Prediction of Atrial Fibrillation recurrence after catheter ablation. An explicative machine learning approach incorporating epicardial adipose tissue volume.

Prediction of Atrial Fibrillation recurrence after catheter ablation. An explicative machine learning approach incorporating epicardial adipose tissue volume. | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Prediction of Atrial Fibrillation recurrence after catheter ablation. An explicative machine learning approach incorporating epicardial adipose tissue volume. José Miguel Castro-García, María Javiera Garfias-Baladrón, Antonio Adarve-Castro, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4577588/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose: Atrial fibrillation (AF) is a common arrhythmia with increasing prevalence and significant clinical impact. Catheter ablation has emerged as a treatment option for drug-resistant AF, with variable success rates. This study aimed to develop a machine learning-based predictive model incorporating interatrial, periatrial, and epicardial adipose tissue volumes to predict AF recurrence after pulmonary vein ablation. Methods: This retrospective cohort study included patients who underwent a first ablation procedure between 2017 and 2022. Computed tomography (CT) scans were used to measure left atrial volume (LAV), periatrial (PAT), interatrial (IAT) and (EAT) epicardial adipose tissue volumes. Two models were created and trained under three machine learning techniques. Receiver Operating Characteristic (ROC) curve analysis, accuracy, precision, recall and F1-score were evaluated. SHapley Additive exPlanations (SHAP) analysis was also conducted. Results: From the initial 85 patients, 69 with complete follow-up and CT scan quality were included. Persistent AF, increased left atrial, PAT and IAT volumes were significantly associated with recurrence. The model including clinical and radiological variables achieved accuracies of 0.86, 0.66, and 0.86 and AUCs of 0.91, 0.87, and 0.92 in the testing group by using MLP Classifier Neural Network, Naïve Bayes, and Logistic Regression, respectively. SHAP analysis emphasized the LAV, PAT volume and AF type for recurrence prediction. Conclusion: This study presents a machine learning explicative approach incorporating cardiac adipose tissue volumes for predicting AF post-ablation recurrence. The logistic regression model including clinical and radiological variables demonstrated the highest performance, highlighting the potential of using multimodal data for post-ablation recurrence prediction. Atrial fibrillation ablation techniques machine learning recurrence. Figures Figure 1 Figure 2 Introduction Atrial fibrillation (AF) is the most common arrhythmia globally, with an incidence that increases in parallel with age, reaching 10–17% in individuals over 78 years old (1) and an estimated prevalence of 37,574 million cases worldwide until 2017, a number that might increase by > 60% in 2050 (2). The clinical importance of its diagnosis and treatment lies in both preventing embolic events through anticoagulation and controlling heart rhythm or rate through either permanent pharmacological treatment, cardioversion or interventional ablative techniques. Due to frequent relapses after electrical or drug-induced cardioversion and the increase of interventional cardiology procedures, catheter ablation of pulmonary veins has become a widely accepted treatment choice for symptomatic AF cases resistant to medications [ 3 ], attaining success rates between 59.9–65.4% within 12 months after radiofrequency or cryoablation [ 4 ]. Certain risk factors, such as the type of AF, have been associated with lower success rates after ablation. In a study by Deng et al. [ 5 ], the success rate was 70% for paroxysmal AF patients and dropped to 50% for those with persistent AF. Aligned with the above, a longer duration of the arrhythmia has been linked to a larger left atrial [ 6 ] volume, which is an independent risk factor for recurrence after ablation [ 7 ], probably due to the accompanying cardiac remodeling, increased fibrosis and atypical microcircuits development. Other factors such as metabolic syndrome, obesity and advanced age have also been associated with a decreased rate of success [ 8 – 10 ]. In the last decade, the relevance of the quality and quantity of epicardial adipose tissue has been linked to AF recurrence after an ablation procedure, with a higher risk found in patients with greater amounts and density of epicardial (EAT), periatrial (PAT) and interatrial adipose tissue (IAT) based on computed tomography (CT) scans and other techniques [ 11 – 24 ]. Due to the multiple risk factors that can influence the outcome following catheter ablation, the complex data management, and the need of a recurrence prediction model, some artificial intelligence models based on supervised learning have been created previously. Some studies [ 25 , 26 ] have included blood markers, other clinical and radiological variables, as well as electrocardiographic [ 27 , 28 ] and electrophysiological [ 29 ], variables resulting in some complex models. While cardiac adipose tissue volumes are a known risk factor for AF recurrence, their inclusion for analysis using machine learning techniques has not been performed before and could offer a refined approach to the complex interplay of these novel risk factors. Machine learning models address better the analysis of non-linear patterns and interactions that may not be apparent in traditional analyses. This study applies these advanced analytical capabilities to refine the predictive accuracy for AF recurrence after catheter ablation, incorporating volumetric analysis of the left atrium, interatrial, periatrial and epicardial adipose tissue. Material and methods The institutional review board approved this study (under the registry PI2023-045) according to human rights declarations and regulations. Due to its retrospective design and reliance on medical record review, informed consent was waived. This retrospective cohort study included all consecutive patients with atrial fibrillation (AF) who underwent radiofrequency or cryoablation of pulmonary veins after a contrast CT scan between 2017 and 2021. CT scans with significant motion artifacts or insufficient left atrial enhancement were excluded. The first cohort included patients who experienced recurrence within 18 months after ablation, while the second cohort included those that maintained sinus rhythm during that period. Demographic and clinical data, including age, gender, body mass index (BMI), comorbidities (type 2 diabetes mellitus, dyslipidemia, and systemic hypertension) and AF type (paroxysmal or persistent) were collected. The presence of sinus rhythm or AF recurrence was documented during the 3-, 6-, 12- and 18-months follow-up visits. Computed Tomography (CT) Scans and Cardiac Adipose Tissue Analysis: CT scans were conducted using a 64-slice detector scanner (Phillips Brilliance CT). Images were acquired through a retrospective electrocardiographically synchronized acquisition protocol with dose modulation at 120 kV, 64 × 0.625 mm collimation, and 0.35 s rotation time. A standard dose of 80 ml iodinated contrast with a concentration of 400 mg/ml was injected at 4 ml/s, followed by 30 ml saline at the same rate of injection. Tracking bolus technique was used, placing the region of interest in the left atrium with a trigger threshold of 130 Hounsfield units (HU). CT datasets were reconstructed with a slice thickness of 1.5 mm. Left atrial volume (LAV), epicardial (EAT), periatrial (PAT) and interatrial adipose tissue (IAT) volume were quantified during the cardiac phase that corresponded to the 75% of the R-R interval by manual semi-automatic segmentation performed with Philips IntelliSpace software 12.1 segmentation and subsegmentation tools (Fig. 1 ). All the CT scans were evaluated and segmented by a radiologist from the cardiac imaging department of a third level hospital (JMCG). EAT volume was traced from the top of the aortic arch to the diaphragm, PAT was quantified from the roof of the coronary sinus to the roof of the left atrium and IAT was obtained by tracing the interatrial septum area. Subsegmentation of adipose tissue was based on an attenuation range between − 195 and − 15 HU. Statistical Analysis The normality of all quantitative variables was assessed using the Kolmogorov-Smirnov test. Parametrically distributed variables were described with mean and standard deviation, while non-parametric variables were described with median and quartiles. Qualitative variables were described by absolute and relative frequency. An initial univariate analysis compared AF recurrence and risk factors, including age, gender, BMI, AF type (paroxysmal or persistent), LAV, EAT, PAT and IAT volume. The chi-square test was used for qualitative variables, Student's t-test for parametric quantitative variables, and Mann-Whitney U test for non-parametric quantitative variables. Variables with p-value < 0.05 were considered statistically significant. Statistical analysis was performed using Statistical Package for Social Sciences (SPSS) version 22.0. Feature Selection: Feature selection was conducted through exhaustive automated process that tested all possible combinations of clinical and radiological variables to determine which contributed most significantly to the model's accuracy. This process was performed using Python 3.12 and the scikit-learn library (version 1.1). Two different models were created, the first one (A) including both clinical and radiological variables and the second one (B) including only radiological variables. The variables included in model A were: age, gender, BMI, AF type (paroxysmal or persistent), left atrial volume, PAT, IAT and EAT volume. Model B included left atrial, PAT, IAT and EAT volume. Data Preprocessing: No patients with missing information were included in this study. Before training the model, the distribution of the training group was balanced by using the technique of random under-sampling. This involved reducing the number of instances in the overrepresented class to match the number of instances in the underrepresented class. These was conducted by applying the RandomUnderSampler method from imblearn library. Finally, feature scaling was performed in the dataset in order to ensure that all features contributed evenly to the model's performance. The MinMaxScaler function was applied to both the training and test groups, transforming the feature values to a common scale within the range of 0 to 1. Model training: To create the training set, the preprocessed and balanced data was divided by using the Train_test_split function which randomly divided the whole sample in approximately 80% (54 patients) for training and 20% (15 patients) for testing. Three different ML techniques were used to train the model by using the data of the training set. The first one was a Multi-layer Perceptron (MLP) classifier configured with a single hidden layer comprising 50 units and utilizing ReLU activation function. The Adam solver was used for optimization, and the model underwent a maximum of 2500 iterations during training. This training was evaluated by using a 5-fold cross-validation and calculating the average accuracy, precision, recall, and F1-score within the training group. The second technique employed was Logistic Regression with balanced class weights and the third was Gaussian Naïve Bayes. In both cases performance evaluation was carried out by calculating the average accuracy, precision, recall, and F1-score after a 5-folds cross-validation. Model Evaluation and SHAP Analysis: After the training and validation stages, both models were tested by calculating the Receiver Operating Characteristic (ROC) curve, accuracy, precision, recall, and F1 score (Table 3). Furthermore, SHAP (SHapley Additive exPlanations) analysis was performed, in order to clarify the importance of each variable within all the techniques applied to the two different models (A and B). Results A total of 85 patients were collected initially. From these, only 70 patients had a complete follow-up and adequate Computed Tomography (CT) scan quality. 