poor prognosis. If the screening is carried out without discrimination, patients may receive unnecessary examinations, wasting medical resources and increasing medical risks. Therefore, accurately differentiating central vertigo from peripheral vertigo is crucial.

[0003] Currently, a variety of assessment tools have been developed to assist in the diagnosis of vertigo. Standardized tools such as TriAGe+ and CATCH2 scores help identify patient characteristics related to the diagnostic process. [7, 8] However, due to the overlapping symptoms of central and peripheral vertigo, symptoms alone are not sufficient to distinguish these two types of vertigo. To accurately distinguish, comprehensive neurological, vestibular, oculomotor, and postural examinations must be carried out. [9] Technologies such as head impulse testing and fixation suppression of nystagmus recorded by video oculography (VOG) have been proven effective in predicting vestibular stroke.

[10] However, the application of these diagnostic methods requires neurology experts or otolaryngology experts and is restricted by specific equipment, making it difficult to widely promote these methods in the emergency diagnosis environment of acute vertigo attacks and primary care. In addition, due to the diversity of the causes of vertigo, acute vertigo also needs to be evaluated in combination with laboratory tests and physiological indicators.

[0004] Although some progress has been made in diagnostic models based on deep learning and questionnaires, these tools are still difficult to identify which types of vertigo are related to more serious conditions. [11, 12] Machine learning algorithms, especially those using large datasets, are being increasingly used in the diagnosis of vertigo. Multicenter studies have shown that the Light Gradient Boosting Machine (LightGBM) based on questionnaires can effectively predict vestibular diseases.

[13] Although these machine learning methods can identify the top ten factors most relevant to dizziness, it remains unclear how to utilize these factors for diagnosis. Currently, there is a lack of practical tools that can effectively distinguish central vertigo from peripheral vertigo, which also complicates the achievement of precise treatment. Emergency physicians globally have listed the improvement of dizziness diagnostic tools as a priority development direction.

[14] Therefore, there is an urgent need for an evidence-based, cost-effective, simple, and feasible acute dizziness diagnosis method that can assist doctors in quickly and effectively identifying central vertigo in the emergency setting or primary healthcare institutions, reducing missed and misdiagnoses.

[0005] Therefore, this study aims to develop and validate a simple, practical, and reliable diagnostic model for central vertigo. Using multi-center data, combining symptoms, objective blood markers, and physiological indicators, logistic regression analysis and multiple machine learning algorithms were employed to select the optimal model to assist in clinical decision-making.

[0006] References:

[0007] 1. Kerber KA, Callaghan BC, Telian SA, et al. Dizziness symptom type prevalence and overlap: a US nationally representative survey. Am J Med. 2017;130(12):1465.

[0008] 2. Karatas M. Central vertigo and dizziness: epidemiology, differential diagnosis, and common causes. Neurologist. 2008;14(6):355-64.

[0009] 3. Balatsouras DG, Koukoutsis G, Fassolis A, et al. Benign paroxysmal positional vertigo in the elderly: current insights. Clin Interv Aging. 2018;13:2251-66.

[0010] 4. Spiegel R, Kirsch M, Rosin C, et al. Dizziness in the emergency department: an update on diagnosis. Swiss Med Wkly. 2017;147:w14565.

[0011] 5. Newman-Toker DE, Hsieh YH, Camargo CA Jr, et al. Spectrum of dizziness visits to US emergency departments: cross-sectional analysis from a nationally representative sample. Mayo Clin Proc. 2008; 83:765–775.

[0012] 6. Royl G, Ploner CJ, Leithner C. Dizziness in the emergency room: diagnoses and misdiagnoses. Eur Neurol. 2011; 66:256–263.

[0013] 7. Kuroda R, Nakada T, Ojima T, et al. The TriAGe+score for vertigo or dizziness: a diagnostic model for stroke in the emergency department. J Stroke Cerebrovasc Dis. 2017; 26:1144–1153.

[0014] 8. Zwergal A, Mo¨hwald K, Hadzhikolev H, et al. Development of a diagnostic index test for stroke as a cause of vertigo, dizziness and imbalance in the emergency room: first results from the prospective EMVERT trial. Clin Neurophysiol. 2018; 129:e54.

[0015] 9. Kattah JC, Talkad AV, Wang DZ, et al. HINTS to diagnose stroke in the acute vestibular syndrome: three-step bedside oculomotor examination more sensitive than early MRI diffusion-weighted imaging. Stroke. 2009;40:3504–3510.

[0016] 10. Georgios Mantokoudis, Thomas Wyss, Ewa Zamaro, et al. Stroke Prediction Based on the Spontaneous Nystagmus Suppression Test in Dizzy Patients: A Diagnostic Accuracy Study. Neurology. 2021;97(1):e42-e51.

[0017] 11. Hyo-Jung Kim, Jeong-Mi Song, Liqun Zhong, et al. Questionnaire-based diagnosis of benign paroxysmal positional vertigo. Neurology. 2020;94(9):e942-e949.

[0018] 12. Peixia Wu, Xuebing Liu, Qi Dai, et al. Diagnosing the benign paroxysmal positional vertigo via 1D and deep-learning composite model. J Neurol. 2023;270(8):3800-3809.

[0019] 13.Fangzhou Yu, Peixia Wu, Haowen Deng, et al. A Questionnaire-Based Ensemble Learning Model to Predict the Diagnosis of Vertigo: Model Development and Validation Study. J Med Internet Res. 2022;24(8):e34126.

