A method, device and system for predicting the occurrence of pulmonary fibrosis after viral pneumonia

Abstract

The present invention belongs to the field of intelligent medical care, and specifically relates to a method, device and system for predicting the occurrence of pulmonary fibrosis after viral pneumonia. The method includes: obtaining CT images of patients with viral pneumonia; extracting the CT images to obtain CT parameters, and the CT parameters include one or more of the following: PV, CV, GV, PM; the CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is obtained according to the output label of the model. The present invention uses quantitative CT parameters derived from AI tools to predict whether pulmonary fibrosis will occur after viral pneumonia, which is conducive to taking therapeutic intervention measures as soon as possible according to the prediction results.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and more specifically, to a method, device, system, and computer-readable storage medium for predicting the occurrence of pulmonary fibrosis after viral pneumonia. Background Art

[0002] The number of viral pneumonia survivors now exceeds hundreds of millions, and persistent poor health after acute viral pneumonia (called long-term viral pneumonia) has become a major problem, especially long-term lung abnormalities. Lung abnormalities mainly manifest as ground-glass opacities (GGO), consolidation, fibrotic changes, and cystic changes on chest computed tomography (CT). In some studies, some fibrotic changes showed little improvement during a follow-up period of one to two years, suggesting that this change may be an irreversible pulmonary sequelae. Currently, there is no clear treatment for pulmonary fibrosis (PCPF) after viral pneumonia, but earlier understanding of its risk factors and predictors is crucial for potential anti-fibrotic treatments.

[0003] Pulmonary fibrosis after viral pneumonia is closely related to endothelial and epithelial damage, so microcirculatory disorders caused by endothelial damage may also be related to the occurrence of pulmonary fibrosis. In addition, studies using artificial intelligence (AI)-based pulmonary vascular segmentation have found that in patients during acute viral pneumonia infection, the cross-sectional area of ​​blood vessels is less than 5 mm 2 (i.e., BV5) is significantly reduced, which corresponds to microcirculatory disorders in acute viral pneumonia. It is worth noting that BV5% (calculated as the ratio of BV5 to total BV) can be used as a marker for intubation or death in viral pneumonia and abnormal gas diffusion after viral pneumonia. However, the distribution of vascular volume in patients recovering from viral pneumonia and its relationship with PCPF remain to be explored. In addition to BV5% as a possible predictor, previous studies have found that the initial CT visual score is an independent predictor of PCPF. At the same time, as a more objective, quantitative and stable tool, the quantitative parameters of pneumonia obtained based on artificial intelligence segmentation tools have been widely used in the diagnosis, evaluation and follow-up of viral pneumonia. However, whether these parameters have a predictive effect on the occurrence of pulmonary fibrosis after viral pneumonia has not yet been explored. Summary of the invention

[0004] The present invention provides a method for predicting the occurrence of pulmonary fibrosis after viral pneumonia, which uses the quantitative CT parameters of acute pneumonia obtained based on AI stability and objectivity to predict the occurrence of pulmonary fibrosis after viral pneumonia, and then can take intervention treatment as early as possible based on the prediction results to reduce the occurrence of fibrosis after viral pneumonia.

[0005] A method for predicting the occurrence of pulmonary fibrosis after viral pneumonia, comprising:

[0006] Obtain CT images of patients with viral pneumonia;

[0007] Extracting the CT image to obtain CT parameters, wherein the CT parameters include one or more of the following: PV, CV, GV, PM;

[0008] The CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0009] Further, the CT parameters include one of the following: PV and CV, PV and GV, PV and PM, CV and GV, CV and PM, GV and PM, PV, CV and GV, PV, CV and PM, PV, GV and PM, CV, GV and PM, PV, CV, GV and PM.

[0010] Furthermore, the CT parameters also include one or more of the following: CT visual score, BV5%.

[0011] Furthermore, clinical data of viral pneumonia patients are simultaneously acquired, the clinical data are extracted to obtain clinical features, the CT parameters and the clinical features are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0012] Furthermore, laboratory data of patients with viral pneumonia are obtained at the same time, and the laboratory data are extracted to obtain test features, which include one or more of the following: clinical type, anemia, hsCRP, and the CT parameters and the test features are input into the model for classification, and whether pulmonary fibrosis will occur is obtained based on the output label of the model.

[0013] Furthermore, the model includes one or more of the following: linear regression, logistic regression, decision tree, random forest, support vector machine, k nearest neighbor, naive Bayes.

[0014] Furthermore, the model construction method comprises the following steps:

[0015] Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, wherein the dataset includes whether fibrosis occurs after viral pneumonia;

[0016] Step 2, extracting CT parameters of the data set;

[0017] Step 3: Input the features into the classifier to obtain a predicted classification result, compare the predicted classification result with the label in the data set, optimize the model according to the comparison result, and obtain a prediction model.