30 (42%) were women and 40 (58%) men. The total patients that experienced recurrence after catheter ablation were 20 (29%), and the ones that did not recur during the 18 months follow-up were 49 (71%) (Tables 1 , 2 ). From the total of patients that experienced recurrence, 7 (35%) had paroxysmal AF and 13 (65%) had persistent AF, with a significant higher risk of recurrence (p < 0.05) being observed for those with persistent AF (OR 1.99; IC: 1.19–3.33). Table 1 Clinical and radiological characteristics of all the included patients. Variables Recurrence No recurrence Clinical variables 20 (28.6%) 50 (71.4%) Age 60.3 (+/-8.2) 57 (+/-8.7) Sex Male 10 (25%) 30 (75%) Female 10 (33.3%) 20 (66.7%) BMI 28.91 (+/-6.38) 30.39 (+/-5.31) DM2 Yes 6 (46.2%) 7 (53.8%) No 14 (24.6%) 43 (75.4%) Hypertension Yes 11 (52.4%) 10 (47.6%) No 27 (55.1%) 22 (44.9%) Dyslipidemia Yes 10 (27%) 27 (73%) No 10 (30.3%) 23 (69.7%) AF type Persistent 13 (44.8%) 16 (55.2%) Paroxysmal 7 (17.1%) 34 (82.9%) Type of procedure Radiofrequency 15 (27.8%) 39 (72.2%) Cryoablation 5 (33.2%) 11 (66.8%) Treatment Beta-blocker Yes 13 (32.7%) 28 (68.3%) No 7 (24.1%) 22 (75.9%) Calcium antagonist Yes 5 (41.7%) 7 (58.3%) No 15 (25.9%) 43 (74.1%) Oral anticoagulant Yes 13 (36.1%) 23 (63.9%) No 26 (76.5%) 8 (23.5%) ACEi / ARA* Yes 10 (34.5%) 19 (65.5%) No 10 (24.4%) 31 (75.6%) Radiological variables Left atrial volume 133.19 (105.54-168.62) 88.32 (72.85-109.15) Epicardial adipose tissue volume 265.93 (200.63-312.91) 205.82 (117.59-288.94) Periatrial adipose tissue volume 14.68 (11.08–23.79) 8.96 (4.05–11.99) Interatrial adipose tissue volume 2.25 (1.04–4.25) 1.36 (0.62–2.63) *ACEi: Angiotensin-Converting Enzyme inhibitors, ARA: Angiotensin Receptor Antagonists Table 2 Univariate analysis of atrial fibrillation recurrence risk for all the clinical and radiological variables. Variables OR (IC) p Clinical variables Age - > 0.05 Sex Male 0.63 (0.22–1.81) > 0.05 Female BMI - > 0.05 DM2 Yes 2.57 (0.73–8.95) > 0.05 No Hypertension Yes 0.996 (0.35–2.83) > 0.05 No Dyslipidemia Yes 0.81 (0.28–2.31) > 0.05 No AF type Persistent 1.99 (1.19–3.33) 0.05 Cryoablation Treatment Beta-blocker Yes 0.71 (0.24–2.12) > 0.05 No Calcium antagonist Yes 0.42 (0.11–1.57) > 0.05 No Oral anticoagulant Yes 0.476 (0.16–1.39) > 0.05 No ACEi / ARA* Yes 0.63 (0.22–1.81) > 0.05 No Radiological variables Left atrial volume - 0.05 Periatrial adipose tissue volume - < 0.01 Interatrial adipose tissue volume - < 0.05 *ACEi: Angiotensin-Converting Enzyme inhibitors, ARA: Angiotensin Receptor Antagonists Left atrium (LAV), interatrial (IAT), periatrial (PAT) and epicardial adipose tissue (EAT) volume differences between both groups. CT imaging revealed volumetric differences between patients with and without AF recurrence, particularly in left atrial volume (LAV), periatrial adipose tissue (PAT), and interatrial adipose tissue (IAT) volumes (Fig. 1 ). Patients with recurrence exhibited significantly larger LAV (median [IQR]: 133.19 [105.54-168.62] mL) compared to those without recurrence (88.32 [72.85-109.15] mL, p < 0.001). Similarly, PAT and IAT volumes were markedly higher in patients who experienced recurrence (PAT: 14.68 [11.08–23.79] vs. 8.96 [4.05–11.99] cm³, p < 0.01; IAT: 2.25 [1.04–4.25] vs. 1.36 [0.62–2.63] cm³, p < 0.05). Model Performance and Clinical Relevance: Due to the total sample of 70 patients being divided into 55 for training and 15 for testing the model, and the application of a class balancing technique, learning was based on the use of 5-fold cross-validation with different patients from those 55 in each fold. Specifically, the model was trained on 42 patients who did not experience recurrence and 13 who did, testing it on 8 patients without recurrence and 7 who experienced recurrence. Regarding models’ training and validation, the highest values of average precision, accuracy, f1 score and recall after 5-fold cross-validation were obtained for model A, using the Logistic Regression (LR) technique. When evaluating the models in the testing groups, the accuracies of model A using a neural network (NN), Naïve Bayes (NB), and Logistic Regression (LR) were 0.86, 0.66 and 0.86, respectively (Table 3), while the AUCs were 0.91, 0.87 and 0.92, respectively (Table 3). The highest ability to predict patients that experienced recurrence was also achieved by LR, correctly identifying all the patients that experienced recurrence (Table 3). Table 3 Model A and B validation and performance in the test group. Model A. Radiological and clinical variables Training group Test group Accuracy Precision Recall F1-score Accuracy Precision Recall F1-score AUC* Neural Network 0.64 0.566 0.7 0.613 0.866 1.0 0.714 0.833 0.911 Naïve Bayes 0.64 0.566 0.7 0.613 0.666 0.666 0.571 0.615 0.875 Logistic Regression 0.64 0.566 0.7 0.613 0.866 0.777 1.0 0.875 0.928 Model B. Radiological variables Training group Test group Accuracy Precision Recall F1-score Accuracy Precision Recall F1-score AUC* Neural Network 0.76 0.7 0.733 0.693 0.8 0.83 0.714 0.769 0.875 Naïve Bayes 0.72 0.7 0.733 0.693 0.666 0.666 0.571 0.615 0.839 Logistic Regression 0.72 0.7 0.733 0.693 0.733 0.8 0.571 0.66 0.875 SHAP Analysis: The results obtained after applying the SHAP analysis were similar for the 3 ML techniques for model A. The 3 most important features using the NN and NB techniques were LAV, PFT and IFT volumes. The Logistic Regression (LR), while aligning with the importance of the LAV and PFT volumes, diverged by identifying the type of atrial fibrillation (AF type) as a significant predictor, indicating LR might be particularly effective in utilizing categorical data to improve its predictive accuracy (Fig. 2 ). In Model B, the LAV and PFT volumes remained the most influential features in predicting outcomes using the NN and LR methods. However, the NB method's emphasis on the LAV and IFT volumes, as opposed to PFT. The consistent appearance of LAV and PFT volumes as top features in the majority of models reinforces their relevance and higher predictive utility (Fig. 2 ). Discussion This study presents a novel approach for atrial fibrillation (AF) recurrence prediction after catheter ablation, focusing on the role of epicardial adipose tissue volumes along with traditional clinical parameters. Our results demonstrate the potential of integrating radiological markers with clinical variables to improve the accuracy of predicting AF recurrence. A meta-analysis conducted in July 2018 highlighted the importance of left atrium PAT and EFT volumes as predictive markers (30). Other studies have linked increased volume and density of epicardial [ 12 , 13 , 15 , 16 , 21 , 23 , 31 ], interatrial [ 19 ], and periatrial [ 20 , 22 ] adipose tissue with AF recurrence. In our study, only IAT and PAT volumes were significantly associated with recurrence, and although not statistically significant the presence of a large volume of EAT volume was the fifth most important variable for AF recurrence prediction when introduced into the SHAP explicative analysis using the Naive Bayes technique. The task of effectively predicting recurrence after AF ablation continues to be a significant challenge. Existing prognostic models for AF recurrence have a high variability in model performance, highlighting the complexity and the need for robust validation and calibration of predictive models. A previous score (MB-LATER) validated by Potpara et al. in 2018 [ 32 ] on 226 patients for late recurrence prediction showed a good performance (AUC: 0.62 [95% CI: 0.54–0.69], p = 0.003) for AF recurrence prediction after more 12 months. This score includes gender, presence of bundle branch block, left atrial diameter > 46 mm, type of AF, recurrence during the blanking period (first 3 months), and preablation history of persistent AF. Other scores such as CAAP-AF, APPLE and SUCCESS have assessed the predictive value of different variables. CAAP-AF, developed by Winkle et al (33) includes variables such as sex, age, type of AF, Left atrial diameter (LAD), coronary artery disease and number of antiarrhythmic drugs failed. This scored has been evaluated in a previous study that included 283 patients with AF, reaching a sensitivity and specificity of 64% and 68%. The APPLE score, encompassing age, type of AF, estimated glomerular filtration rate (eGFR), LAD, and left ventricular ejection fraction (LVEF) in first and/or repeat ablation populations achieved an AUC of 0.62 in a previous study involving 192 patients [ 34 ]​​. This research also introduced the SUCCESS score, enhancing the APPLE score by adding a point for each prior ablation, which showed improvement in receiver operating characteristic analysis, reaching 0.657. From the previous research analyzing the association between the epicardial adipose tissue volume and density, we could only find one (15) incorporating the pericoronary fat tissue average density into a model. This radiological and clinical model included age, sex, body mass index, atrial fibrillation (AF) type, NT- pro-BNP level, left atrial volume index, left ventricular end- diastolic dimension, and early AF recurrence obtaining and AUC of 0.726. This study employed regression techniques to predict outcomes in a group of patients. The attempt to predict recurrence in patients with atrial fibrillation has also been explored using supervised Machine Learning (ML) techniques. These techniques tend to fit better to variables that have a non-linear interaction between them compared to statistical techniques and allow a more refined tunning of the hyperparameters used to train a model. In a previous study, derived from the ESCEHRA Atrial Fibrillation Ablation Long-Term Registry (AFA-LT) a machine-learning scoring system was created, in order to predict recurrence after 1 year and included a total of 3128 patients [ 35 ]. By evaluating four machine learning techniques (decision tree, random forest, AdaBoost, and k-nearest neighbour) a final model using the random forest technique showed an AUC 0.721, 95% confidence interval (CI: 0.680–0.764) in the testing group by training the model with 19 clinical variables. The most important variables used in this model were left ventricular end-diastolic volume, eGFR, BMI, age, LA diameter and LVEF. Similar variables have been used in other studies, for example, Xue Zhou et al [ 36 ] created a convolutional neural network trained with 4 predictors: LAV, left atrial appendage volume (LAAV), type of AF and N-terminal pro-BNP based on 310 patients with a 25-month follow up obtaining a C-index of 0.75 (0.72–0.79) in the validation set and 0.76 (0.72–0.79) in the training set. The LAV and LAAV were calculated from the pre-ablation CT scan. A limitation of this study was the high proportion of not collected variables, for example, the presence of previous ablations, which might be an important risk factor for recurrence. To our knowledge no previous studies have incorporated cardiac adipose tissue into a machine learning (ML) prediction strategy. Our study was conducted at a single center with a limited number of patients undergoing their first AF ablation. To address the challenge of working with a small dataset of 69 patients, our study employed several methodological strategies to ensure a good performance. Firstly, data normalization was conducted to ensure equitable contribution of all variables to the model learning process. Secondly, we applied 5-fold cross-validation during training, which maximizes the use of our limited dataset by ensuring all observations are utilized for both training and validation. This technique helps in