Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Mar 1.
Published in final edited form as: Circ Arrhythm Electrophysiol. 2021 Feb 12;14(3):e009265. doi: 10.1161/CIRCEP.120.009265

Machine Learning-derived Fractal Features of Shape and Texture of the Left Atrium and Pulmonary Veins from Cardiac CT Scans are Associated with Risk of Recurrence of Atrial Fibrillation Post-ablation

Marjan Firouznia 1, Albert K Feeny 2, Michael A LaBarbera 2, Meghan McHale 3,4, Catherine Cantlay 4, Natalie Kalfas 4, Paul Schoenhagen 2,3,6, Walid Saliba 3, Patrick Tchou 3, John Barnard 5, Mina K Chung 2,3,4,*, Anant Madabhushi 1,7,*
PMCID: PMC8015207  NIHMSID: NIHMS1673743  PMID: 33576688

Abstract

Background -

We hypothesized that computerized morphologic analysis of the LA and pulmonary veins (PVs) via fractal measurements of shape and texture features of the LA myocardial wall could predict AF recurrence after ablation.

Methods -

Pre-ablation contrast CT scans were collected for 203 patients who underwent AF ablation. The LA body, PVs, and myocardial wall were segmented using a semi-automated region growing method. Twenty-eight fractal-based shape and texture-based features were extracted from resulting segments. The top features most associated with post-ablation recurrence were identified using feature selection and subsequently evaluated with a Random Forest classifier. Feature selection and classifier construction were performed on a discovery cohort (D1) of 137 patients; classifiers were subsequently validated on an independent set (D2) of 66 patients. Dedicated classifiers to capture the fractal and morphologic properties of LA body (CLA), PVs (CPV), and LA myocardial (CLAM) tissue were constructed, as well as a model (CAll) capturing properties of all segmented compartments. Fractal-based models were also compared against a model employing machine estimation of LA volume. To assess the effect of clinical parameters, such as AF type and catheter technique, a clinical model (Cclin) was also compared against CAll.

Results -

Statistically significant differences were observed for fractal features of CLA, CLAM and CAll in distinguishing AF recurrence (p<0.001) on D1. Using the five top features, CAll had the best prediction performance (AUC=0.81 [95% Confidence Interval (CI): 0.78–0.85]), followed by CPV (AUC=0.78 [95% CI: 0.74–0.80]) and CLA (AUC=0.70 [95% CI: 0.63–0.78]) on D2. The clinical parameter model Cclin yielded an AUC=0.70 [95% CI: 0.65–0.77], while the atrial volume model yielded an AUC=0.59. Combining CAll and Cclin on D2 improved the AUC to 0.87 [95% CI: 0.82–0.93].

Conclusions -

Fractal measurements of the LA, PVs, and atrial myocardium on CT scans were associated with likelihood of post-ablation AF recurrence.

Keywords: atrial fibrillation arrhythmia, atrial fibrillation heart failure, machine learning, catheter ablation, computed tomography, left atrium remodeling, recurrent atrial fibrillation, pulmonary veins, Atrial Fibrillation, Imaging

Graphical Abstract

graphic file with name nihms-1673743-f0005.jpg

Introduction

Atrial fibrillation (AF) is a common cardiac arrhythmia that affects one to two percent of the population1. Atrial fibrillation induces morphological changes of the left atrium (LA), which can manifest as changes in atrial volume2, shape3, the atrial wall, and PVs4. Remodeling of these structural changes after AF ablation has been detected using computed tomography (CT)5. Several models6,7 have been proposed to predict AF recurrence after ablation using clinical features, such as age8, hypertension9, and persistence of AF and LA scarring10,11. Further investigation into the different LA morphologies associated with AF recurrence support the use of imaging in identifying patients for ablation. Several investigators have reported that LA geometry has an important role in assessing AF incidence and recurrence using various cardiac imaging modalities12. Cardiac CT can assess atrial wall thickness and PV and LA anatomy before catheter ablation procedures for AF13. Varela et al. demonstrated that shape features of LA can be useful for predicting AF recurrence from screening CT scans14. Whitaker et al., demonstrated that changes in myocardial structure in the LA are an important part of this pathological remodeling process15.

Although Euclidean geometry is useful for describing smooth and regular shapes16, fractal dimension provides a novel approach to quantitatively characterize irregularity or roughness in shape and texture17. Complex biologic structures have self-similar properties that enable them to be quantified by their fractal dimension (FD)18–20.