[0020] 14.Eagles D, Stiell IG, Clement CM, et al. International survey of emergency physicians' priorities for clinical decision rules. Acad Emerg Med. 2008; 15:177–182. Summary of the Invention

[0021] The first aspect of the present application provides a method for constructing a central vertigo diagnosis model, which includes:

[0022] Step S1: Obtain the data of dizziness patients who meet the inclusion and exclusion criteria, and randomly divide them into training group data and validation group data according to a certain proportion;

[0023] Step S2: Based on the training group data, use the least absolute shrinkage and selection operator regression method to screen out the risk factors related to central vertigo;

[0024] Step S3: Based on the risk factors related to central vertigo, use a variety of machine learning algorithms to construct different models and train the models;

[0025] Step S4: Based on the validation group data, verify the performance of each constructed model, and select the model with the best comprehensive performance as the central vertigo diagnosis model.

[0026] Further, in step S2, the risk factors related to central vertigo include: the correlation with body position or head position, visual rotation, hearing loss, head swelling or headache, gender, age, systolic blood pressure, and disease course (more than 3 days).

[0027] Further, in step S3, the machine learning algorithms include logistic regression, random forest, k-nearest neighbor, linear discriminant analysis (LDA), naive Bayes, neural network, quadratic discriminant analysis (QDA), and support vector machine (SVM).

[0028] Further, in step S4, the discrimination, calibration, and clinical application value of the model are internally and externally verified through the ROC curve, calibration curve, clinical decision curve, and clinical impact curve, respectively. The model with the best comprehensive performance is the logistic regression model, and the logistic regression model is presented through a diagnostic formula and / or nomogram.

[0029] Further, the diagnostic formula for predicting the probability of central vertigo formation in the current patient is:

[0030]

[0031] where P is the probability value of predicting the formation of central vertigo in the current patient;

[0032] The assignments of the various risk factors in the diagnostic formula are shown in the following table:

[0033]

[0035] The second aspect of the present application provides a central vertigo diagnosis system, which includes:

[0036] A data input module, which is used to obtain data on risk factors related to central vertigo;

[0037] A prediction module, which is used to predict the probability value of suffering from central vertigo according to a pre-constructed central vertigo diagnosis model;

[0038] An output module, which is used to output the probability value, diagnostic conclusion, and its corresponding clinical decision.

[0039] Further, the risk factors related to central vertigo obtained by the data input module include: the correlation with body position or head position, visual rotation, hearing loss, head swelling or headache, gender, age, systolic blood pressure, and disease duration (more than 3 days).

[0040] Further, the pre-constructed central vertigo diagnosis model is a logistic regression model, and the logistic regression model is presented through a diagnostic formula and / or nomogram.

[0041] Further, the diagnostic formula for predicting the probability of central vertigo formation in the current patient is:

[0042]

[0043] where P is the probability value of predicting the formation of central vertigo in the current patient;

[0044] The assignments of the various risk factors in the diagnostic formula are shown in the following table:

[0045]

[0047] The third aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method for constructing a central vertigo diagnosis model are implemented.

[0048] Combining all the above technical solutions, the advantages and positive effects of the present invention are as follows:

[0049] The present invention screens out eight key diagnostic indicators from the clinical data of patients, and selects the logistic regression model with the best comprehensive performance from a variety of machine learning algorithms as the final central vertigo diagnosis model. This model can provide an auxiliary diagnostic tool for clinicians. The method is simple, convenient and fast, especially in clinical scenarios such as emergency departments and primary hospitals, helping to identify central vertigo earlier and more accurately. The diagnostic model constructed by the present invention has significant advantages in improving the diagnostic discrimination and calibration of central vertigo. This model can provide an effective means for early diagnosis in clinical practice, avoid serious consequences caused by missed diagnosis and misdiagnosis, and has broad potential for promotion and application.

[0050] Specifically:

[0051] 1. At present, there is no relevant model for diagnosing central vertigo patients through common clinical indicators in clinical practice. Vertigo and dizziness are common symptoms that may be caused by a variety of peripheral and central reasons, so the diagnosis is challenging and time-consuming. Existing diagnostic techniques, such as eye movement assessment and cranial MRI scanning, cannot effectively achieve rapid diagnosis. Therefore, there is an urgent need to develop an efficient and practical diagnostic tool. The present invention will fill the industry gap and help in the early diagnosis of central vertigo in clinical scenarios such as emergency departments and primary hospitals.

[0052] 2. The present invention establishes a diagnostic model by comparing the clinical symptoms, laboratory tests and other indicators of patients with confirmed central vertigo and peripheral vertigo. The 8 indicators extracted, including the correlation of head position or body position, visual rotation, hearing loss, headache, gender, age, systolic blood pressure, and disease course (more than 3 days), are easy to obtain data, and it can be completed in about 1 minute. The method is simple, convenient and fast, especially suitable for primary medical institutions, providing a reference basis for better diagnosing central vertigo clinically.