preventing overfitting and enhances the performance on new data. Thirdly, random undersampling was utilized to address class imbalance within our dataset, reducing the size of the majority class to minimize bias towards the more frequent class and enabling more effective learning from the characteristics of both groups. Only patients with complete clinical information and a minimum follow-up of 18 months were included in our study. However, the discrimination found within our dataset needs further validation across diverse populations and other centers to confirm its broader applicability. We did not include the type of ablation procedure performed to the patients (pulmonary vein isolation, Cavo tricuspid isthmus ablation, SVC isolation, etc.) which is another limitation of our study. Finally, our study period was limited to the first 18 months after the ablation, for which our model was trained. The weight assignment to the included variables is based on the recurrence within this timeframe which could yield different results for periods extending beyond 18 months. Conclusion In conclusion, our study, underscores the potential of combining clinical data with cardiac adipose tissue volumes acquired from the CT scan prior to ablation in order to enhance the prediction of AF recurrence. Our study highlights the importance of multimodal data integration and the need for ongoing development and validation of predictive models which could ultimately lead to more tailored therapeutic strategies, improving patient outcomes in AF. Declarations Funding No funding was received for this research. Competing interests The authors declare that they have no competing interests. Authors' contributions (All authors contributed to the study conception and design). JMCG: Data collection, statistical analysis and manuscript writing AUV: Study design and manuscript revision MJGB, APR and HTB: Data collection AAC: Manuscript writing. Ethics approval and consent to participate This study was performed in line with the principles of the Declaration of Helsinki. The institutional review board of the General Universitary Hospital of Alicante, Spain approved this study (under the registry PI2023-045) according to human rights declarations and regulations. Consent to participate Informed consent was waived by the ethics committee. References Zoni-Berisso M, Lercari F, Carazza T, Domenicucci S (2014) Epidemiology of atrial fibrillation: European perspective. Clin Epidemiol 6:213–220. 10.2147/clep.s47385 Lippi G, Sanchis-Gomar F, Cervellin G (2021) Global epidemiology of atrial fibrillation: An increasing epidemic and public health challenge. Int J Stroke 16(2):217–221. 10.1177/1747493019897870 Calkins H, Hindricks G, Cappato et al (2018) 2017 HRS/EHRA/ECAS/APHRS/SOLAECE expert consensus statement on catheter and surgical ablation of atrial fibrillation. Europace 20:e1–e160. 10.1093/europace/eux274 Chen YH, Lu ZY, Xiang Y et al (2017) Cryoablation vs. radiofrequency ablation for treatment of paroxysmal atrial fibrillation: A systematic review and meta-analysis. Europace 19:784–794. 10.1093/europace/euw330 Deng H, Bai Y, Shantsila A, Fauchier L, Potpara TS, Lip GYH (2017) Clinical scores for outcomes of rhythm control or arrhythmia progression in patients with atrial fibrillation: A systematic review. Clin Res Cardiol 106:813–823. 10.1007/s00392-017-1123-0 Castro-García JM, Arenas-Jiménez JJ, Adarve-Castro A, Trigueros-Buil H, Garfias-Baladrón MJ, Ureña-Vacas A (2023) Factores de riesgo clínicos y radiológicos para recurrencia de fibrilación auricular tras la ablación de venas pulmonares. Radiología. 10.1016/j.rx.2023.06.008 Mahajan R, Lau D, Brooks A et al (2021) Atrial fibrillation and obesity. J Am Coll Cardiol EP 7:630–641. 10.1016/j.jacep.2020.11.015 Chang SL, Tuan TC, Tai CT et al (2009) Comparison of outcome in catheter ablation of atrial fibrillation in patients with versus without the metabolic syndrome. Am J Cardiol 103:67–72. 10.1016/j.amjcard.2008.08.042 Cai L, Yin Y, Ling Z et al (2013) Predictors of late recurrence of atrial fibrillation after catheter ablation. Int J Cardiol 164:82–87. 10.1016/j.ijcard.2011.06.094 Wang TJ, Parise H, Levy D et al (2004) Obesity and the risk of new-onset atrial fibrillation. JAMA 292:2471–2477. 10.1001/jama.292.20.2471 Wong CX, Abed HS, Molaee P et al (2011) Pericardial fat is associated with atrial fibrillation severity and ablation outcome. J Am Coll Cardiol 57(17):1745–1751. 10.1016/j.jacc.2010.11.045 Tsao HM, Hu WC, Wu MH et al (2011) Quantitative analysis of quantity and distribution of epicardial adipose tissue surrounding the left atrium in patients with atrial fibrillation and effect of recurrence after ablation. Am J Cardiol 107(10):1498–1503. 10.1016/j.amjcard.2011.01.027 Nagashima K, Okumura Y, Watanabe I et al (2011) Association between epicardial adipose tissue volumes on 3-dimensional reconstructed CT images and recurrence of atrial fibrillation after catheter ablation. Circ J 75(11):2559–2565. 10.1253/circj.cj-11-0554 Kocyigit D, Gurses KM, Yalcin MU et al (2015) Periatrial epicardial adipose tissue thickness is an independent predictor of atrial fibrillation recurrence after cryoballoon-based pulmonary vein isolation. J Cardiovasc Comput Tomogr 9(4):295–302. 10.1016/j.jcct.2015.03.011 Nogami K, Sugiyama T, Kanaji Y et al (2021) Association between pericoronary adipose tissue attenuation and outcome after second-generation cryoballoon ablation for atrial fibrillation. Br J Radiol 94:20210361. 10.1259/bjr.20210361 Goldenberg GR, Hamdan A, Barsheshet A et al (2021) Epicardial fat and the risk of atrial tachy-arrhythmia recurrence post pulmonary vein isolation: a computed tomography study. Int J Cardiovasc Imaging 37:2785–2790. 10.1007/s10554-021-02244-w El Mahdiui M, Simon J, Smit JM et al (2021) Posterior left atrial adipose tissue attenuation assessed by computed tomography and recurrence of atrial fibrillation after catheter ablation. Circ Arrhythm Electrophysiol 14(7):e009135. 10.1161/circep.120.009135 Jian B, Li Z, Wang J, Zhang C (2022) Correlation analysis between heart rate variability, epicardial fat thickness, visfatin and AF recurrence post radiofrequency ablation. BMC Cardiovasc Disord 22:65. 10.1186/s12872-022-02496-x Samanta R, Houbois CP, Massin SZ, Seidman M, Wintersperger BJ, Chauhan VS (2021) Interatrial septal fat contributes to interatrial conduction delay and atrial fibrillation recurrence following ablation. Circ Arrhythm Electrophysiol 14(8):e010235. 10.1161/circep.121.010235 Ciuffo L, Nguyen H, Marques MD et al (2019) Periatrial fat quality predicts atrial fibrillation ablation outcome. Circ Cardiovasc Imaging 12(4):e008764. 10.1161/circimaging.118.008764 Maeda M, Oba K, Yamaguchi S et al (2018) Usefulness of epicardial adipose tissue volume to predict recurrent atrial fibrillation after radiofrequency catheter ablation. Am J Cardiol 122(10):1694–1700. 10.1016/j.amjcard.2018.08.005 Masuda M, Mizuno H, Enchi Y et al (2015) Abundant epicardial adipose tissue surrounding the left atrium predicts early rather than late recurrence of atrial fibrillation after catheter ablation. J Interv Card Electrophysiol 44(1):31–37. 10.1007/s10840-015-0031-3 Stojanovska J, Kazerooni EA, Sinno M et al (2015) Increased epicardial fat is independently associated with the presence and chronicity of atrial fibrillation and radiofrequency ablation outcome. Eur Radiol 25(8):2298–2309. 10.1007/s00330-015-3643-1 Kim TH, Park J, Park JK et al (2014) Pericardial fat volume is associated with clinical recurrence after catheter ablation for persistent atrial fibrillation, but not paroxysmal atrial fibrillation: An analysis of over 600 patients. Int J Cardiol 176:841–846. 10.1016/j.ijcard.2014.08.015 Ma Y, Zhang D, Xu J et al (2023) Explainable machine learning model reveals its decision-making process in identifying patients with paroxysmal atrial fibrillation at high risk for recurrence after catheter ablation. BMC Cardiovasc Disord 23:91. 10.1186/s12872-023-03087-0 Zhou X, Nakamura K, Sahara N et al (2022) Deep Learning-Based Recurrence Prediction of Atrial Fibrillation After Catheter Ablation. Circ J 86(2):299–308. 10.1253/circj.CJ-21-0622 Tang S, Razeghi O, Kapoor R et al (2022) Machine Learning–Enabled Multimodal Fusion of Intra-Atrial and Body Surface Signals in Prediction of Atrial Fibrillation Ablation Outcomes. Circ Arrhythm Electrophysiol, 15. Published online July 22, 2022. 10.1161/CIRCEP.122.010850 Baalman SWE, Lopes RR, Ramos LA, Neefs J et al (2021) Prediction of Atrial Fibrillation Recurrence after Thoracoscopic Surgical Ablation. Using Mach Learn Techniques Diagnostics 11(10):1787. 10.3390/diagnostics11101787 Roney CH, Sim I, Yu J et al (2022) Circulation: Arrhythmia Electrophysiol 15(2). 10.1161/CIRCEP.121.010253 . Predicting Atrial Fibrillation Recurrence by Combining Population Data and Virtual Cohorts of Patient-Specific Left Atrial Models Sepehri Shamloo A, Dagres N, Dinov B et al (2019) Is epicardial fat tissue associated with atrial fibrillation recurrence after ablation? A systematic review and meta-analysis. Int J Cardiol Heart Vasc 26:22:132–138. 10.1016/j.ijcha.2019.01.003 Huber AT, Fankhauser S, Chollet L et al (2022) The relationship between enhancing left atrial adipose tissue at CT and recurrent atrial fibrillation. Radiology 305:56–65. 10.1148/radiol.212644 Potpara TS, Mujovic N, Sivasambu B et al (2019) Validation of the MB-LATER score for prediction of late recurrence after catheter-ablation of atrial fibrillation. Int J Cardiol. 1:276:130–135. 10.1016/j.ijcard.2018.08.018 . Epub 2018 Aug 11. PMID: 30126656 Sanhoury M, Moltrasio M, Tundo F et al (2017) Predictors of arrhythmia recurrence after balloon cryoablation of atrial fibrillation: the value of CAAP-AF risk scoring system. J Interv Card Electrophysiol 49(2):129–135. 10.1007/s10840-017-0248-4 Epub 2017 Apr 18. PMID: 28417287 Jud FN, Obeid S, Duru F, Haegeli LM (2019) A novel score in the prediction of rhythm outcome after ablation of atrial fibrillation: The SUCCESS score. Anatol J Cardiol 21(3):142–149. 10.14744/AnatolJCardiol.2018.76570 Saglietto A, Gaita F, Blomstrom-Lundqvist C et al (2023) AFA-Recur: an ESC EORP AFA-LT registry machine-learning web calculator predicting atrial fibrillation recurrence after ablation. Europace 25(1):92–100. 10.1093/europace/euac145 Zhou X, Nakamura K, Sahara N et al (2022) Deep Learning-Based Recurrence Prediction of Atrial Fibrillation After Catheter Ablation. Circ J 86:2:299–308. 