The objective of this work was to evaluate the role of fractal related features pertaining to shape and textural differences of the LA and PVs on pre-treatment CT scans between AF patients with or without recurrent AF following ablation.

Methods

All data and supporting materials have been provided with the published article. In this study, we first extracted fractal features to characterize morphological and texture variation of the LA and PVs. Fractal-based classification models were constructed individually for the LA and PV models. A combined model involving features from both LA and PVs was also constructed. We then used fractal features to develop a random forest machine classifier to predict the likelihood of AF recurrence from CT scans. We evaluated whether fractal features provided additional predictive value beyond clinical variables, including age, sex, LA volume, left ventricular ejection fraction, body mass index, hypertension, sinus rhythm at time of ablation, AF type (paroxysmal, persistent), and ablation type (irrigated radiofrequency, cryoballoon ablation). An overview of our methodology is illustrated in Figure 1. This study was approved by the Institutional Review Board of the Cleveland Clinic for retrospective medical records review.

Figure 1.

Figure 1.

Open in a new tab

Overview of methodology. CT images were retrospectively collected. Regions of interest were segmented using Syngo.Via to obtain segmented masks for all models. Three-dimensional (3D) fractal features were extracted from the LA and PV segmentations. Next, we used Wilcoxon rank-sum tests to select the top features to train a random forest classifier and validate it on an independent dataset (n=66). ROC curves of the fractal-based radiomic classifier were trained on 203 patients with 137 patients for training (D1) and 66 patients for testing (D2). The different classifier models including those for the LA body, PV, and the combined model CAll were evaluated on D2 using ROC curve analysis. The details of LA body, PV, and tissue model are reported in the Supplemental section (Supplemental Table II).

Patient Population

This was a retrospective study involving two cohorts of patients with AF who underwent catheter ablation, each with a routine pulmonary vein CT scans prior to ablation (Details are provided in supplemental section (Supplemental I))21. Patients with CT scans with excessive artifacts or poor contrast (Figure 2) that hindered the ability to accurately segment the different tissue compartments were also excluded. In all, from a total of 239 patients, 36 patients were excluded (see flowchart for inclusion and exclusion criteria in Figure 2). A discovery cohort (D1) of 137 patients was identified from the ablation cohort from 2013-2016, and the independent validation set (D2) of 66 patients was constructed from the lone atrial fibrillation cohort from 2003-2009.

Figure 2.

Figure 2.

Open in a new tab

Flowchart showing inclusion and exclusion criteria for this study.

LA Segmentation and Feature Extraction

Binary segmentation models were created in Syngo.Via (Siemens, Munich, Germany) imaging software using a semi-automated 3D region growing algorithm to create a unified segmentation of the LA body, LA appendage, and full branching structure of the PVs (Details are provided in supplemental section (Supplemental Table II and Supplemental III)). The segmentation models are depicted in Figure 1, Step 2. The binary segmentation models corresponding to the PV and LA were subsequently exported to MATLAB 2018b (Mathworks, Natick, MA) for additional feature analysis.

Fractal analysis

Shape-based features of LA and PVs

We extracted 1D22, 2D23, and 3D fractal dimensions24 to assess shape related features of the LA. Fractal dimension (FD) is defined as the negative gradient of an ordinary least-squares fit line to the logarithm of box size and box count. Higher values of FDs are reflective of greater levels of anatomic complexity and surface variation. A more detailed description of 1D, 2D and 3D features is provided in the (Details are provided in Supplemental section (Supplemental IV and Supplemental Table III)). A total of 26 shape-based features were extracted from the PV and LA body models.

Texture features of Tissue model

Two fractal features related to texture variations of CT intensities in the LA wall were also extracted. This was done by employing the 3D fast Fourier transform (FFT)23 of fractal dimension. The CT scans were transformed to FD images using the differential box-counting (DBC) algorithm25.