[0053] 3. The present invention reports for the first time the screening of clinical indicators based on machine learning to establish a diagnostic model for central vertigo. This model has good diagnostic discrimination, calibration, and great clinical popularization value. It is of great clinical guiding significance for early identification of relevant risk factors of central vertigo in clinical practice, for auxiliary diagnosis before time-consuming and expensive examinations such as cranial MR plain scan are completed, and then for timely clinical intervention. The present invention also presents the model in the form of a nomogram or a computer system, making the model easier to be clinically promoted. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart for screening vertigo patients;

[0055] Figure 2 In FIG. A, it is the selection of the optimal parameter (lambda) in the Lasso model, using ten-fold cross-validation and the minimum standard criterion; in FIG. B, it is the Lasso coefficient curve of the features;

[0056] Figure 3 It is the calibration curve of various models in the internal validation group;

[0057] Figure 4 It is the decision curve of various models in the internal validation group for diagnosing central vertigo;

[0058] Figure 5 It is the calibration curve of various models in the external validation group;

[0059] Figure 6 It is the decision curve of various models in the external validation group for diagnosing central vertigo;

[0060] Figure 7 It is the nomogram of the central vertigo diagnostic model of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The advantages of the present invention are further elaborated below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art should understand that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the protection scope of the present invention.

[0062] Example 1 Construction Method of Central Vertigo Diagnostic Model

[0063] This example provides a construction method of a central vertigo diagnostic model, including steps S1 - S4:

[0064] Step S1: Obtain the data of dizziness patients who meet the inclusion and exclusion criteria, and randomly divide them into training group data and validation group data according to a certain proportion.

[0065] 1. Research method:

[0066] 1.1 Research object

[0067] As Figure 1 , in the initial screening, 515 vertigo patients from our hospital were enrolled between October 2017 and February 2022 as the training cohort and the internal validation cohort. In addition, in 2023, 108 patients were included from other hospitals as the external validation cohort. All participants were examined by a board-certified neurologist, including the patient's medical history, blood tests, and neuroimaging tests. All patients underwent brain magnetic resonance imaging (MRI), magnetic resonance angiography (MRA), or computed tomography angiography (CTA).

[0068] The inclusion criteria were as follows: age > 18 years; central vertigo was defined as restricted diffusion changes on brain MRI or any transient neurological deficit presumably caused by a lesion without evidence of acute infarction, but with > 50% occlusion or stenosis of the major intracranial vessels; peripheral vertigo was clearly diagnosed as benign paroxysmal positional vertigo (BPPV), vestibular neuritis, Meniere's disease, vestibular migraine according to the published criteria and laboratory test results, and had undergone video electronystagmography (VNG) positional tests, caloric stimulation tests, head impulse tests, etc.

[0069] The exclusion criteria included: cervical spondylosis; obstructive sleep apnea syndrome; primary aldosteronism; hypothyroidism; hypertension and Shy-Drager syndrome; having a history of severe physical illness or alcoholism; liver and kidney dysfunction; having a history of severe trauma or surgery within one year; refusing to sign the informed consent form.

[0070] The study protocol and informed consent form had been reviewed and approved by the local institutional research ethics committee (CHEC2023-101). All participants signed a written informed consent form before enrollment.

[0071] 1.2 Data collection

[0072] Relevant data of all included subjects were collected, including: demographic characteristics, medical history, clinical characteristics, serum indicators, and other indicators.

[0073] Demographic characteristics included gender, age, body mass index (BMI).

[0074] Medical history included smoking history, drinking history, hypertension, diabetes, hyperlipidemia, coronary heart disease or stroke, arrhythmia, cancer, etc.

[0075] Clinical characteristics included disease course, onset form, diet-related, position or head position-related, visual rotation, nausea / vomiting / cold sweat, chest tightness / palpitations, tinnitus / earplug sensation, hearing loss, headache, sense of imbalance, etc.

[0076] Serum indicators include routine blood tests (white blood cells, red blood cells, platelets, lymphocyte ratio, neutrophil ratio, hemoglobin, neutrophil count, lymphocyte count, granulocyte flow ratio, monocyte flow ratio, platelet / lymphocyte ratio, monocyte count, eosinophil count, basophil count, mean corpuscular hemoglobin content, mean corpuscular volume, mean platelet volume, red blood cell distribution width SD, platelet distribution width SD), biochemical tests (urea, alanine aminotransferase, aspartate aminotransferase, glucose, creatinine, uric acid, homocysteine, glomerular filtration rate, glycated hemoglobin, low-density lipoprotein, high-density lipoprotein, triglycerides, total cholesterol, apolipoprotein A1, apolipoprotein B, albumin, prealbumin), and coagulation function tests (plasma D-dimer, thrombin time, prothrombin time, fibrinogen, fibrin degradation products, activated partial thromboplastin time, international normalized ratio). Blood samples of all participants were collected within 8 hours after admission.

[0077] Other indicators: systolic blood pressure, diastolic blood pressure, electrocardiogram (normal, arrhythmia, myocardial ischemia, arrhythmia combined with myocardial ischemia).

[0078] 1.3 Statistical analysis

[0079] Continuous variables were expressed as mean (standard deviation, SD) or median (interquartile range, IQR). Categorical variables were expressed as number n (percentage %). For continuous variables that conform to normal distribution or do not conform to normal distribution, Student's t-test or Mann-Whitney U-test was used respectively, and Pearson chi-square test or Mann-Whitney U-test was used for categorical variables or ordinal variables. All risk factors were analyzed.

[0080] Patients were divided into a training cohort and a validation cohort during treatment, with a ratio of 7:3, and random grouping was performed using the R package "splitTools" (seed = 1217). Patients were randomly divided into a training cohort (n = 296) and an internal validation cohort (n = 126). The external validation cohort included 108 patients (see Figure 1 ). In the training cohort, 46.3% (137 / 296) of the patients were diagnosed with central vertigo. We collected basic demographic data, clinical symptoms, medical history, blood tests (including routine blood tests, biochemical tests, and coagulation function tests), and electrocardiograms.