10.1253/circj.CJ-21-0622 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4577588","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":321018706,"identity":"42d4485c-73bc-44fa-916e-14c65256e5cd","order_by":0,"name":"José Miguel Castro-García","email":"data:image/png;base64,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","orcid":"","institution":"Saint Michael´s Hospital","correspondingAuthor":true,"prefix":"","firstName":"José","middleName":"Miguel","lastName":"Castro-García","suffix":""},{"id":321018709,"identity":"a81f0b0f-11b6-4c99-a2e4-6f29a28bce03","order_by":1,"name":"María Javiera Garfias-Baladrón","email":"","orcid":"","institution":"General Universitary Hospital of Alicante","correspondingAuthor":false,"prefix":"","firstName":"María","middleName":"Javiera","lastName":"Garfias-Baladrón","suffix":""},{"id":321018710,"identity":"ad2942c2-2ae4-4e3f-9727-c3f25555b8c0","order_by":2,"name":"Antonio Adarve-Castro","email":"","orcid":"","institution":"Virgin of Victory Universitary Hospital","correspondingAuthor":false,"prefix":"","firstName":"Antonio","middleName":"","lastName":"Adarve-Castro","suffix":""},{"id":321018711,"identity":"0a47a258-db8f-49f0-9a63-fa9282daa8c2","order_by":3,"name":"Helena Trigueros-Buil","email":"","orcid":"","institution":"General Universitary Hospital of Alicante","correspondingAuthor":false,"prefix":"","firstName":"Helena","middleName":"","lastName":"Trigueros-Buil","suffix":""},{"id":321018712,"identity":"54f2321f-babe-4ace-8fc8-358b16be061f","order_by":4,"name":"Álvaro Palazón-Ruíz","email":"","orcid":"","institution":"General Universitary Hospital of Alicante","correspondingAuthor":false,"prefix":"","firstName":"Álvaro","middleName":"","lastName":"Palazón-Ruíz","suffix":""},{"id":321018713,"identity":"688cc864-9347-4253-af10-7ab63fd5346b","order_by":5,"name":"Almudena Ureña-Vacas","email":"","orcid":"","institution":"General Universitary Hospital of Alicante","correspondingAuthor":false,"prefix":"","firstName":"Almudena","middleName":"","lastName":"Ureña-Vacas","suffix":""}],"badges":[],"createdAt":"2024-06-13 16:47:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4577588/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4577588/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":59597732,"identity":"1b9c8e3a-f89f-4788-9df7-c5851cfc37f7","added_by":"auto","created_at":"2024-07-03 16:15:42","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":394903,"visible":true,"origin":"","legend":"\u003cp\u003eVariables Obtained from Radiological Analysis: Images labeled with 'A' in the top row display cases of patients who experienced recurrence, characterized by larger volumes of epicardial adipose tissue, periatrial adipose tissue, interatrial adipose tissue, and left atrial volume. Conversely, images marked with 'B' in the bottom row depict patients without recurrence, presenting smaller volumes of this variables.\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4577588/v1/0eb9cfb9fa5b48aff32f73f7.jpg"},{"id":59597731,"identity":"4c9d1e05-9b48-4e5a-880b-f3cf0c43d08c","added_by":"auto","created_at":"2024-07-03 16:15:42","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":296741,"visible":true,"origin":"","legend":"\u003cp\u003eVariables weights for all the predicted cases assigned by SHAP analysis for both models (A and B). Each dot represents the influence of a specific feature on a prediction. Variables within the model are represented on the Y-axis and SHAP values (weights) on the X-axis. Positive values in the X axis increase the likelihood of recurrence and negative values increase the likelihood of stability. Colors indicate the variable value for each patient in which the prediction was made: blue for lower values and red for higher values. The red dots for the variables that had two different classes (AF type, comorbidities, sex) represent the patients who had persistent AF, dyslipidemia, hypertension and were male.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4577588/v1/51cadc4d318eb67a1ac6c40c.jpg"},{"id":87309355,"identity":"8e62f8f3-e2ea-4ade-a07b-25d5d8aec030","added_by":"auto","created_at":"2025-07-22 14:38:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1535958,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4577588/v1/5a52ac91-dbfa-46d0-831d-0b573b0d3e32.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Prediction of Atrial Fibrillation recurrence after catheter ablation. An explicative machine learning approach incorporating epicardial adipose tissue volume.","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAtrial fibrillation (AF) is the most common arrhythmia globally, with an incidence that increases in parallel with age, reaching 10\u0026ndash;17% in individuals over 78 years old (1) and an estimated prevalence of 37,574\u0026nbsp;million cases worldwide until 2017, a number that might increase by \u0026gt;\u0026thinsp;60% in 2050 (2). The clinical importance of its diagnosis and treatment lies in both preventing embolic events through anticoagulation and controlling heart rhythm or rate through either permanent pharmacological treatment, cardioversion or interventional ablative techniques.\u003c/p\u003e \u003cp\u003eDue to frequent relapses after electrical or drug-induced cardioversion and the increase of interventional cardiology procedures, catheter ablation of pulmonary veins has become a widely accepted treatment choice for symptomatic AF cases resistant to medications [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], attaining success rates between 59.9\u0026ndash;65.4% within 12 months after radiofrequency or cryoablation [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Certain risk factors, such as the type of AF, have been associated with lower success rates after ablation. In a study by Deng et al. [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], the success rate was 70% for paroxysmal AF patients and dropped to 50% for those with persistent AF.\u003c/p\u003e \u003cp\u003eAligned with the above, a longer duration of the arrhythmia has been linked to a larger left atrial [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] volume, which is an independent risk factor for recurrence after ablation [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], probably due to the accompanying cardiac remodeling, increased fibrosis and atypical microcircuits development. Other factors such as metabolic syndrome, obesity and advanced age have also been associated with a decreased rate of success [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn the last decade, the relevance of the quality and quantity of epicardial adipose tissue has been linked to AF recurrence after an ablation procedure, with a higher risk found in patients with greater amounts and density of epicardial (EAT), periatrial (PAT) and interatrial adipose tissue (IAT) based on computed tomography (CT) scans and other techniques [\u003cspan additionalcitationids=\"CR12 CR13 CR14 CR15 CR16 CR17 CR18 CR19 CR20 CR21 CR22 CR23\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDue to the multiple risk factors that can influence the outcome following catheter ablation, the complex data management, and the need of a recurrence prediction model, some artificial intelligence models based on supervised learning have been created previously.\u003c/p\u003e \u003cp\u003eSome studies [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] have included blood markers, other clinical and radiological variables, as well as electrocardiographic [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] and electrophysiological [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], variables resulting in some complex models.\u003c/p\u003e \u003cp\u003eWhile cardiac adipose tissue volumes are a known risk factor for AF recurrence, their inclusion for analysis using machine learning techniques has not been performed before and could offer a refined approach to the complex interplay of these novel risk factors. Machine learning models address better the analysis of non-linear patterns and interactions that may not be apparent in traditional analyses. This study applies these advanced analytical capabilities to refine the predictive accuracy for AF recurrence after catheter ablation, incorporating volumetric analysis of the left atrium, interatrial, periatrial and epicardial adipose tissue.\u003c/p\u003e"},{"header":"Material and methods","content":"\u003cp\u003e The institutional review board approved this study (under the registry PI2023-045) according to human rights declarations and regulations. Due to its retrospective design and reliance on medical record review, informed consent was waived.\u003c/p\u003e \u003cp\u003eThis retrospective cohort study included all consecutive patients with atrial fibrillation (AF) who underwent radiofrequency or cryoablation of pulmonary veins after a contrast CT scan between 2017 and 2021. CT scans with significant motion artifacts or insufficient left atrial enhancement were excluded.\u003c/p\u003e \u003cp\u003eThe first cohort included patients who experienced recurrence within 18 months after ablation, while the second cohort included those that maintained sinus rhythm during that period. Demographic and clinical data, including age, gender, body mass index (BMI), comorbidities (type 2 diabetes mellitus, dyslipidemia, and systemic hypertension) and AF type (paroxysmal or persistent) were collected. The presence of sinus rhythm or AF recurrence was documented during the 3-, 6-, 12- and 18-months follow-up visits.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eComputed Tomography (CT) Scans and Cardiac Adipose Tissue Analysis:\u003c/h2\u003e \u003cp\u003eCT scans were conducted using a 64-slice detector scanner (Phillips Brilliance CT). Images were acquired through a retrospective electrocardiographically synchronized acquisition protocol with dose modulation at 120 kV, 64 \u0026times; 0.625 mm collimation, and 0.35 s rotation time. A standard dose of 80 ml iodinated contrast with a concentration of 400 mg/ml was injected at 4 ml/s, followed by 30 ml saline at the same rate of injection.\u003c/p\u003e \u003cp\u003eTracking bolus technique was used, placing the region of interest in the left atrium with a trigger threshold of 130 Hounsfield units (HU). CT datasets were reconstructed with a slice thickness of 1.5 mm.\u003c/p\u003e \u003cp\u003eLeft atrial volume (LAV), epicardial (EAT), periatrial (PAT) and interatrial adipose tissue (IAT) volume were quantified during the cardiac phase that corresponded to the 75% of the R-R interval by manual semi-automatic segmentation performed with Philips IntelliSpace software 12.1 segmentation and subsegmentation tools (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAll the CT scans were evaluated and segmented by a radiologist from the cardiac imaging department of a third level hospital (JMCG). EAT volume was traced from the top of the aortic arch to the diaphragm, PAT was quantified from the roof of the coronary sinus to the roof of the left atrium and IAT was obtained by tracing the interatrial septum area. Subsegmentation of adipose tissue was based on an attenuation range between \u0026minus;\u0026thinsp;195 and \u0026minus;\u0026thinsp;15 HU.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eThe normality of all quantitative variables was assessed using the Kolmogorov-Smirnov test. Parametrically distributed variables were described with mean and standard deviation, while non-parametric variables were described with median and quartiles. Qualitative variables were described by absolute and relative frequency. An initial univariate analysis compared AF recurrence and risk factors, including age, gender, BMI, AF type (paroxysmal or persistent), LAV, EAT, PAT and IAT volume.\u003c/p\u003e \u003cp\u003eThe chi-square test was used for qualitative variables, Student's t-test for parametric quantitative variables, and Mann-Whitney U test for non-parametric quantitative variables. Variables with p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 were considered statistically significant. Statistical analysis was performed using Statistical Package for Social Sciences (SPSS) version 22.0.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eFeature Selection:\u003c/h2\u003e \u003cp\u003eFeature selection was conducted through exhaustive automated process that tested all possible combinations of clinical and radiological variables to determine which contributed most significantly to the model's accuracy. This process was performed using Python 3.12 and the scikit-learn library (version 1.1).\u003c/p\u003e \u003cp\u003eTwo different models were created, the first one (A) including both clinical and radiological variables and the second one (B) including only radiological variables.\u003c/p\u003e \u003cp\u003eThe variables included in model A were: age, gender, BMI, AF type (paroxysmal or persistent), left atrial volume, PAT, IAT and EAT volume. Model B included left atrial, PAT, IAT and EAT volume.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eData Preprocessing:\u003c/h2\u003e \u003cp\u003eNo patients with missing information were included in this study. Before training the model, the distribution of the training group was balanced by using the technique of random under-sampling. This involved reducing the number of instances in the overrepresented class to match the number of instances in the underrepresented class. These was conducted by applying the RandomUnderSampler method from imblearn library.