Statistical analysis

We extracted a total of 28 features (26 shape features, two texture features), details of these features are in the Supplemental section (Supplemental Table IV). Feature selection was implemented to avoid the curse of dimensionality and reduce the risk of model overfitting. The variable ranking feature selection25 was used to identify the top five most discriminating features with the lowest p-value (<0.001, using the Wilcoxon rank sum tests) for each of the individual models in D1, followed by training a Random Forest (RF) classifier in conjunction with these top identified features. Dedicated random forest classifiers to capture the fractal and morphologic properties of LA body (CLA), PVs (CPV), and LA myocardial (CLAM) tissue were constructed, as well as model (CAll) capturing the associated properties of all the segmented compartments. To evaluate the predictive contribution of the PV alone, we created an additional classifier model (CPV-LA) by subtracting the PV segmentation from the segmentation of the LA, leaving only the branching structures of the PVs and the LA appendage (Figure 1). We also constructed a machine model (CLAV) employing computer estimated LA volume from the individual segmentations of the LA in order to compare the approach reported in26 with CAll, CPV, CLA. We used the Statistical Parametric Mapping (SPM) toolbox27 to estimate the volume of LA from CT scans. For clinical variables, the t test was implemented as a feature selection method and the features with the lowest p-value (<0.001, using the Wilcoxon rank sum tests) were used for building CClin. To determine clinical parameters that are associated with AF recurrence, we considered 20 clinical characteristics for CClin. The feature selection was used to identify the top clinical features, these are reported in Supplemental Table I and include age, LA volume, AF type, and catheter technique.

In addition, we also created an integrated clinical and FD model (CClin+FD). For CClin+FD, we used the top five selected FD features, including two shape-based features from PV model, two shape-based features from the Lumen, one 3D texture feature, and four clinical parameters including Age, AF type, Hypertension, and Catheter Technique.

The different models were all validated on the independent test set D2 (n=66). The various models were compared in terms of area (AUC) under the receiver operating characteristic curve (ROC). All AUCs were presented with bootstrap bias-corrected 95% confidence intervals (CIs).

Results

Experiment 1: Evaluate the role of fractal related features of the individual compartments of the heart in their association with likelihood of post-ablation AF recurrence

To characterize morphological and texture variation of the LA in patients, we extracted 28 fractal features from the training set, including 13 fractal features from PV, 13 fractal features from LA, and two texture-based features from LA wall. To assess the role of the fractal features and main differences between AF+ and AF−, 3D fractal features were shown in Figures 3 and 4. We considered 3D fractal-based shape features from LA and PVs and 3D fractal-based texture features from the LA wall to discriminate between AF+ and AF− patients. The 3D shape-based features from LA and PV were extracted as illustrated in Figure 3. The 3D fractal shape analysis of PV and LA body models are illustrated in Figure 4. The fractal texture features were found to be significantly different between AF+ and AF− patients (p-value<0.001). As illustrated in Figure 4, patients with AF recurrence tended to have a higher expression of 3D fractal features. As shown in Figure 3, AF− patients tend to have a lower fractal dimension, whereas AF+ patients tend to have a higher fractal dimension, the higher FD values reflecting greater levels of anatomic complexity and surface variation. Figure 4 shows 3D fractal analysis of myocardial wall to characterize texture variations. The heat maps in Figure 4 show significant differences in texture fractal features between AF+ and AF− patients, the feature expression being higher in AF+ compared to AF− patients.

Figure 3.

Figure 3.

Open in a new tab

3D Shape-based fractal features extracted from segmentation masks of LA and PVs. Images show feature extraction from masks using 3D fractal features to characterize shape of LA body and PV models. (A) PV model of LA consists of LA body and PVs and (B) LA body model consists of the LA body, appendage, and pulmonary vein ostia. Images show 3D fractal analysis of LA body and PV masks in patient with AF and without AF recurrence.

Figure 4.

Figure 4.

Open in a new tab

Illustration of 3D texture-based fractal features of myocardial wall of LA extracted from CT scans. Images show 3D fractal features of CT scans and 3D heatmaps of the features. 3D fractal dimension represent texture information of the myocardial tissue of LA. Top row represents CT scan, FD features, and FD heat map of the patient with AF recurrence (AF+). Bottom row represents CT scan, FD features, and FD heat map of the patient without AF recurrence (AF−).

Correlation heat maps of features are shown in supplemental section (Supplemental V)). The segmentation associated sensitivity of the classifier are reported in Supplemental section (Supplemental VI).

Experiment 2: A fractal-based machine learning classifier to predict AF recurrence

The objective of Experiment 2 was to develop and test a machine learning classifier to predict AF recurrence post-ablation using fractal features.

On D1, the model CPV had best prediction (AUC 0.78 [95% CI: 0.74–0.80]) with four top selected features (median of 3D FD, mean of 2D FD, entropy, and Gaussian mixture model of state space reconstruction), followed by the CLA (AUC 0.70 [95% CI: 0.63–0.78]) with four top selected features (median of 3D FD, mean of 2D FD, entropy, and Gaussian mixture model of state space reconstruction as reported in Supplemental Table IV).