[0081] There were significant differences in indicators such as gender, age, systolic blood pressure, disease course, onset form, related to body position or head position, visual rotation, nausea, tinnitus, hearing loss, headache, hypertension, diabetes, coronary heart disease or stroke, urea, glucose, creatinine, uric acid, homocysteine, renal tubular filtration rate, lymphocyte ratio, single flow ratio, and thrombin time, and all P < 0.05. No significant differences were found in other indicators (see Table 1).

[0082] Table 1 Baseline information in the training cohort

[0083] Items Classification Trainset Central vertigo (n = 137) Peripheral vertigo (n = 159) P value Statistical magnitude Gender Male 152(51.35) 87(63.50) 65(40.88) 0.0001 15.08 Female 144(48.65) 50(36.50) 94(59.12) Total 296(100.00) 137(100.00) 159(100.00) Age 66.43±11.35 69.82±11.10 63.50±10.75 <.0001 4.96 BMI 24.19±3.28 24.23±3.52 24.16±3.07 0.8632 0.17 Systolic pressure 132.33±19.12 136.56±21.81 128.69±15.63 0.0004 3.60 Diastolic pressure 79.79±10.68 80.74±11.36 78.98±10.04 0.1589 1.41 Course ≤3 days 88(29.73) 27(19.71) 61(38.36) 0.0005 12.26 >3 days 208(70.27) 110(80.29) 98(61.64) Total 296(100.00) 137(100.00) 159(100.00) Onset form Single 51(17.23) 14(10.22) 37(23.27) 0.0102 9.17 Recurrent 220(74.32) 109(79.56) 111(69.81) Persistent 25(8.45) 14(10.22) 11(6.92) Total 296(100.00) 137(100.00) 159(100.00) Related to eating Yes 3(1.01) 1(0.73) 2(1.26) 0.6512 0.20 No 293(98.99) 136(99.27) 157(98.74) Total 296(100.00) 137(100.00) 159(100.00) Related to body position or head position Yes 154(52.03) 42(30.66) 112(70.44) <.0001 46.67 No 142(47.97) 95(69.34) 47(29.56) Total 296(100.00) 137(100.00) 159(100.00) Visual rotation Yes 166(56.08) 49(35.77) 117(73.58) <.0001 42.73 No 130(43.92) 88(64.23) 42(26.42) Total 296(100.00) 137(100.00) 159(100.00) Nausea / vomiting / cold sweat Yes 169(57.09) 58(42.34) 111(69.81) <.0001 22.68 No 127(42.91) 79(57.66) 48(30.19) Total 296(100.00) 137(100.00) 159(100.00) Chest tightness / palpitations Yes 35(11.82) 17(12.41) 18(11.32) 0.7725 0.08 No 261(88.18) 120(87.59) 141(88.68) Total 296(100.00) 137(100.00) 159(100.00) Tinnitus / aural fullness Yes 52(17.57) 17(12.41) 35(22.01) 0.0304 4.69 No 244(82.43) 120(87.59) 124(77.99) Total 296(100.00) 137(100.00) 159(100.00) Hearing loss Yes 22(7.43) 3(2.19) 19(11.95) 0.0014 10.19 No 274(92.57) 134(97.81) 140(88.05) Total 296(100.00) 137(100.00) 159(100.00) Headache Yes 26(8.78) 17(12.41) 9(5.66) 0.0408 4.18 No 270(91.22) 120(87.59) 150(94.34) Total 296(100.00) 137(100.00) 159(100.00) Sense of imbalance Yes 89(30.07) 48(35.04) 41(25.79) 0.0835 2.99 No 207(69.93) 89(64.96) 118(74.21) Total 296(100.00) 137(100.00) 159(100.00) Smoking Yes 65(21.96) 35(25.55) 30(18.87) 0.1663 1.92 No 231(78.04) 102(74.45) 129(81.13) Total 296(100.00) 137(100.00) 159(100.00) Drinking Yes 27(9.12) 9(6.57) 18(11.32) 0.1569 2.00 No 269(90.88) 128(93.43) 141(88.68) Total 296(100.00) 137(100.00) 159(100.00) Hypertension Yes 167(56.42) 88(64.23) 79(49.69) 0.0118 6.33 No 129(43.58) 49(35.77) 80(50.31) Total 296(100.00) 137(100.00) 159(100.00) Diabetes Yes 54(18.24) 32(23.36) 22(13.84) 0.0344 4.47 No 242(81.76) 105(76.64) 137(86.16) Total 296(100.00) 137(100.00) 159(100.00) Hyperlipidemia Yes 39(13.18) 13(9.49) 26(16.35) 0.0817 3.03 No 257(86.82) 124(90.51) 133(83.65) Total 296(100.00) 137(100.00) 159(100.00) Coronary heart disease or stroke Yes 49(16.55) 32(23.36) 17(10.69) 0.0035 8.55 No 247(83.45) 105(76.64) 142(89.31) Total 296(100.00) 137(100.00) 159(100.00) Arrhythmia Yes 28(9.46) 9(6.57) 19(11.95) 0.1148 2.49 No 268(90.54) 128(93.43) 140(88.05) Total 296(100.00) 137(100.00) 159(100.00) Cancer Yes 15(5.07) 6(4.38) 9(5.66) 