\u003c/p\u003e \u003cp\u003eFinally, feature scaling was performed in the dataset in order to ensure that all features contributed evenly to the model's performance. The MinMaxScaler function was applied to both the training and test groups, transforming the feature values to a common scale within the range of 0 to 1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eModel training:\u003c/h2\u003e \u003cp\u003eTo create the training set, the preprocessed and balanced data was divided by using the Train_test_split function which randomly divided the whole sample in approximately 80% (54 patients) for training and 20% (15 patients) for testing.\u003c/p\u003e \u003cp\u003eThree different ML techniques were used to train the model by using the data of the training set. The first one was a Multi-layer Perceptron (MLP) classifier configured with a single hidden layer comprising 50 units and utilizing ReLU activation function. The Adam solver was used for optimization, and the model underwent a maximum of 2500 iterations during training. This training was evaluated by using a 5-fold cross-validation and calculating the average accuracy, precision, recall, and F1-score within the training group.\u003c/p\u003e \u003cp\u003eThe second technique employed was Logistic Regression with balanced class weights and the third was Gaussian Na\u0026iuml;ve Bayes. In both cases performance evaluation was carried out by calculating the average accuracy, precision, recall, and F1-score after a 5-folds cross-validation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eModel Evaluation and SHAP Analysis:\u003c/h2\u003e \u003cp\u003eAfter the training and validation stages, both models were tested by calculating the Receiver Operating Characteristic (ROC) curve, accuracy, precision, recall, and F1 score (Table\u0026nbsp;3). Furthermore, SHAP (SHapley Additive exPlanations) analysis was performed, in order to clarify the importance of each variable within all the techniques applied to the two different models (A and B).\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eA total of 85 patients were collected initially. From these, only 70 patients had a complete follow-up and adequate Computed Tomography (CT) scan quality. 30 (42%) were women and 40 (58%) men. The total patients that experienced recurrence after catheter ablation were 20 (29%), and the ones that did not recur during the 18 months follow-up were 49 (71%) (Tables\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e, \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e). From the total of patients that experienced recurrence, 7 (35%) had paroxysmal AF and 13 (65%) had persistent AF, with a significant higher risk of recurrence (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) being observed for those with persistent AF (OR 1.99; IC: 1.19\u0026ndash;3.33).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eClinical and radiological characteristics of all the included patients.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003eVariables\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRecurrence\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNo recurrence\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eClinical variables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20 (28.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e50 (71.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e60.3 (+/-8.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e57 (+/-8.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e30 (75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (33.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e20 (66.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28.91 (+/-6.38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e30.39 (+/-5.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDM2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6 (46.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7 (53.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14 (24.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e43 (75.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (52.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (47.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27 (55.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e22 (44.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDyslipidemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (27%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27 (73%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (30.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23 (69.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAF type\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePersistent\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (44.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e16 (55.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eParoxysmal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7 (17.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34 (82.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eType of procedure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRadiofrequency\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15 (27.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e39 (72.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCryoablation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5 (33.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11 (66.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eTreatment\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBeta-blocker\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (32.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28 (68.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7 (24.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e22 (75.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCalcium antagonist\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5 (41.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7 (58.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15 (25.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e43 (74.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOral anticoagulant\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13 (36.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23 (63.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26 (76.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8 (23.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eACEi / ARA*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (34.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19 (65.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10 (24.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31 (75.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eRadiological variables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLeft atrial volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e133.19 (105.54-168.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e88.32 (72.85-109.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEpicardial adipose tissue volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e265.93 (200.63-312.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e205.82 (117.59-288.94)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePeriatrial adipose tissue volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14.68 (11.08\u0026ndash;23.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8.96 (4.05\u0026ndash;11.99)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInteratrial adipose tissue volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.25 (1.04\u0026ndash;4.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.36 (0.62\u0026ndash;2.63)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e*ACEi: Angiotensin-Converting Enzyme inhibitors, ARA: Angiotensin Receptor Antagonists\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eUnivariate analysis of atrial fibrillation recurrence risk for all the clinical and radiological variables.\u0026nbsp;\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003eVariables\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOR (IC)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eClinical variables\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.63 (0.22\u0026ndash;1.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDM2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.57 (0.73\u0026ndash;8.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHypertension\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.996 (0.35\u0026ndash;2.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDyslipidemia\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.81 (0.28\u0026ndash;2.31)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAF type\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePersistent\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.99 (1.19\u0026ndash;3.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;\u0026thinsp;0.05\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eParoxysmal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eType of procedure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRadiofrequency\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.3 (0.38\u0026ndash;4.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCryoablation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eTreatment\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBeta-blocker\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.71 (0.24\u0026ndash;2.12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCalcium antagonist\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.42 (0.11\u0026ndash;1.57)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOral anticoagulant\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.476 (0.16\u0026ndash;1.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eACEi / ARA*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.63 (0.22\u0026ndash;1.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003eRadiological variables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLeft atrial volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;\u0026thinsp;0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEpicardial adipose tissue volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026gt;\u0026thinsp;0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePeriatrial adipose tissue volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;\u0026thinsp;0.01\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInteratrial adipose tissue volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026lt;\u0026thinsp;0.05\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e*ACEi: Angiotensin-Converting Enzyme inhibitors, ARA: Angiotensin Receptor Antagonists\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eLeft atrium (LAV), interatrial (IAT), periatrial (PAT) and epicardial adipose tissue (EAT) volume differences between both groups.