[0018] A prediction system for viral pneumonia, comprising:

[0019] Acquisition module: Acquisition of CT images of patients with viral pneumonia;

[0020] Feature extraction module: extract the CT image to obtain CT parameters, wherein the CT parameters include one or more of the following: PV, CV, GV, PM;

[0021] Prediction module: The CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0022] A device for predicting the occurrence of fibrosis after viral pneumonia, comprising: a memory and a processor, wherein the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, any one of the above-mentioned methods for predicting the occurrence of pulmonary fibrosis after viral pneumonia is implemented.

[0023] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, any one of the above-mentioned methods for predicting the occurrence of pulmonary fibrosis after viral pneumonia is implemented.

[0024] Advantages of this application:

[0025] 1. This application innovatively explores the predictive effect of quantitative CT parameters derived from AI tools on the occurrence of pulmonary fibrosis after viral pneumonia;

[0026] 2. This application determined through multivariate regression analysis that PV, CV, GV, and PM are independent predictors of pulmonary fibrosis after viral pneumonia;

[0027] 3. This application predicts the occurrence of pulmonary fibrosis after viral pneumonia by performing machine learning modeling on quantitative CT parameters, clinical data, and test features derived from AI tools. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 is a schematic flow chart of a method for predicting the occurrence of pulmonary fibrosis after viral pneumonia provided by an embodiment of the present invention;

[0030] Figure 2 is a schematic flow chart of a system for predicting the occurrence of pulmonary fibrosis after viral pneumonia provided by an embodiment of the present invention;

[0031] Figure 3 is a schematic block diagram of a device for predicting the occurrence of pulmonary fibrosis after viral pneumonia provided by an embodiment of the present invention;

[0032] Figure 4 This is a representative image of abnormal CT fibrosis provided by an embodiment of the present invention, including honeycomb shadows (rectangles in Figure a), tractional bronchiectasis and / or bronchiolar dilatation (arrows in Figure b), lattice shadows (rectangles in Figure b), and structural distortion (circles in Figure c). The above images were all found in the follow-up chest CT 3 months later;

[0033] Figure 5 The present invention provides a volume-rendered chest CT image of a 52-year-old woman in the acute phase (a) and follow-up (b), and color-coded segmentation of the pulmonary vessels by cross-sectional area: (a) BV5% of the acute phase CT is 27.76%; (b) BV5% of the follow-up CT is 51.14%; different color codes represent cross-sectional areas <5 mm 2 (BV5), 5 to 10 mm 2 (BV5–10) and >10 mm 2 (BV10) blood vessel volume, where BV5% represents the percentage of blood vessel volume with a cross-sectional area less than 5 mm2 to the total blood vessel volume.

[0034] Figure 6 Representative images of quantitative analysis of pneumonia lesions provided by embodiments of the present invention: acute phase (a) and follow-up (b), CT scan pneumonia lesions of a 75-year-old woman with critical COVID-19 can be divided into ground glass opacity (GGO) and consolidation, shown in blue and yellow, respectively. The original acute phase and follow-up images of the patient are shown in Figure (c) and Figure (d), respectively.

[0035] Figure 7 The present invention provides a comparison of receiver operating characteristic curves (ROC) for predicting the occurrence of PCPF using PV (pneumonia volume), PM (pneumonia mass), CV (consolidation volume), GV (ground-glass opacity volume) and CT visual score;

[0036] Figure 8 The present invention provides a change in the distribution of pulmonary vascular volume: (a) The bar graph shows the changes in the blood volume ratios of three different groups during the acute phase and follow-up CT scans. The whole blood volume is divided into three groups, including: 2 Blood vessel volume (BV5), cross-sectional area between 5 and 10 mm 2 Blood vessels with a volume between 5 and 10 mm2 (BV5–10) and a cross-sectional area greater than 10 mm2 2(b) Box plot depicting the cross-sectional area of ​​​​less than 5mm in the acute phase and follow-up CT in the group with viral pneumonia fibrosis and the group without fibrosis. 2 Changes in the proportion of pulmonary vascular volume to the total pulmonary vascular volume (BV5%). DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0038] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The sequence numbers of the operations, such as S101, S102, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0040] Figure 1 The present invention provides a schematic flow chart of a method for predicting the occurrence of pulmonary fibrosis after viral pneumonia. Specifically, the method comprises the following steps:

[0041] S101: Obtain CT images of patients with viral pneumonia;

[0042] In one embodiment, the viral pneumonia patient is an acute viral pneumonia patient.

[0043] In one embodiment, CT images of patients with acute viral pneumonia are obtained.

[0044] In one embodiment, a CT image of a patient with acute viral pneumonia is obtained and the clinical data of the patient is obtained at the same time.