0.6164 0.25 No 281(94.93) 131(95.62) 150(94.34) Total 296(100.00) 137(100.00) 159(100.00) Urea 5.54±1.59 5.85±1.68 5.27±1.47 0.0017 3.17 Alanine aminotransferase 23.97±17.25 24.62±18.72 23.40±15.91 0.5456 0.60 Aspartate aminotransferase 20.16±9.04 20.57±8.63 19.80±9.39 0.4655 0.73 Glucose 5.96±1.51 6.29±1.81 5.67±1.12 0.0004 3.58 Creatinine 73.84±19.24 78.88±21.14 69.50±16.30 <.0001 4.30 Uric acid 109.14±158.00 67.42±137.69 145.10±165.76 <.0001 -4.34 Homocysteine 13.57±16.82 15.84±21.20 11.61±11.53 0.0306 2.17 Glomerular filtration rate 89.42±20.91 85.80±22.74 92.54±18.71 0.0055 -2.80 Glycosylated hemoglobin 6.12±1.26 6.21±1.25 6.04±1.26 0.2337 1.19 Low - density lipoprotein 2.79±0.87 2.71±0.85 2.85±0.89 0.1537 -1.43 High - density lipoprotein 1.26±0.34 1.27±0.39 1.26±0.30 0.9457 0.07 Triglyceride 1.63±0.97 1.56±0.85 1.69±1.06 0.2633 -1.12 Total cholesterol 4.70±1.21 4.59±1.00 4.79±1.37 0.1462 -1.46 Leukocyte 6.31±1.94 6.39±2.08 6.23±1.81 0.4929 0.69 Erythrocyte 4.49±0.56 4.52±0.60 4.46±0.53 0.3549 0.93 Platelet 210.50±64.84 205.25±59.11 215.02±69.27 0.1966 -1.29 Lymphocyte ratio 30.39±8.75 29.13±8.15 31.48±9.12 0.0212 -2.32 Neutrophil ratio 58.34±10.06 59.16±10.54 57.64±9.60 0.1947 1.30 Hemoglobin 136.24±15.70 137.30±17.03 135.33±14.45 0.2834 1.07 Neutrophil count 3.85±1.86 4.03±2.15 3.69±1.57 0.1132 1.59 Lymphocyte count 1.86±0.61 1.82±0.65 1.90±0.57 0.2278 -1.21 Grain-flow ratio 2.35±1.90 2.57±2.32 2.16±1.42 0.0676 1.83 Single flow ratio 4.12±1.61 3.87±1.40 4.34±1.75 0.0136 -2.48 Platelets / lymphocytes 125.80±65.82 129.32±74.40 122.77±57.49 0.3942 0.85 Monocyte count 0.49±0.16 0.49±0.15 0.48±0.17 0.4610 0.74 Eosinophilic count 0.20±0.31 0.19±0.28 0.21±0.33 0.6051 -0.52 Basophil counts 0.07±0.36 0.10±0.53 0.04±0.08 0.1729 1.37 Mean erythrocytehemoglobin content 30.45±2.04 30.46±2.02 30.44±2.07 0.9256 0.09 Mean erythrocyte volume 90.62±6.73 90.53±8.55 90.68±4.65 0.8481 -0.19 Mean platelet volume 12.51±17.95 13.66±23.83 11.51±10.54 0.3043 1.03 Erythrocyte distributionwidth SD 15.85±31.26 17.42±37.98 14.50±24.08 0.4236 0.80 Platelet distributionwidth SD 13.47±2.82 13.56±2.98 13.39±2.68 0.5967 0.53 Apolipoprotein A1 1.20±0.25 1.20±0.27 1.21±0.23 0.5795 -0.55 Apolipoprotein B 0.90±0.77 0.87±0.80 0.93±0.75 0.4718 -0.72 albumin 40.19±5.03 39.64±6.19 40.66±3.71 0.0807 -1.75 prealbumin 219.53±45.92 215.45±44.77 223.05±46.74 0.1561 -1.42 Plasma D-dimer 0.44±0.45 0.46±0.43 0.43±0.48 0.4774 0.71 Thrombin time 16.93±1.34 17.12±1.39 16.76±1.27 0.0212 2.32 Prothrombin time 13.36±0.88 13.39±1.09 13.34±0.66 0.6289 0.48 Fibrinogen 3.19±0.75 3.12±0.66 3.25±0.81 0.1600 -1.41 Fibrin degradationproducts 2.93±2.32 3.12±2.91 2.77±1.63 0.1960 1.30 Activated partialthromboplastin time 36.66±3.92 36.45±3.74 36.84±4.08 0.3904 -0.86 Prothrombin internationalnormalized ratio 1.02±0.11 1.02±0.14 1.02±0.06 0.8409 -0.20 Electrocardiogram Normal 206(69.59) 95(69.34) 111(69.81) 0.4343 2.73 Arrhythmia 41(13.85) 23(16.79) 18(11.32) Myocardial ischemia 38(12.84) 15(10.95) 23(14.47) Arrhythmia plusmyocardial ischemia 11(3.72) 4(2.92) 7(4.40) Total 296(100.00) 137(100.00) 159(100.00)

[0084] A P-value < 0.05 was considered statistically significant. Statistical analysis was performed using SAS version 9.4 (SAS Institute Inc) and R version 4.0.4 (R Foundation for Statistical Computing).

[0085] Step S2: Based on the training group data, the least absolute shrinkage and selection operator regression method was used to screen out the risk factors related to central vertigo.

[0086] The R package "glmnet" was used to screen out the risk factors related to central vertigo in vertigo patients using the least absolute shrinkage and selection operator (LASSO) regression method.