The AUC for CPV alone was 0.72 with four top selected features, including 3D FD of shape, entropy, Gaussian mixture model, and 2D FD of shape. CAll yielded an AUC of 0.81 [95% CI: 0.78–0.85]) with features corresponding to median of 3D FD of Lumen and PV, mean of 2D FD, and entropy. Variable ranking importance of features are shown in supplemental section (Supplemental VII)).

Experiment 3: Integrating fractal features with the clinical variables and evaluating effect of ablation

The objective of the third experiment was to assess whether fractal-based imaging features added predictive value beyond common clinical parameters that are associated with AF recurrence: age, LA volume, AF type, and catheter technique.

The clinical features were reported in supplemental section (Supplemental VIII)). The clinical variables were used to train model with D1 (n=137) and test the model with independent validation D2. The model Cclin yielded an AUC on the test data of 0.70 and the corresponding model for all fractal features CAll, had an AUC of 0.81 [95% CI: 0.78–0.85]). The AUC for CClin+FD was 0.87 [95% CI: 0.82–0.93] on D2.

Discussion

Atrial fibrillation is a highly prevalent condition with substantial morbidity and mortality. It leads to complex changes in the LA that may contribute to AF recurrence28,29.

This work identified a set of quantitative fractal-based features that characterized the morphology and cardiac structure of the LA and PVs and texture changes of the LA wall that associated with the likelihood of developing recurrent AF. In Experiment 1, we focused on identifying those fractal features that best discriminate patients who had AF recurrence from those who did not. Fractal features corresponding to the LA body and PVs on segmented masks and the texture variation of the atrial wall were identified as the most discriminating features.

In Experiment 2, on the test set D2, the model CAll was identified as being most predictive of AF recurrence. Previous studies on the relationship between the anatomical structure of the PVs and LA and the occurrence and development of AF have mainly focused on anatomical variations of PVs and LA30–32. In an important study conducted by Syed et al., measurements of PV ostia were performed by CT, resulting in different numbers and positions of PVs, with poor correlation of diameter measurements among the imaging modalitie33. Wei et al., demonstrated a relationship between PV structural characteristics and LA diameter with AF recurrence34. This study reported that enlargement of LA is an independent risk factor for post-ablation recurrence of AF34.

In Experiment 3, we explored the effect of clinical and procedural variables on likelihood of recurrence and investigated the role of radiomic features as a function of AF type and catheter technique. Clinically, features previously predictive of recurrence are persistent AF and hypertension; here clinical models performed similarly to previous clinical predictive models of AF recurrence. Nakatani et al., found that heterogeneity in the left atrial wall thickness contributes to AF recurrence after catheter ablation35. The results suggest that the shape-based features of CLA and CPV and texture-based feature of atrial wall were most significant in predicting AF recurrence. Recent studies have shown that the LA wall thickness in patients with persistent AF is lower than that of patients with paroxysmal AF15; based on fractal analysis, the AF type is highly correlated with 3D texture features of atrial wall. The fractal feature based model was found to significantly outperform the model based on LA volume in predicting AF recurrence.

The fractal features improve the prediction performance of CClin; the correlation map between the clinical factors and the FD features were not found to be significant, as shown in the heat maps, further supporting the concept that FD features represent structural content and predictive value beyond clinical parameters, including LA volume and persistent AF, the main clinical factors previously reported to predict AF ablation success. The analyses also showed fractal features outperformed LA volume model, which had an AUC of only 0.59. FD features represent structural content and predictive value beyond clinical parameters, including LA volume and persistent AF.

Limitations

Various limitations of this study should be recognized. In this paper, we had a limited sample size with a single center design and retrospective data assessment. Even though we used independent validation cases, all the scans came from a single site. Multisite validation will need to be done to establish the generalizability of the approach. While an evaluation of the sensitivity of the approach to segmentation performance was done, we did not explicitly address or evaluate the sensitivity of the approach to different CT slice thicknesses, reconstruction kernels and scanners. LA fibrosis by late gadolinium enhancement has also been shown to be important factors in arrhythmia and patient stratification14 by CMR algorithms, which cannot be used on CT scans, though it is possible that textural differences on fractal CT analyses could be detecting similar substrates.

Conclusions

Fractal analysis of 3D morphology of the left atrium and pulmonary veins from CT scans were found to be predictive of AF recurrence following catheter ablation and may provide utility in improving patient selection or targeting for AF ablation. This work could also be used in determining follow-up needs for AF risk after ablation. The type of AF (paroxysmal vs. persistent) is highly correlated with 3D texture features of CLAM. Beyond type of AF, however, fractal analyses provided significantly improved prediction performance over clinical features, supporting the concept that FD features represent structural content and predictive value beyond clinical features, including LA volume and AF persistence.