[0045] In one embodiment, CT images of patients with acute viral pneumonia are obtained and laboratory test data of the patients are obtained at the same time.

[0046] In one embodiment, CT images of patients with acute viral pneumonia are obtained, and clinical data and laboratory test data of the patients are obtained simultaneously.

[0047] This retrospective study reviewed consecutive patients who visited the fever clinic and emergency department of our hospital between November 28, 2022 and January 8, 2023. Inclusion criteria were: (1) patients diagnosed with viral pneumonia by real-time polymerase chain reaction (RT-PCR) or antigen test results in our hospital; (2) the availability of DICOM data of thin-section chest CT (slice thickness ≤ 1 mm) at the initial visit and at the follow-up visit 3 months later; and (3) the availability of clinical and laboratory test results in the hospital. Exclusion criteria were: (1) patients with insufficient CT image quality for quantitative analysis (n = 21); (2) the interval between follow-up and initial visit was less than 1 month (n = 138); (3) pregnancy (n = 6); and (4) children and adolescents (age < 18 years) (n = 7). A total of 122 patients were included in the study.

[0048] Information was collected through the digital hospital information system (HIS), including demographic data (age, sex), preexisting comorbidities, initial symptoms, initial laboratory test results [blood routine, high-sensitivity C-reactive protein level (hsCRP)], and clinical type at baseline (mild / moderate, severe / critical).

[0049] Platelet count less than 100 × 10 9 / L indicates low platelet count, and less than 120mg / dL for men and less than 110mg / dL for women indicates anemia. hsCRP above 50mg / L indicates elevated hsCRP. The upper WBC count is set at more than 9.5×10 9 / L. According to the diagnosis and treatment plan released by the National Health Commission in January 2023, the preliminary clinical classification is divided into mild / moderate and severe / critical. The results are independently evaluated by senior physicians specializing in respiratory diseases.

[0050] Methods: This retrospective study included chest CT images obtained at the time of initial diagnosis and follow-up chest CT images within 1 to 3 months after the initial diagnosis. Chest CT scans were performed in a single breath-hold using a variety of multidetector CT (MDCT) scanners: Somatom Definition Flash or Somatom Force (Siemens, Forchheim, Germany), Discovery CT750 HD (General Electric, Milwaukee, WI), IQon CT (Philips, The Netherlands), InsitumCT 768 (Sinovision, China), or Aquilion 64 (Toshiba, Tokyo, Japan). Patients were scanned in the supine position with a tube voltage set to 120 kV and automatic tube current exposure control. All images were reconstructed in the axial plane with a slice thickness / interval value of 1 mm / 1 mm, 0.625 mm / 0.625 mm, or 0.5 mm / 0.5 mm. Images were transferred to a picture archiving and communication system (PACS) for further data analysis.

[0051] S102: extracting the CT image to obtain CT parameters, wherein the CT parameters include one or more of the following: PV, CV, GV, PM;

[0052] In one embodiment, the CT parameters further include one or more of the following: CT visual score, BV5%.

[0053] In one embodiment, the CT parameters are: PV and CV.

[0054] In one embodiment, the CT parameters are: PV and GV.

[0055] In one embodiment, the CT parameters are: PV and PM.

[0056] In one embodiment, the CT parameters are: CV and GV.

[0057] In one embodiment, the CT parameters are: CV and PM.

[0058] In one embodiment, the CT parameters are: GV and PM.

[0059] In one embodiment, the CT parameters are: PV and CV and GV.

[0060] In one embodiment, the CT parameters are: PV and CV and PM.

[0061] In one embodiment, the CT parameters are: PV and GV and PM.

[0062] In one embodiment, the CT parameters are: CV and GV and PM.

[0063] In one embodiment, the CT parameters are: PV and CV and GV and PM.

[0064] In one embodiment, the CT parameters are: PV and CT visual score.

[0065] In one embodiment, the CT parameters are: PV and CV and CT visual score.

[0066] In one embodiment, the CT parameters are: PV and CT and PM and visual score.

[0067] In one embodiment, the CT parameters are: PV and CV and GV and PM and CT visual score.

[0068] In one embodiment, the CT parameters are: PV and CV and GV and PM and CT visual score and hsCRF.

[0069] Image analysis was performed using preselected lung (width, 1200HU; horizontal, -600HU) and mediastinal (width, 450HU; horizontal, 50HU) window settings. Two senior radiologists with extensive experience in chest image analysis independently reviewed the images on the PACS system without prior knowledge of the patients' clinical history or laboratory results. Ambiguous findings were discussed by the two to reach a consensus conclusion. For each patient, qualitative CT features were classified using terms from the Fleischner Society glossary. Identified features included ground-glass opacities, consolidation, pleural thickening, pleural effusion, pericardial effusion, regional lymphadenopathy, reticular opacities, emphysema, bronchial or bronchiolar wall thickening, linear atelectasis, pleural parenchymal bands, and bronchiectasis and / or bronchioles. In addition, the presence of reticular opacities, traction bronchiectasis and / or bronchioles, architectural distortion, or honeycombing indicated the presence of fibrotic changes ( Figure 4 Then the patients were divided into fibrosis group and non-fibrosis group according to the presence or absence of fibrosis changes in follow-up CT images.