[0087] LASSO analysis was used to select variables: Taking the binomial deviance as the criterion, the optimal hyperparameter λ was determined in the LASSO regression through ten-fold cross-validation. Based on the optimal λ, 15 features with non-zero coefficients were screened out as the best predictors of central vertigo. In the LASSO regression, λ (the regularization parameter) was selected based on the one-standard-error criterion (the right dotted line) and the minimum criterion (the left dotted line) (see Figure 2 A in). The LASSO coefficient profile plot of candidate features. The cross curve represents the number of features retained at this log(λ) value. According to the one-standard-error criterion, 28 predictive variables (including dummy variables) with non-zero coefficients were selected (see Figure 2 B in).

[0088] Fifteen variables were selected through LASSO regression, including related to body position or head position, visual rotation, hearing loss, headache, gender, age, systolic blood pressure, disease course (> 3 days), activated partial thromboplastin time, urea, onset form, nausea / vomiting / cold sweat, coronary heart disease or stroke, glucose, and renal tubular filtration rate. These variables from the LASSO analysis were used to construct a multivariable Logistic regression model (stepwise method, sls = 0.10, sle = 0.05).

[0089] Based on 15 risk factors and combined with the suggestions of 3 neurologists, we further selected several indicators more closely related to central vertigo from 15 variables to construct two logistic regression models (full model, simplified model).

[0090] Step S3: Based on the risk factors related to central vertigo, different models are constructed and trained using a variety of machine learning algorithms respectively.

[0091] The R package "mlr3verse" is used for model training, and the algorithms used include logistic regression, random forest, k-nearest neighbor, linear discriminant analysis (LDA), naive Bayes, neural network, quadratic discriminant analysis (QDA), and support vector machine (SVM).

[0092] The variables included in the full model are related to body position or head position, visual rotation, hearing loss, headache, gender, age, systolic blood pressure, disease course (>3 days), activated partial thromboplastin time, and urea. The variables included in the simplified model are related to body position or head position, visual rotation, hearing loss, headache, gender, age, systolic blood pressure, and disease course (>3 days) (see Table 2 for the variable assignment table). The relevant variables in both models are included in the stepwise multivariate Logistic regression analysis, and all P values are <0.05. Therefore, we constructed simplified and full Logistic regression models for the diagnosis of central vertigo (see Table 3).

[0093] Table 2 Variable Assignment Table

[0094]

[0095] Table 3 Univariate and Multivariate Logistic Regression Analysis of the Full Model and the Simplified Model

[0096]

[0097] Logistic regression model - simplified model formula:

[0098]

[0099] Coefficient meaning:

[0100] (1) When the correlation with body position or head position is "unrelated", the occurrence probability is about times that when the correlation with body position or head position is "related".

[0101] (2) When visual rotation is "none", the occurrence probability is about times that when visual rotation is "present".

[0102] (3) When hearing loss is "none", the occurrence probability is about that when hearing loss is "present" times

[0103] (4) When the head swelling or headache is "none", the occurrence probability is about times that when the head swelling or headache is "present".

[0104] (5) When the gender is "female", the occurrence probability is about times that when the gender is "male".

[0105] (6) When the age is "≥74", the occurrence probability is about times that when the age is "≤73".

[0106] (7) When the systolic blood pressure is ">129", the occurrence probability is about times that when the systolic blood pressure is "≤129".

[0107] (8) When the course of disease is "chronic", the occurrence probability is about times that when the course of disease is "acute".

[0108] Logistic regression model - full model formula:

[0109]

[0110] (1) When the correlation with body position or head position is "uncorrelated", the occurrence probability is about times that when the correlation with body position or head position is "correlated".

[0111] (2) When the visual rotation is "none", the occurrence probability is about times that when the visual rotation is "present".

[0112] (3) When the hearing loss is "none", the occurrence probability is about times that when the hearing loss is "present".

[0113] (4) When the head swelling or headache is "none", the occurrence probability is about times that when the head swelling or headache is "present".

[0114] (5) When the gender is "female", the occurrence probability is about times that when the gender is "male".

[0115] (6) When the urea condition is "≤5.55", the occurrence probability is about times

[0116] (7) When the age is "≥74", the occurrence probability is about times that when the age is "≤73".

[0117] When the systolic blood pressure is ">129", the probability of occurrence is about times that when the systolic blood pressure is "≤129".

[0118] (9)When the activated partial thromboplastin time is ">38.15", the probability of occurrence is about times that when the activated partial thromboplastin time is "≤38.15".

[0119] (10)When the course of disease is "chronic", the probability of occurrence is about times that when the course of disease is "acute".

[0120] Step S4: Based on the validation group data, perform performance verification on each constructed model, and select the model with the best comprehensive performance as the central vertigo diagnosis model.

[0121] The established model is verified in the internal cohort and the external cohort. The discrimination, calibration and clinical application value of the model are evaluated through the ROC curve, calibration curve, clinical decision curve and clinical impact curve respectively, and through internal validation and external validation. The discrimination ability of the model is evaluated through the receiver operating characteristic (ROC) curve, and calculated using the R package "pROC" and Brier score. The calibration of different models is evaluated through the calibration curve in the R package "rms". The clinical benefit of the model is evaluated through decision curve analysis (DCA), and the calculation method is calculated by the R package "rmda". Based on the above indicators, the best model is selected.