\u003c/p\u003e\n\u003cp\u003eCT imaging revealed volumetric differences between patients with and without AF recurrence, particularly in left atrial volume (LAV), periatrial adipose tissue (PAT), and interatrial adipose tissue (IAT) volumes (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). Patients with recurrence exhibited significantly larger LAV (median [IQR]: 133.19 [105.54-168.62] mL) compared to those without recurrence (88.32 [72.85-109.15] mL, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Similarly, PAT and IAT volumes were markedly higher in patients who experienced recurrence (PAT: 14.68 [11.08\u0026ndash;23.79] vs. 8.96 [4.05\u0026ndash;11.99] cm\u0026sup3;, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; IAT: 2.25 [1.04\u0026ndash;4.25] vs. 1.36 [0.62\u0026ndash;2.63] cm\u0026sup3;, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003eModel Performance and Clinical Relevance:\u003c/h2\u003e\n \u003cp\u003eDue to the total sample of 70 patients being divided into 55 for training and 15 for testing the model, and the application of a class balancing technique, learning was based on the use of 5-fold cross-validation with different patients from those 55 in each fold. Specifically, the model was trained on 42 patients who did not experience recurrence and 13 who did, testing it on 8 patients without recurrence and 7 who experienced recurrence.\u003c/p\u003e\n \u003cp\u003eRegarding models\u0026rsquo; training and validation, the highest values of average precision, accuracy, f1 score and recall after 5-fold cross-validation were obtained for model A, using the Logistic Regression (LR) technique. When evaluating the models in the testing groups, the accuracies of model A using a neural network (NN), Na\u0026iuml;ve Bayes (NB), and Logistic Regression (LR) were 0.86, 0.66 and 0.86, respectively (Table\u0026nbsp;3), while the AUCs were 0.91, 0.87 and 0.92, respectively (Table\u0026nbsp;3). The highest ability to predict patients that experienced recurrence was also achieved by LR, correctly identifying all the patients that experienced recurrence (Table\u0026nbsp;3).\u003c/p\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable\u0026nbsp;3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eModel A and B validation and performance in the test group.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ctbody\u003e\n \u003ctr style=\"height: 35px;\"\u003e\n \u003ctd style=\"width: 443.086px; height: 35px;\" colspan=\"10\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel A. Radiological and clinical variables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 35px;\"\u003e\n \u003ctd style=\"width: 68px; height: 35px;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd style=\"width: 164.086px; height: 35px;\" colspan=\"4\"\u003e\n \u003cp\u003e\u003cem\u003eTraining group\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 201px; height: 35px;\" colspan=\"5\"\u003e\n \u003cp\u003e\u003cem\u003eTest group\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 48px;\"\u003e\n \u003ctd style=\"width: 68px; height: 48px;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 48px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 48px;\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 48px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 48px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC*\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 48px;\"\u003e\n \u003ctd style=\"width: 68px; height: 48px;\"\u003e\n \u003cp\u003eNeural Network\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 48px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.566\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 48px;\"\u003e\n \u003cp\u003e0.613\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.866\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 48px;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 48px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.911\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 35px;\"\u003e\n \u003ctd style=\"width: 68px; height: 35px;\"\u003e\n \u003cp\u003eNa\u0026iuml;ve Bayes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 35px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 35px;\"\u003e\n \u003cp\u003e0.566\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 35px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 35px;\"\u003e\n \u003cp\u003e0.613\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 35px;\"\u003e\n \u003cp\u003e0.666\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 35px;\"\u003e\n \u003cp\u003e0.666\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 35px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 35px;\"\u003e\n \u003cp\u003e0.615\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 35px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.875\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 48.0118px;\"\u003e\n \u003ctd style=\"width: 68px; height: 48.0118px;\"\u003e\n \u003cp\u003eLogistic Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.566\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.613\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.866\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.777\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 48.0118px;\"\u003e\n \u003cp\u003e1.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48.0118px;\"\u003e\n \u003cp\u003e0.875\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 48.0118px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.928\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 35px;\"\u003e\n \u003ctd style=\"width: 443.086px; height: 35px;\" colspan=\"10\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel B. Radiological variables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 35px;\"\u003e\n \u003ctd style=\"width: 68px; height: 35px;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd style=\"width: 164.086px; height: 35px;\" colspan=\"4\"\u003e\n \u003cp\u003e\u003cem\u003eTraining group\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 201px; height: 35px;\" colspan=\"5\"\u003e\n \u003cp\u003e\u003cem\u003eTest group\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 48px;\"\u003e\n \u003ctd style=\"width: 68px; height: 48px;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 48px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 48px;\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 48px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 48px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC*\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 48px;\"\u003e\n \u003ctd style=\"width: 68px; height: 48px;\"\u003e\n \u003cp\u003eNeural Network\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 48px;\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003e0.733\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 48px;\"\u003e\n \u003cp\u003e0.693\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 48px;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 48px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.875\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 35px;\"\u003e\n \u003ctd style=\"width: 68px; height: 35px;\"\u003e\n \u003cp\u003eNa\u0026iuml;ve Bayes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 35px;\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 35px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 35px;\"\u003e\n \u003cp\u003e0.733\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 35px;\"\u003e\n \u003cp\u003e0.693\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 35px;\"\u003e\n \u003cp\u003e0.666\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 35px;\"\u003e\n \u003cp\u003e0.666\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 35px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 35px;\"\u003e\n \u003cp\u003e0.615\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 35px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.839\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr style=\"height: 48px;\"\u003e\n \u003ctd style=\"width: 68px; height: 48px;\"\u003e\n \u003cp\u003eLogistic Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 48px; height: 48px;\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003e0.733\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34.0857px; height: 48px;\"\u003e\n \u003cp\u003e0.693\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.733\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px; height: 48px;\"\u003e\n \u003cp\u003e0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 34px; height: 48px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 33px; height: 48px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 36px; height: 48px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.875\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eSHAP Analysis:\u003c/h2\u003e\n \u003cp\u003eThe results obtained after applying the SHAP analysis were similar for the 3 ML techniques for model A. The 3 most important features using the NN and NB techniques were LAV, PFT and IFT volumes. The Logistic Regression (LR), while aligning with the importance of the LAV and PFT volumes, diverged by identifying the type of atrial fibrillation (AF type) as a significant predictor, indicating LR might be particularly effective in utilizing categorical data to improve its predictive accuracy (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eIn Model B, the LAV and PFT volumes remained the most influential features in predicting outcomes using the NN and LR methods. However, the NB method\u0026apos;s emphasis on the LAV and IFT volumes, as opposed to PFT. The consistent appearance of LAV and PFT volumes as top features in the majority of models reinforces their relevance and higher predictive utility (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study presents a novel approach for atrial fibrillation (AF) recurrence prediction after catheter ablation, focusing on the role of epicardial adipose tissue volumes along with traditional clinical parameters. Our results demonstrate the potential of integrating radiological markers with clinical variables to improve the accuracy of predicting AF recurrence.\u003c/p\u003e \u003cp\u003eA meta-analysis conducted in July 2018 highlighted the importance of left atrium PAT and EFT volumes as predictive markers (30). Other studies have linked increased volume and density of epicardial [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], interatrial [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], and periatrial [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] adipose tissue with AF recurrence. In our study, only IAT and PAT volumes were significantly associated with recurrence, and although not statistically significant the presence of a large volume of EAT volume was the fifth most important variable for AF recurrence prediction when introduced into the SHAP explicative analysis using the Naive Bayes technique.\u003c/p\u003e \u003cp\u003eThe task of effectively predicting recurrence after AF ablation continues to be a significant challenge. Existing prognostic models for AF recurrence have a high variability in model performance, highlighting the complexity and the need for robust validation and calibration of predictive models.\u003c/p\u003e \u003cp\u003eA previous score (MB-LATER) validated by Potpara et al. in 2018 [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] on 226 patients for late recurrence prediction showed a good performance (AUC: 0.62 [95% CI: 0.54\u0026ndash;0.69], p\u0026thinsp;=\u0026thinsp;0.003) for AF recurrence prediction after more 12 months. This score includes gender, presence of bundle branch block, left atrial diameter\u0026thinsp;\u0026gt;\u0026thinsp;46 mm, type of AF, recurrence during the blanking period (first 3 months), and preablation history of persistent AF.\u003c/p\u003e \u003cp\u003eOther scores such as CAAP-AF, APPLE and SUCCESS have assessed the predictive value of different variables. CAAP-AF, developed by Winkle et al (33) includes variables such as sex, age, type of AF, Left atrial diameter (LAD), coronary artery disease and number of antiarrhythmic drugs failed. This scored has been evaluated in a previous study that included 283 patients with AF, reaching a sensitivity and specificity of 64% and 68%.