[0070] All of these abnormalities on chest CT images were quantitatively assessed for lung involvement at the initial visit by a radiologist with senior experience using a semiquantitative assessment system. The extent of pneumonia in each of the five lobes was assessed using a scale of 0 to 5: 0 indicates the absence of disease, 1 indicates less than 5% involvement; 2 indicates involvement between 5% and 24%; 3 corresponds to involvement between 25% and 49%; 4 is used for involvement between 50% and 74%; and 5 indicates more than 75% involvement. The total CT visual score represents the overall lung involvement and is the sum of the scores for each lobe, with a possible range from 0 (indicating no disease) to 25 (indicating maximal involvement).

[0071] Automatic lung segmentation was achieved using an encoder-decoder neural network. The vascular volumes were then divided into three groups based on their cross-sectional areas. These three groups included 2 Blood vessel volume (BV5), cross-sectional area of ​​5 mm 2 Up to 10mm 2 Blood vessels with a volume between 5 and 10 mm2 (BV5–10) and a cross-sectional area greater than 10 mm2 2 Vascular volume (BV10) ( Figure 5 The blood volume percentages of the three groups (BV5%, BV5-10% and BV10%) were calculated relative to the total pulmonary vascular volume. Pneumonia lesions (including GGO and consolidation) were also quantitatively analyzed based on the segmentation results of the neural network ( Figure 6 ).

[0072] These quantitative parameters include the pneumonia volume (PV) of both lungs in ml, the mean attenuation of pneumonia lesions (PA) in HU, the GGO volume (GV) in ml, the consolidation volume (CV) in both lungs in ml. The pneumonia mass (PM) is calculated as follows:

[0073]

[0074] Among them, the unit of PM is g, the unit of PV is ml, and the unit of PA is HU.

[0075] Statistical analysis: Categorical variables are presented as frequencies and percentages. Non-normally distributed continuous variables were described using median and interquartile range (IQR), whereas normally distributed continuous variables were summarized by mean and standard deviation (SD). All statistical analyses were performed in R (version 4.3.1). Intergroup differences were assessed using the chi-square test for categorical data, the analysis of variance test for normally distributed continuous data, and the Kruskal-Wallis or Wilcoxon nonparametric test for continuous data that did not follow a normal distribution. P values ​​were adjusted for multiple comparisons using the false discovery rate (FDR) method. To address missing data in laboratory test results, multiple imputation was performed using the “mice” package in R.

[0076] Receiver operating characteristic (ROC) curve analysis was performed for all the above AI-generated quantitative parameters and CT visual scores to compare their effectiveness in predicting pulmonary fibrosis (PCPF). The optimal cutoff values ​​were determined by maximizing the Youden index. The corresponding area under the curve (AUC), accuracy, precision, specificity, sensitivity, and F1 score were calculated. Univariate logistic regression analysis was applied to each CT parameter, demographic information (age, sex), initial laboratory results, and clinical classification to evaluate their association with PCPF. In this step, comorbidities and neutrophil and lymphocyte counts were removed from the variable set as categorical variables, as they showed highly uneven distribution. Thereafter, multivariate logistic regression models included age, sex, clinical type, and abnormal laboratory test results (anemia and elevated hsCRP levels) as covariates to determine whether CT parameters could independently predict PCPF. Finally, the Delong test was performed in R to test the differences in AUC values ​​between CT parameters.

[0077] Demographic, clinical, and laboratory characteristics: A total of 122 patients (73 males, 49 females; median age, 68.50 [60.00, 78.75] years) were included in this study as shown in Table 1. No patient reported a history of viral pneumonia.

[0078] The median and IQR interval between follow-up CT and initial presentation was 54 days and [43, 75], respectively. All patients were divided into a nonfibrotic group (n = 54) and a fibrotic group (n = 68) according to the presence or absence of de novo fibrotic changes on follow-up CT. In the fibrotic group, patients were generally older (median [IQR]: 74.50 [66.00, 84.25] vs. 64.00 [51.75, 69.00], P < 0.001), were more often classified as severe / critical at initial presentation (57.35% vs. 18.52%, P < 0.001), and were more likely to have at least one pre-existing comorbidity than the nonfibrotic group (100% vs. 87.04%, P = 0.008) (Table 1).