[0122] (1)Internal validation based on the Logistic regression model and new machine learning algorithms

[0123] To determine the accuracy of our central vertigo diagnosis model, a new machine learning (ML) algorithm was used to verify in the internal validation cohort. The support vector machine (SVM) performed the best, with an AUC value of 0.81. In contrast, the performance of K-nearest neighbor (AUC = 0.8), Logistic regression-full model (AUC = 0.78), quadratic discriminant analysis (AUC = 0.78), Logistic regression-simplified model (AUC = 0.76), linear discriminant analysis (AUC = 0.73), neural network (AUC = 0.73), naive Bayes model (AUC = 0.72), random forest (AUC = 0.69) was relatively poor (see Table 3). However, the Brier score of the Logistic regression-simplified model was 0.18, which was the lowest among all models, indicating that this model had better comprehensive performance (see Table 4).

[0124] Table 4 AUC values and Brier scores of the Logistic regression model and new machine learning algorithms in internal validation

[0125]

[0126] Table 5 Parameters of the internal validation calibration curve

[0127]

[0128] The diagnostic model constructed by the present invention shows the area under the ROC curve in improving the diagnostic discrimination performance of central vertigo. The area under the ROC curve can prove which model plays a better role in diagnosing central vertigo. For calibration, the closer the slope of the calibration curve is to 1, the better the effect. See Table 5.

[0129] As Figure 3 shown, the calibration curves of each model in the internal validation group are plotted: the horizontal axis represents the predicted risk of central vertigo, and the vertical axis represents the actually diagnosed central vertigo; the diagonal dotted line represents the perfect prediction of the ideal model; the solid line represents the performance of each model. The closer the solid line is to the diagonal dotted line, the more accurate the prediction of the model.

[0130] In the calibration curves of the internal validation group, both the support vector machine model and the Logistic regression - simplified model showed good calibration effects ( Figure 3 ). The decision curve of central vertigo shows that the Logistic regression model performs higher in terms of clinical benefit ( Figure 4 ).

[0131] (2) External validation based on the Logistic regression model and new machine learning algorithms

[0132] To evaluate the generalization ability and actual performance of our central vertigo diagnostic model, we applied new machine learning (ML) algorithms in the external validation cohort. The Logistic regression - simplified model performed the best, with an AUC value of 0.81, superior to linear discriminant analysis (AUC = 0.8), Logistic regression - full model (AUC = 0.79), random forest (AUC = 0.79), naive Bayes model (AUC = 0.79), quadratic discriminant analysis (AUC = 0.78), support vector machine (AUC = 0.75), K - nearest neighbor (AUC = 0.71), and neural network (AUC = 0.67). At the same time, the Brier score of the Logistic regression - simplified model was 0.19, which was the lowest among all models, indicating that this model had better comprehensive performance (see Table 6).

[0133] Table 6 AUC values and Brier scores of the Logistic regression model and new machine learning algorithms in external validation

[0134]

[0135] Table 7 Parameters of the external validation calibration curve

[0136]

[0137] For the diagnostic model constructed by the present invention, in terms of improving the diagnostic discrimination performance of central vertigo, the area under the ROC curve can prove which model plays a better role in diagnosing central vertigo. In terms of calibration, the closer the slope of the calibration curve is to 1, the better the effect. See Table 7.

[0138] The Logistic regression model also showed a high model efficiency in the calibration assessment ( Figure 5 ). The decision curve of central vertigo shows that Logistic regression performs better in terms of clinical benefits ( Figure 6 ). Through the internal and external validation evaluations of different indicators and considering the interpretability of the model, the Logistic regression-simplified model was finally selected as the final model.

[0139] Step S5: Establish a nomogram for the Logistic regression-simplified model

[0140] Finally, based on the eight variables finally screened out, the simplified logistic regression model was transformed into a nomogram ( Figure 7 ), in order to present the diagnostic model for easy understanding and application of the model. The risk of central vertigo is related to the total score of the scale, including factors related to the correlation with body position or head position, visual rotation, hearing loss, headache, gender, age, systolic blood pressure, and disease course (>3 days).

[0141] The scoring form is simply called the "grap3h2" scoring form, which is composed of the first letters of the important English words included in the nomogram indicators. Specifically as follows: Correlation of head or body position (P), Visual rotation (R), hearing loss (H), headache (H), gender (G), age (A), systolic pressure (P), course of disease / process (P) (more than 3 days).

[0142] Example 2 Central vertigo diagnosis system

[0143] Based on the central vertigo diagnosis model already constructed in Example 1, this example provides a central vertigo system. Exemplarily, for patients clinically suspected of having central vertigo, before undergoing examinations such as head magnetic resonance imaging, by evaluating these 8 variables (non-invasive, common clinical indicators, without increasing the patient's additional costs and medical risks), the risk of the patient developing central vertigo can be predicted according to the nomogram. For example, a 75-year-old male patient has no correlation with head position or body position, no visual rotation, no hearing loss, headache, a systolic blood pressure of 140 mmHg, and a disease course of more than 3 days. By looking up the nomogram, the total score of the above indicators is about 430 points, and the corresponding risk of developing central vertigo exceeds 95%. In addition, this model can also be presented in the form of a diagnostic formula, and the diagnostic formula can be made into a computer program. By directly inputting the above 8 indicators, the probability of the patient developing central vertigo can be obtained, which is more convenient for clinical use.

[0144] The diagnostic system includes: a data input module, a prediction module, and an output module.