\u003c/p\u003e \u003cp\u003eThe APPLE score, encompassing age, type of AF, estimated glomerular filtration rate (eGFR), LAD, and left ventricular ejection fraction (LVEF) in first and/or repeat ablation populations achieved an AUC of 0.62 in a previous study involving 192 patients [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]​​. This research also introduced the SUCCESS score, enhancing the APPLE score by adding a point for each prior ablation, which showed improvement in receiver operating characteristic analysis, reaching 0.657.\u003c/p\u003e \u003cp\u003eFrom the previous research analyzing the association between the epicardial adipose tissue volume and density, we could only find one (15) incorporating the pericoronary fat tissue average density into a model. This radiological and clinical model included age, sex, body mass index, atrial fibrillation (AF) type, NT- pro-BNP level, left atrial volume index, left ventricular end- diastolic dimension, and early AF recurrence obtaining and AUC of 0.726. This study employed regression techniques to predict outcomes in a group of patients.\u003c/p\u003e \u003cp\u003eThe attempt to predict recurrence in patients with atrial fibrillation has also been explored using supervised Machine Learning (ML) techniques. These techniques tend to fit better to variables that have a non-linear interaction between them compared to statistical techniques and allow a more refined tunning of the hyperparameters used to train a model.\u003c/p\u003e \u003cp\u003eIn a previous study, derived from the ESCEHRA Atrial Fibrillation Ablation Long-Term Registry (AFA-LT) a machine-learning scoring system was created, in order to predict recurrence after 1 year and included a total of 3128 patients [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. By evaluating four machine learning techniques (decision tree, random forest, AdaBoost, and k-nearest neighbour) a final model using the random forest technique showed an AUC 0.721, 95% confidence interval (CI: 0.680\u0026ndash;0.764) in the testing group by training the model with 19 clinical variables. The most important variables used in this model were left ventricular end-diastolic volume, eGFR, BMI, age, LA diameter and LVEF.\u003c/p\u003e \u003cp\u003eSimilar variables have been used in other studies, for example, Xue Zhou et al [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e] created a convolutional neural network trained with 4 predictors: LAV, left atrial appendage volume (LAAV), type of AF and N-terminal pro-BNP based on 310 patients with a 25-month follow up obtaining a C-index of 0.75 (0.72\u0026ndash;0.79) in the validation set and 0.76 (0.72\u0026ndash;0.79) in the training set. The LAV and LAAV were calculated from the pre-ablation CT scan. A limitation of this study was the high proportion of not collected variables, for example, the presence of previous ablations, which might be an important risk factor for recurrence.\u003c/p\u003e \u003cp\u003eTo our knowledge no previous studies have incorporated cardiac adipose tissue into a machine learning (ML) prediction strategy. Our study was conducted at a single center with a limited number of patients undergoing their first AF ablation. To address the challenge of working with a small dataset of 69 patients, our study employed several methodological strategies to ensure a good performance. Firstly, data normalization was conducted to ensure equitable contribution of all variables to the model learning process. Secondly, we applied 5-fold cross-validation during training, which maximizes the use of our limited dataset by ensuring all observations are utilized for both training and validation. This technique helps in preventing overfitting and enhances the performance on new data. Thirdly, random undersampling was utilized to address class imbalance within our dataset, reducing the size of the majority class to minimize bias towards the more frequent class and enabling more effective learning from the characteristics of both groups.\u003c/p\u003e \u003cp\u003eOnly patients with complete clinical information and a minimum follow-up of 18 months were included in our study. However, the discrimination found within our dataset needs further validation across diverse populations and other centers to confirm its broader applicability.\u003c/p\u003e \u003cp\u003eWe did not include the type of ablation procedure performed to the patients (pulmonary vein isolation, Cavo tricuspid isthmus ablation, SVC isolation, etc.) which is another limitation of our study. Finally, our study period was limited to the first 18 months after the ablation, for which our model was trained. The weight assignment to the included variables is based on the recurrence within this timeframe which could yield different results for periods extending beyond 18 months.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, our study, underscores the potential of combining clinical data with cardiac adipose tissue volumes acquired from the CT scan prior to ablation in order to enhance the prediction of AF recurrence. Our study highlights the importance of multimodal data integration and the need for ongoing development and validation of predictive models which could ultimately lead to more tailored therapeutic strategies, improving patient outcomes in AF.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cul\u003e\n\u003cli\u003eFunding\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eNo funding was received for this research.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eCompeting interests\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eAuthors' contributions (All authors contributed to the study conception and design).\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eJMCG: Data collection, statistical analysis and manuscript writing\u003c/p\u003e\n\u003cp\u003eAUV: Study design and manuscript revision\u003c/p\u003e\n\u003cp\u003eMJGB, APR and HTB: Data collection\u003c/p\u003e\n\u003cp\u003eAAC: Manuscript writing.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eEthics approval and consent to participate\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThis study was performed in line with the principles of the Declaration of Helsinki. The institutional review board of the General Universitary Hospital of Alicante, Spain approved this study (under the registry PI2023-045) according to human rights declarations and regulations.\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eConsent to participate\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eInformed consent was waived by the ethics committee.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eZoni-Berisso M, Lercari F, Carazza T, Domenicucci S (2014) Epidemiology of atrial fibrillation: European perspective. Clin Epidemiol 6:213\u0026ndash;220. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2147/clep.s47385\u003c/span\u003e\u003cspan address=\"10.2147/clep.s47385\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLippi G, Sanchis-Gomar F, Cervellin G (2021) Global epidemiology of atrial fibrillation: An increasing epidemic and public health challenge. Int J Stroke 16(2):217\u0026ndash;221. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/1747493019897870\u003c/span\u003e\u003cspan address=\"10.1177/1747493019897870\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCalkins H, Hindricks G, Cappato et al (2018) 2017 HRS/EHRA/ECAS/APHRS/SOLAECE expert consensus statement on catheter and surgical ablation of atrial fibrillation. Europace 20:e1\u0026ndash;e160. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/europace/eux274\u003c/span\u003e\u003cspan address=\"10.1093/europace/eux274\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen YH, Lu ZY, Xiang Y et al (2017) Cryoablation vs. radiofrequency ablation for treatment of paroxysmal atrial fibrillation: A systematic review and meta-analysis. Europace 19:784\u0026ndash;794. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/europace/euw330\u003c/span\u003e\u003cspan address=\"10.1093/europace/euw330\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng H, Bai Y, Shantsila A, Fauchier L, Potpara TS, Lip GYH (2017) Clinical scores for outcomes of rhythm control or arrhythmia progression in patients with atrial fibrillation: A systematic review. Clin Res Cardiol 106:813\u0026ndash;823. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00392-017-1123-0\u003c/span\u003e\u003cspan address=\"10.1007/s00392-017-1123-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCastro-Garc\u0026iacute;a JM, Arenas-Jim\u0026eacute;nez JJ, Adarve-Castro A, Trigueros-Buil H, Garfias-Baladr\u0026oacute;n MJ, Ure\u0026ntilde;a-Vacas A (2023) Factores de riesgo cl\u0026iacute;nicos y radiol\u0026oacute;gicos para recurrencia de fibrilaci\u0026oacute;n auricular tras la ablaci\u0026oacute;n de venas pulmonares. Radiolog\u0026iacute;a. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.rx.2023.06.008\u003c/span\u003e\u003cspan address=\"10.1016/j.rx.2023.06.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMahajan R, Lau D, Brooks A et al (2021) Atrial fibrillation and obesity. J Am Coll Cardiol EP 7:630\u0026ndash;641. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacep.2020.11.015\u003c/span\u003e\u003cspan address=\"10.1016/j.jacep.2020.11.015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChang SL, Tuan TC, Tai CT et al (2009) Comparison of outcome in catheter ablation of atrial fibrillation in patients with versus without the metabolic syndrome. Am J Cardiol 103:67\u0026ndash;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.amjcard.2008.08.042\u003c/span\u003e\u003cspan address=\"10.1016/j.amjcard.2008.08.042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai L, Yin Y, Ling Z et al (2013) Predictors of late recurrence of atrial fibrillation after catheter ablation. Int J Cardiol 164:82\u0026ndash;87. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijcard.2011.06.094\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcard.2011.06.094\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang TJ, Parise H, Levy D et al (2004) Obesity and the risk of new-onset atrial fibrillation. JAMA 292:2471\u0026ndash;2477. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.292.20.2471\u003c/span\u003e\u003cspan address=\"10.1001/jama.292.20.2471\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWong CX, Abed HS, Molaee P et al (2011) Pericardial fat is associated with atrial fibrillation severity and ablation outcome. J Am Coll Cardiol 57(17):1745\u0026ndash;1751. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacc.2010.11.045\u003c/span\u003e\u003cspan address=\"10.1016/j.jacc.2010.11.045\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTsao HM, Hu WC, Wu MH et al (2011) Quantitative analysis of quantity and distribution of epicardial adipose tissue surrounding the left atrium in patients with atrial fibrillation and effect of recurrence after ablation. Am J Cardiol 107(10):1498\u0026ndash;1503. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.amjcard.2011.01.027\u003c/span\u003e\u003cspan address=\"10.1016/j.amjcard.2011.01.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNagashima K, Okumura Y, Watanabe I et al (2011) Association between epicardial adipose tissue volumes on 3-dimensional reconstructed CT images and recurrence of atrial fibrillation after catheter ablation. Circ J 75(11):2559\u0026ndash;2565. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1253/circj.cj-11-0554\u003c/span\u003e\u003cspan address=\"10.1253/circj.cj-11-0554\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKocyigit D, Gurses KM, Yalcin MU et al (2015) Periatrial epicardial adipose tissue thickness is an independent predictor of atrial fibrillation recurrence after cryoballoon-based pulmonary vein isolation. J Cardiovasc Comput Tomogr 9(4):295\u0026ndash;302. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jcct.2015.03.011\u003c/span\u003e\u003cspan address=\"10.1016/j.jcct.2015.03.