[0079] Table 1 Comparison of components of demographic and clinical characteristics

[0080]

[0081] Note: Unless otherwise stated, data are presented as the number of patients, and percentages are expressed in brackets. IQR: interquartile range; * indicates P < 0.05.

[0082] According to the clinical and laboratory test results, the fibrosis group was more likely to have anemia (47.06% vs. 14.81%, P = 0.001) and elevated hsCRP (51.47% vs. 24.07%, P = 0.01) than the non-fibrosis group. The fibrosis group showed more chest tightness than the non-fibrosis group (50.00% vs. 27.78%, P = 0.02).

[0083] Comparison of qualitative CT features, CT visual scores, and quantitative CT parameters at baseline: As shown in Table 2, the CT visual score of patients in the fibrosis group was significantly higher than that of the non-fibrosis group (median [IQR]: 14.00 [7.75, 17.00] vs. 7.00 [3.00, 9.75], P < 0.001). The incidence of GGO (100.00% vs. 83.33%, P = 0.002) and consolidation (52.94% vs. 27.78%, P = 0.009) on acute CT in patients in the fibrosis group was higher than that in patients in the non-fibrosis group. In terms of quantitative CT parameters, the baseline PV, CV, GV, and PM levels of patients in the fibrosis group were higher than those in the non-fibrosis group (all P < 0.001). In terms of blood volume distribution, the BV5% of patients in the acute CT in patients with fibrosis was significantly lower than that in patients in the non-fibrosis group (P = 0.003).

[0084] Table 2 Comparison of components of initial CT features

[0085]

[0086]

[0087] Table 3 lists the results of ROC curve analysis and the optimal cutoff values ​​for all CT parameters.

[0088] Table 3 Cutoff values, AUC (95% confidence interval), accuracy, sensitivity, specificity, precision, and F1 score for quantitative CT parameters and CT visual scores

[0089]

[0090] Univariate logistic regression analysis showed that age greater than 68.5 years (P<0.001), male sex (P=0.0497), severe / critical clinical classification (P<0.001), anemia (P<0.001), and elevated hsCRP (P=0.02) at the first visit were more likely to develop PCPF. Using these variables as covariates, multivariate logistic regression analysis was performed for each CT parameter. The multivariate results in Table 4 indicated that PV (adjusted OR 3.28, 95% CI: 1.20-9.31, P = 0.02), CV (adjusted OR 3.77, 95% CI: 1.37-10.75, P = 0.01), GV (adjusted OR 3.77, 95% CI: 1.37-10.75, P = 0.01) OR 3.38, 95% CI: 1.26-9.38, P = 0.02), CT visual score (adjusted OR 12.06, 95% CI: 3.15-58.89, P < 0.001) and PM (adjusted OR 3.58, 95% CI: 1.28-10.46, P = 0.02) were independent predictors of PCPF at their respective cutoff values. In contrast, although BV5% was negatively correlated with the presence of PCPF in univariate logistic regression analysis (OR 0.33, 95% CI: 0.16-0.70, P = 0.004), it was not an independent predictor in the subsequent multivariate model after adjustment for the above-mentioned confounders (adjusted OR 0.72, 95% CI: 0.29-1.87, P = 0.50).

[0091] Table 4 Univariate and multivariate logistic regression analysis of quantitative CT parameters and CT visual scores

[0092]

[0093]

[0094] Note: OR: odds ratio; CI: confidence interval; Adjusted models used age, gender, clinical type, and abnormal laboratory test characteristics (anemia and elevated hsCRP) as covariates; PV: pneumonia volume; CV: consolidation volume; GV: ground glass volume; PM: pneumonia weight; BV5%: the percentage of blood volume in vessels with a cross-sectional area of ​​less than 5 mm2 to the pulmonary vascular volume; PA: CT attenuation.

[0095] ROC curves of PV, PM, CV, GV and CT visual scores are shown in Figure 2. Figure 7As shown in Table 3, CV showed the highest AUC value (0.80, 95% CI: 0.72-0.88) compared with PV (0.79, 95% CI: 0.71-0.86), GV (0.78, 95% CI: 0.70-0.86), PM (0.79, 95% CI: 0.71-0.87) and CT visual score (0.77, 95% CI: 0.69-0.85). However, the AUC values ​​of these AI-derived CT parameters were not significantly different from those of CT visual score by Delong test (all P>0.05). However, relative to these AI-derived quantitative CT parameters, CT visual score had the lowest sensitivity (0.54) (PV: 0.69, CV: 0.75, GV: 0.68, PM: 0.69).

[0096] Comparison of CT features between initial and follow-up chest CT: At the 3-month follow-up chest CT, 20 of 122 patients (16.4%) had normal CT findings. Persistent ground-glass opacities (GGO) were observed in 103 of 122 patients (84.43%), while changes were observed in 15 of 122 patients (12.30%). Fibrotic changes were observed in 68 of 122 patients (55.74%).