[0145] The data input module is used to obtain data on risk factors related to central vertigo. The risk factors related to central vertigo obtained by the data input module include: the correlation with body position or head position, visual rotation, hearing loss, head distension or headache, gender, age, systolic blood pressure, and disease course (more than 3 days).

[0146] The prediction module is used to predict the probability value of having central vertigo according to the pre-constructed central vertigo diagnosis model. The pre-constructed central vertigo diagnosis model is a logistic regression model, and the logistic regression model is presented by a diagnostic formula and / or a nomogram (such as Figure 7 shown).

[0147] The diagnostic formula for predicting the probability of the current patient developing central vertigo is:

[0148]

[0149] where P is the probability value of predicting the current patient developing central vertigo.

[0150] The output module is used to output probability values, diagnostic conclusions, and their corresponding clinical decisions. After asking the patient about their medical history, based on their clinical symptoms, it is determined whether dizziness is related to head position, whether there is rotational vertigo, hearing loss, head swelling or headache, the duration of the illness, and the blood pressure at the onset of the illness. Based on the patient's demographic information, their gender and age are obtained. For example, a 75-year-old male patient has no correlation with head position or body position, no rotational vertigo, no hearing loss, has a headache, a systolic blood pressure of 140 mmHg, and an illness duration of more than 3 days. By looking up the nomogram, the total score of the above indicators is about 430 points, and the risk of central vertigo formation corresponding to this exceeds 95%, indicating that there is a 95% probability that this dizzy patient will be diagnosed with central vertigo. Exemplarily, the probability value, diagnostic conclusion, and their corresponding clinical decisions are directly output on the user interface; or, a diagnostic report containing the probability value, diagnostic conclusion, and their corresponding clinical decisions is output.

[0151] Embodiment 3 Computer-readable storage medium

[0152] A computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it implements the steps in the method for constructing a central vertigo diagnosis model as in Embodiment 1.

[0153] A computer-readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (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 versatile disc (DVD), memory stick, floppy disk, mechanical encoding devices, such as punched cards or raised structures in grooves storing instructions thereon, and any suitable combination of the above. The computer-readable program instructions described herein can be downloaded from the computer-readable storage medium to various computing / processing devices, or downloaded to an external computer or external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0154] It should be noted that the embodiments of the present invention have better implementability and do not impose any form of limitation on the present invention. Any person skilled in the art may use the technical content disclosed above to modify or transform it into equivalent effective embodiments. However, as long as it does not depart from the technical solution of the present invention, any modification, equivalent change or modification made to the above embodiments based on the technical essence of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for constructing a central vertigo diagnosis model, characterized in that Including: Step S1: Obtain the data of dizziness patients meeting the inclusion and exclusion criteria, and randomly divide them into training group data and validation group data according to a certain ratio; Step S2: Based on the training group data, use the least absolute shrinkage and selection operator regression method to screen out the risk factors related to central vertigo; Step S3: Based on the risk factors related to central vertigo, use a variety of machine learning algorithms to construct different models and train the models; Step S4: Based on the validation group data, verify the performance of each constructed model, and select the model with the best comprehensive performance as the central vertigo diagnosis model.

2. The method for constructing a central vertigo diagnosis model according to claim 1, wherein In step S2, the risk factors related to central vertigo include: the correlation with body position or head position, visual rotation, hearing loss, head swelling or headache, gender, age, systolic blood pressure, and disease course (more than 3 days).

3. The method for constructing a central vertigo diagnosis model according to claim 2, wherein In step S3, the machine learning algorithms include logistic regression, random forest, k-nearest neighbor, linear discriminant analysis (LDA), naive Bayes, neural network, quadratic discriminant analysis (QDA), and support vector machine (SVM).

4. The method for constructing a central vertigo diagnosis model according to claim 3, characterized in that In step S4, the discrimination, calibration, and clinical application value of the model are internally and externally verified through the ROC curve, calibration curve, clinical decision curve, and clinical impact curve. The model with the best comprehensive performance is the logistic regression model, and the logistic regression model is presented through a diagnostic formula and / or nomogram.

5. The method for constructing a central vertigo diagnosis model according to claim 4, wherein The diagnostic formula for predicting the formation probability of central vertigo in the current patient is: ， where P is the probability value of predicting the formation of central vertigo in the current patient; The assignment of each risk factor in the diagnostic formula is shown in the following table: 。 6. A central vertigo diagnosis system, characterized in that, Including: A data input module for obtaining data on risk factors related to central vertigo; A prediction module for predicting the probability value of having central vertigo according to the pre-constructed central vertigo diagnosis model; An output module for outputting the probability value, diagnostic conclusion, and its corresponding clinical decision.

7. The central vertigo diagnosis system according to claim 6, characterized in that, The risk factors related to central vertigo obtained by the data input module include: the correlation with body position or head position, visual rotation, hearing loss, head swelling or headache, gender, age, systolic blood pressure, and disease course (more than 3 days).

8. The central vertigo diagnosis system according to claim 7, characterized in that, The pre-constructed central vertigo diagnosis model is a logistic regression model, and the logistic regression model is presented through a diagnostic formula and / or nomogram.

9. The central vertigo diagnosis system according to claim 8, wherein, The diagnostic formula for predicting the formation probability of central vertigo in the current patient is: ， where P is the probability value of predicting the formation of central vertigo in the current patient; The assignment of each risk factor in the diagnostic formula is shown in the following table: 。 10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps in the method for constructing a central vertigo diagnosis model according to any one of claims 1 to 5.

Images