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNogami K, Sugiyama T, Kanaji Y et al (2021) Association between pericoronary adipose tissue attenuation and outcome after second-generation cryoballoon ablation for atrial fibrillation. Br J Radiol 94:20210361. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1259/bjr.20210361\u003c/span\u003e\u003cspan address=\"10.1259/bjr.20210361\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldenberg GR, Hamdan A, Barsheshet A et al (2021) Epicardial fat and the risk of atrial tachy-arrhythmia recurrence post pulmonary vein isolation: a computed tomography study. Int J Cardiovasc Imaging 37:2785\u0026ndash;2790. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10554-021-02244-w\u003c/span\u003e\u003cspan address=\"10.1007/s10554-021-02244-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEl Mahdiui M, Simon J, Smit JM et al (2021) Posterior left atrial adipose tissue attenuation assessed by computed tomography and recurrence of atrial fibrillation after catheter ablation. Circ Arrhythm Electrophysiol 14(7):e009135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/circep.120.009135\u003c/span\u003e\u003cspan address=\"10.1161/circep.120.009135\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJian B, Li Z, Wang J, Zhang C (2022) Correlation analysis between heart rate variability, epicardial fat thickness, visfatin and AF recurrence post radiofrequency ablation. BMC Cardiovasc Disord 22:65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12872-022-02496-x\u003c/span\u003e\u003cspan address=\"10.1186/s12872-022-02496-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSamanta R, Houbois CP, Massin SZ, Seidman M, Wintersperger BJ, Chauhan VS (2021) Interatrial septal fat contributes to interatrial conduction delay and atrial fibrillation recurrence following ablation. Circ Arrhythm Electrophysiol 14(8):e010235. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/circep.121.010235\u003c/span\u003e\u003cspan address=\"10.1161/circep.121.010235\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCiuffo L, Nguyen H, Marques MD et al (2019) Periatrial fat quality predicts atrial fibrillation ablation outcome. Circ Cardiovasc Imaging 12(4):e008764. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/circimaging.118.008764\u003c/span\u003e\u003cspan address=\"10.1161/circimaging.118.008764\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaeda M, Oba K, Yamaguchi S et al (2018) Usefulness of epicardial adipose tissue volume to predict recurrent atrial fibrillation after radiofrequency catheter ablation. Am J Cardiol 122(10):1694\u0026ndash;1700. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.amjcard.2018.08.005\u003c/span\u003e\u003cspan address=\"10.1016/j.amjcard.2018.08.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMasuda M, Mizuno H, Enchi Y et al (2015) Abundant epicardial adipose tissue surrounding the left atrium predicts early rather than late recurrence of atrial fibrillation after catheter ablation. J Interv Card Electrophysiol 44(1):31\u0026ndash;37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10840-015-0031-3\u003c/span\u003e\u003cspan address=\"10.1007/s10840-015-0031-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStojanovska J, Kazerooni EA, Sinno M et al (2015) Increased epicardial fat is independently associated with the presence and chronicity of atrial fibrillation and radiofrequency ablation outcome. Eur Radiol 25(8):2298\u0026ndash;2309. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00330-015-3643-1\u003c/span\u003e\u003cspan address=\"10.1007/s00330-015-3643-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim TH, Park J, Park JK et al (2014) Pericardial fat volume is associated with clinical recurrence after catheter ablation for persistent atrial fibrillation, but not paroxysmal atrial fibrillation: An analysis of over 600 patients. Int J Cardiol 176:841\u0026ndash;846. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijcard.2014.08.015\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcard.2014.08.015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa Y, Zhang D, Xu J et al (2023) Explainable machine learning model reveals its decision-making process in identifying patients with paroxysmal atrial fibrillation at high risk for recurrence after catheter ablation. BMC Cardiovasc Disord 23:91. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12872-023-03087-0\u003c/span\u003e\u003cspan address=\"10.1186/s12872-023-03087-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou X, Nakamura K, Sahara N et al (2022) Deep Learning-Based Recurrence Prediction of Atrial Fibrillation After Catheter Ablation. Circ J 86(2):299\u0026ndash;308. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1253/circj.CJ-21-0622\u003c/span\u003e\u003cspan address=\"10.1253/circj.CJ-21-0622\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTang S, Razeghi O, Kapoor R et al (2022) Machine Learning\u0026ndash;Enabled Multimodal Fusion of Intra-Atrial and Body Surface Signals in Prediction of Atrial Fibrillation Ablation Outcomes. Circ Arrhythm Electrophysiol, 15. Published online July 22, 2022. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCEP.122.010850\u003c/span\u003e\u003cspan address=\"10.1161/CIRCEP.122.010850\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaalman SWE, Lopes RR, Ramos LA, Neefs J et al (2021) Prediction of Atrial Fibrillation Recurrence after Thoracoscopic Surgical Ablation. Using Mach Learn Techniques Diagnostics 11(10):1787. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/diagnostics11101787\u003c/span\u003e\u003cspan address=\"10.3390/diagnostics11101787\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoney CH, Sim I, Yu J et al (2022) Circulation: Arrhythmia Electrophysiol 15(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCEP.121.010253\u003c/span\u003e\u003cspan address=\"10.1161/CIRCEP.121.010253\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Predicting Atrial Fibrillation Recurrence by Combining Population Data and Virtual Cohorts of Patient-Specific Left Atrial Models\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSepehri Shamloo A, Dagres N, Dinov B et al (2019) Is epicardial fat tissue associated with atrial fibrillation recurrence after ablation? A systematic review and meta-analysis. Int J Cardiol Heart Vasc 26:22:132\u0026ndash;138. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijcha.2019.01.003\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcha.2019.01.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuber AT, Fankhauser S, Chollet L et al (2022) The relationship between enhancing left atrial adipose tissue at CT and recurrent atrial fibrillation. Radiology 305:56\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1148/radiol.212644\u003c/span\u003e\u003cspan address=\"10.1148/radiol.212644\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePotpara TS, Mujovic N, Sivasambu B et al (2019) Validation of the MB-LATER score for prediction of late recurrence after catheter-ablation of atrial fibrillation. Int J Cardiol. 1:276:130\u0026ndash;135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijcard.2018.08.018\u003c/span\u003e\u003cspan address=\"10.1016/j.ijcard.2018.08.018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2018 Aug 11. PMID: 30126656\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSanhoury M, Moltrasio M, Tundo F et al (2017) Predictors of arrhythmia recurrence after balloon cryoablation of atrial fibrillation: the value of CAAP-AF risk scoring system. J Interv Card Electrophysiol 49(2):129\u0026ndash;135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10840-017-0248-4\u003c/span\u003e\u003cspan address=\"10.1007/s10840-017-0248-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003eEpub 2017 Apr 18. PMID: 28417287\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJud FN, Obeid S, Duru F, Haegeli LM (2019) A novel score in the prediction of rhythm outcome after ablation of atrial fibrillation: The SUCCESS score. Anatol J Cardiol 21(3):142\u0026ndash;149. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.14744/AnatolJCardiol.2018.76570\u003c/span\u003e\u003cspan address=\"10.14744/AnatolJCardiol.2018.76570\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaglietto A, Gaita F, Blomstrom-Lundqvist C et al (2023) AFA-Recur: an ESC EORP AFA-LT registry machine-learning web calculator predicting atrial fibrillation recurrence after ablation. Europace 25(1):92\u0026ndash;100. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/europace/euac145\u003c/span\u003e\u003cspan address=\"10.1093/europace/euac145\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou X, Nakamura K, Sahara N et al (2022) Deep Learning-Based Recurrence Prediction of Atrial Fibrillation After Catheter Ablation. Circ J 86:2:299\u0026ndash;308. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1253/circj.CJ-21-0622\u003c/span\u003e\u003cspan address=\"10.1253/circj.CJ-21-0622\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Atrial fibrillation, ablation techniques, machine learning, recurrence.","lastPublishedDoi":"10.21203/rs.3.rs-4577588/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4577588/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePurpose: Atrial fibrillation (AF) is a common arrhythmia with increasing prevalence and significant clinical impact. Catheter ablation has emerged as a treatment option for drug-resistant AF, with variable success rates. This study aimed to develop a machine learning-based predictive model incorporating interatrial, periatrial, and epicardial adipose tissue volumes to predict AF recurrence after pulmonary vein ablation.\u003c/p\u003e\n\u003cp\u003eMethods: This retrospective cohort study included patients who underwent a first ablation procedure between 2017 and 2022. Computed tomography (CT) scans were used to measure left atrial volume (LAV), periatrial (PAT), interatrial (IAT) and (EAT) epicardial adipose tissue volumes. Two models were created and trained under three machine learning techniques. Receiver Operating Characteristic (ROC) curve analysis, accuracy, precision, recall and F1-score were evaluated. SHapley Additive exPlanations (SHAP) analysis was also conducted.\u003c/p\u003e\n\u003cp\u003eResults: From the initial 85 patients, 69 with complete follow-up and CT scan quality were included. Persistent AF, increased left atrial, PAT and IAT volumes were significantly associated with recurrence. The model including clinical and radiological variables achieved accuracies of 0.86, 0.66, and 0.86 and AUCs of 0.91, 0.87, and 0.92 in the testing group by using MLP Classifier Neural Network, Naïve Bayes, and Logistic Regression, respectively. SHAP analysis emphasized the LAV, PAT volume and AF type for recurrence prediction.\u003c/p\u003e\n\u003cp\u003eConclusion: This study presents a machine learning explicative approach incorporating cardiac adipose tissue volumes for predicting AF post-ablation recurrence. The logistic regression model including clinical and radiological variables demonstrated the highest performance, highlighting the potential of using multimodal data for post-ablation recurrence prediction.\u003c/p\u003e","manuscriptTitle":"Prediction of Atrial Fibrillation recurrence after catheter ablation. An explicative machine learning approach incorporating epicardial adipose tissue volume.","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-03 16:15:37","doi":"10.21203/rs.3.rs-4577588/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"eb5a8083-a36e-4ee0-9972-30e1d80008ff","owner":[],"postedDate":"July 3rd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-07-22T14:38:15+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-03 16:15:37","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4577588","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4577588","identity":"rs-4577588","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}