[0097] As shown in Table 5, the proportion of patients with changes in follow-up CT (41.80% vs. 12.30%, P < 0.001) was significantly reduced compared with the initial chest CT. In addition, follow-up CT showed a significant decrease in CT visual score, PV, CV, GV, PM, and PA (all P < 0.001). Follow-up results of microcirculation showed that BV5% increased significantly in follow-up CT compared with PV (P < 0.001), while the follow-up value of BV10% decreased significantly compared with the initial value (P < 0.01) ( Figure 8 Meanwhile, compared with the non-fibrosis group, the BV5% of the patients in the fibrosis group was significantly lower at both the initial (P=0.008) and follow-up CT (P=0.048) ( Figure 8 b).

[0098] Table 5 Comparison of CT findings and CT visual scores between initial CT and 3-month follow-up

[0099]

[0100]

[0101] Note: GGO: ground-glass opacity; IQR: interquartile range; PV: pneumonia volume; CV: consolidation volume; GV: ground-glass opacity volume; PM: pneumonia weight; BV5%: cross-sectional area less than 5 mm 2The blood volume in the blood vessels is the percentage of the pulmonary vascular volume; PA: CT attenuation.

[0102] This retrospective study showed that 55.74% of patients showed pulmonary fibrosis changes at 3-month follow-up. Compared with patients who did not develop PCPF, patients who developed PCPF tended to be older, more likely to be initially classified as severe / critical, and had a higher incidence of pre-existing comorbidities. In addition, anemia and elevated hsCRP levels at the initial visit were more common in the PCPF group. Independent predictors of PCPF included PV, CV, GV, and PM, as well as initial CT visual scores. According to ROC analysis, the AUC values ​​of these five parameters did not show significant differences. At the same time, there was a significant redistribution of lung blood volume in follow-up CT, manifested as an increase in BV5% and a decrease in BV10%. At the same time, compared with patients who did not develop PCPF, patients with PCPF always had lower BV5% in initial and follow-up CT.

[0103] In our study, the proportion of patients with fibrotic changes after viral pneumonia was higher than in some studies (ranging from 1.6% to 28.4%). This difference may be due to different definitions of fibrotic changes on CT. Some studies restricted the definition of pulmonary fibrosis to volume loss and the occurrence of traction bronchiectasis and / or bronchial dilatation, which resulted in a lower incidence of fibrosis after viral pneumonia. In addition, these different results may be related to the younger age of their population and the lower proportion of patients with pre-existing comorbidities, while the elderly patient population in our study had a higher proportion of pre-existing comorbidities. At the same time, most patients in our study had no previous exposure to the virus, which also made them more seriously ill at baseline. Recent studies have found that semiquantitative CT visual scores are a favorable tool for stratifying pneumonia severity and predicting PCPF in patients with viral pneumonia. In our study, PV, CV, GV, PM, and CT visual scores were all independent predictors of PCPF. Based on ROC curve analysis, the quantitative CT parameters derived from AI matched the predictive ability of CT visual scores for PCPF. Furthermore, the sensitivity of these AI-derived parameters was higher than that of CT visual scores, suggesting that AI-derived parameters may be more sensitive predictors relative to CT visual scores. In addition, AI tools offer the potential for faster, more stable, and more objective assessments.

[0104] In our study, we longitudinally evaluated the changes in pulmonary vascular volume distribution after viral pneumonia using an artificial intelligence-based lung segmentation tool for the first time. A significant increase in BV5% and a significant decrease in BV10% were observed on follow-up CT, suggesting that microcirculatory dysfunction had recovered after infection. However, recently, Mohamed I et al. reported that 87.4% of patients had perfusion abnormalities on 6-month follow-up dual-energy CT (DECT) angiography, suggesting that microcirculatory dysfunction may persist for up to 6 months after viral pneumonia. In addition, Kuchler T et al. found that persistent endothelial dysfunction was associated with long-term abnormal symptoms in patients after viral pneumonia. Accordingly, in our study, PCPF patients had lower BV5% on follow-up CT than non-PCPF patients, suggesting that PCPF patients had persistent microcirculatory dysfunction. However, the relationship between persistent microcirculatory dysfunction and PCPF remains to be studied.

[0105] Our study has some limitations. First, the study was conducted in a single center, and the number of study participants was small. Second, no correlation analysis was performed on the temporal changes in CT scans with symptoms and lung function. Third, longer monitoring is needed to estimate whether fibrotic changes and microcirculatory dysfunction are irreversible. In conclusion, our study showed that AI-derived quantitative CT parameters (including PV, CV, GV, and PM) were independent predictors of PCPF, and their predictive ability was comparable to that of CT visual scores. The vessel segmentation results showed that patients with PCPF had persistent microcirculatory dysfunction, suggesting a potential correlation between BV5% and PCPF. Therefore, our results suggest that AI-based CT quantitative analysis may be an effective and important tool for managing the sequelae of viral pneumonia.

[0106] S103: The CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0107] In one embodiment, clinical data of viral pneumonia patients are obtained simultaneously, the clinical data are extracted to obtain clinical features, the CT parameters and the clinical features are input into a model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0108] In one embodiment, laboratory data of patients with viral pneumonia are obtained simultaneously, and the laboratory data are extracted to obtain test features, wherein the test features include one or more of the following: severe / critical clinical type, anemia, and hsCRP. The CT parameters and the test features are input into a model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0109] In one embodiment, the model includes one or more of the following: linear regression, logistic regression, decision tree, random forest, support vector machine, k nearest neighbor, naive Bayes.

[0110] In one embodiment, the method for constructing the model includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, wherein the dataset includes a label indicating whether fibrosis occurs after viral pneumonia; Step 2, extracting CT parameters of the dataset; Step 3, inputting the features into a classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the model according to the comparison result, and obtaining a prediction model.

[0111] In one embodiment, the method for constructing a classifier includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, the dataset including a label indicating whether fibrosis occurs after viral pneumonia; Step 2, extracting features of the dataset, the features including: PV and CV; Step 3, inputting the features into the classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the classifier according to the comparison result, and obtaining a classifier 1 for the occurrence of pulmonary fibrosis after viral pneumonia. In one embodiment, the PV and CV of the patient to be predicted are extracted and input into classifier 1 to predict whether pulmonary fibrosis will occur after viral pneumonia in the patient.

[0112] In one embodiment, the method for constructing a classifier includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, the dataset including a label indicating whether fibrosis occurs after viral pneumonia; Step 2, extracting features of the dataset, the features including: PV, CV and GV; Step 3, inputting the features into the classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the classifier according to the comparison result, and obtaining a classifier 2 for the occurrence of pulmonary fibrosis after viral pneumonia. In one embodiment, the PV, CV and GV of the patient to be predicted are extracted and input into classifier 2 to predict whether pulmonary fibrosis will occur after viral pneumonia in the patient.

[0113] In one embodiment, the classifier construction method includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, the dataset including a label of whether fibrosis occurs after viral pneumonia; Step 2, extracting features of the dataset, the features including: PV and CV and GV and PM; Step 3, inputting the features into the classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the classifier according to the comparison result, and obtaining a classifier 3 for the occurrence of pulmonary fibrosis after viral pneumonia. In one embodiment, the PV and CV and GV and PM of the patient to be predicted are extracted and input into the classifier 3 to predict whether pulmonary fibrosis will occur after viral pneumonia.

[0114] In one embodiment, the method for constructing a classifier includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, the dataset including a label indicating whether fibrosis occurs after viral pneumonia; Step 2, extracting features of the dataset, the features including: PV, CV, GV, PM, and CT visual scores; Step 3, inputting the features into the classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the classifier according to the comparison result, and obtaining a classifier 4 for the occurrence of pulmonary fibrosis after viral pneumonia. In one embodiment, the PV, CV, GV, PM, and CT visual scores of the patient to be predicted are extracted and input into the classifier 4 to predict whether pulmonary fibrosis will occur after viral pneumonia in the patient.

[0115] In one embodiment, the classifier construction method includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, the dataset including a label of whether fibrosis occurs after viral pneumonia; Step 2, extracting features of the dataset, the features including: PV and CV and GV and PM and hsCRF; Step 3, inputting the features into the classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the classifier according to the comparison result, and obtaining a classifier 5 for the occurrence of pulmonary fibrosis after viral pneumonia. In one embodiment, the PV and CV and GV and PM and hsCRF of the patient to be predicted are extracted and input into the classifier 5 to predict whether pulmonary fibrosis will occur after viral pneumonia in the patient.

[0116] In one embodiment, the method for constructing a classifier includes the following steps: Step 1, obtaining a dataset of patients in the recovery period of viral pneumonia, the dataset including a label indicating whether fibrosis occurs after viral pneumonia; Step 2, extracting features of the dataset, the features including: PV and CV and GV and PM and CT visual scores and hsCRF and clinical types, the clinical types being mild / moderate or severe / critical; Step 3, inputting the features into the classifier to obtain a predicted classification result, comparing the predicted classification result with the label in the dataset, optimizing the classifier according to the comparison result, and obtaining a classifier 6 for the occurrence of pulmonary fibrosis after viral pneumonia. In one embodiment, the PV and CV and GV and PM and CT visual scores and hsCRF and clinical types of the patient to be predicted are extracted and input into the classifier 6 to predict whether pulmonary fibrosis will occur after viral pneumonia in the patient.

[0117] Figure 2 It is a schematic flow chart of a system for predicting the occurrence of pulmonary fibrosis after viral pneumonia provided by an embodiment of the present invention.

[0118] A prediction system for the occurrence of pulmonary fibrosis after viral pneumonia, comprising:

[0119] Acquisition module 201: acquiring CT images of patients with viral pneumonia;

[0120] Feature extraction module 202: extracting the CT image to obtain CT parameters, wherein the CT parameters include one or more of the following: PV, CV, GV, PM;

[0121] Prediction module 203: The CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

[0122] Figure 3 It is a schematic block diagram of a device for predicting the occurrence of pulmonary fibrosis after viral pneumonia provided by an embodiment of the present invention.

[0123] A device for predicting the occurrence of pulmonary fibrosis after viral pneumonia, comprising: a memory and a processor;

[0124] The memory is used to store program instructions;

[0125] The processor is used to call program instructions, and when the program instructions are executed, the steps of the method for predicting the occurrence of pulmonary fibrosis after viral pneumonia are implemented.

[0126] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for predicting the occurrence of pulmonary fibrosis after viral pneumonia.

[0127] The validation results of this validation example show that assigning intrinsic weights to indications can moderately improve the performance of the method relative to the default setting.

[0128] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0129] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0130] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0132] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0133] A person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0134] The above is a detailed introduction to a computer device provided by the present invention. For a person skilled in the art, according to the concept of the embodiments of the present invention, there may be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting the occurrence of pulmonary fibrosis after viral pneumonia, characterized in that: The method comprises: obtaining a CT image of a patient with viral pneumonia; The CT image is extracted to obtain CT parameters, which include PM. PM is calculated as follows: Among them, PM is the weight of pneumonia in g, PV is the volume of pneumonia in ml, and PA is the CT attenuation of pneumonia lesions in HU; The CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

2. The method for predicting the occurrence of pulmonary fibrosis after viral pneumonia according to claim 1, characterized in that: The CT parameters also include one of the following: PV and CV, PV and GV, CV and GV, PV, CV and GV.

3. The method for predicting the occurrence of pulmonary fibrosis after viral pneumonia according to claim 1, characterized in that: The CT parameters also include one or more of the following: CT visual score, BV5%.

4. The method for predicting the occurrence of pulmonary fibrosis after viral pneumonia according to claim 1, characterized in that: At the same time, clinical data of patients with viral pneumonia are obtained, the clinical data are extracted to obtain clinical characteristics, the CT parameters and the clinical characteristics are input into the model for classification, and whether pulmonary fibrosis will occur is obtained based on the output label of the model.

5. The method for predicting the occurrence of pulmonary fibrosis after viral pneumonia according to claim 1, characterized in that: At the same time, laboratory data of patients with viral pneumonia are obtained, and the laboratory data are extracted to obtain test features, which include one or more of the following: clinical type, anemia, hsCRP, and the CT parameters and the test features are input into the model for classification, and whether pulmonary fibrosis will occur is obtained based on the output label of the model.

6. The method for predicting the occurrence of pulmonary fibrosis after viral pneumonia according to claim 1, characterized in that: The model includes one or more of the following: linear regression, logistic regression, decision tree, random forest, support vector machine, k nearest neighbor, naive Bayes.

7. The method for predicting the occurrence of pulmonary fibrosis after viral pneumonia according to claim 1, characterized in that: The method for constructing the model comprises the following steps: Step 1: Obtain a dataset of patients in the recovery period of viral pneumonia, wherein the dataset includes a label of whether fibrosis occurs after viral pneumonia; Step 2, extracting CT parameters of the data set; Step 3: Input the features into the classifier to obtain a predicted classification result, compare the predicted classification result with the label in the data set, optimize the model according to the comparison result, and obtain a prediction model.

8. A prediction system for viral pneumonia, characterized in that: include: Acquisition module: Acquisition of CT images of patients with viral pneumonia; Feature extraction module: extract the CT image to obtain CT parameters, the CT parameters include: PM; PM is calculated as follows: Among them, PM is the weight of pneumonia in g, PV is the volume of pneumonia in ml, and PA is the CT attenuation of pneumonia lesions in HU; Prediction module: The CT parameters are input into the model for classification, and whether pulmonary fibrosis will occur is determined based on the output label of the model.

9. A device for predicting the occurrence of fibrosis after viral pneumonia, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; the processor is used to call program instructions, and when the program instructions are executed, a method for predicting the occurrence of pulmonary fibrosis after viral pneumonia as described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the occurrence of pulmonary fibrosis after viral pneumonia as claimed in any one of claims 1 to 7 is implemented.

Images