Supplementary Material

Supplemental Material

What is Known?

  • Atrial fibrillation (AF) is a common cardiac arrhythmia that affects one to two percent of the population.

  • Atrial fibrillation induces morphological changes of the left atrium after AF ablation.

What the Study Adds?

  • Predicting AF recurrence based on morphological changes of the left atrium on CT scans.

  • 3D fractal features extracted from the left atrium, pulmonary veins, and myocardium wall to represent shape and texture variations.

  • A random forest machine classifier based on fractal features is introduced to predict the likelihood of AF recurrence from CT scans.

Acknowledgments

Sources of Funding:

Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health under award numbers U24CA199374, R01CA202752, R01CA208236, R01 CA216579, R01CA220581, and U01CA239055; National Center for Research Resources under award number C06RR12463, the National Heart Lung and Blood Institute of the National Institutes of Health under award number R0HL111314; the VA Merit Review Award IBX004121A from the United States Department of Veterans Affairs Biomedical Laboratory Research and Development Service, the DOD Prostate Cancer Idea Development Award (W81XWH-15-1-0558), the DOD Lung Cancer Investigator-Initiated Translational Research Award (W81XWH-18-1-0440), the DOD Peer Reviewed Cancer Research Program (W81XWH-16-1-0329), the Ohio Third Frontier Technology Validation Fund, the Wallace H. Coulter Foundation Program in the Department of Biomedical Engineering; the American Heart Association Atrial Fibrillation Strategically Focused Research Network under award numbers 18SFRN34110067, 18SFRN34170013; the National Center for Advancing Translational Sciences of the National Institutes of Health under UL1TR002548 and UL1RR024989 Clinical & Translational Science Collaborative of Cleveland; the Cleveland Clinic Department of Cardiovascular Medicine philanthropy research funds, the Tomsich Atrial Fibrillation Research Fund, and the Cleveland Clinic Center of Excellence in Cardiovascular Translational Functional Genomics. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health, the U.S. Department of Veterans Affairs, the Department of Defense, the American Heart Association, or the United States Government.

Nonstandard Abbreviations and Acronyms

AF

Atrial Fibrillation

PVs

Pulmonary veins

LA

Left atrium

FFT

Fast Fourier transform

FD

Fractal dimension

AF+

AF recurrence

AF−

Without AF recurrence

Footnotes

Disclosures: None

References:

  • 1.Pathak R, Lau DH, Mahajan R, Sanders P. Structural and functional remodeling of the left atrium: clinical and therapeutic implications for atrial fibrillation. J Atr Fibrillation. 2013;6:986. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Delgado V, Di Biase L, Leung M, Romero J, Tops LF, Casadei B, Marrouche N, Bax JJ. Structure and function of the left atrium and left atrial appendage: AF and stroke implications. J Am Coll Cardiol. 2017;70:3157–72. [DOI] [PubMed] [Google Scholar]
  • 3.Whitaker J, Rajani R, Chubb H, Gabrawi M, Varela M, Wright M, Niederer S, O’Neill MD. The role of myocardial wall thickness in atrial arrhythmogenesis. Europace. 2016;18:1758–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Park J, Park CH, Uhm JS, Pak HN, Lee MH, Joung B. A thin left atrial antral wall around the pulmonary vein reflects structural remodeling by atrial fibrillation and is associated with stroke. Yonsei Med J. 2017;58:282–289. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Ebersberger U, Bernard ML, Schoepf UJ, Wince WB, Litwin SE, Wang Y, Blanke P, Makowski MR, McQuiston AD, Silverman JR, et al. Cardiac Computed Tomography for Atrial Fibrillation Patients Undergoing Ablation: Implications for the Prediction of Early Recurrence. J Thorac Imaging. 2020;35:186–92. [DOI] [PubMed] [Google Scholar]
  • 6.Bieging ET, Morris A, Wilson BD, McGann CJ, Marrouche NF, Cates J. Left atrial shape predicts recurrence after atrial fibrillation catheter ablation. J Cardiovasc Electrophysiol. 2018;29:966–972. [DOI] [PubMed] [Google Scholar]
  • 7.Gilbert K, Bai W, Mauger C, Medrano-Gracia P, Suinesiaputra A, Lee AM, Sanghvi MM, Aung N, Piechnik SK, Neubauer S, et al. Independent left ventricular morphometric atlases show consistent relationships with cardiovascular risk factors: a UK biobank study. Sci Rep. 2019;9:1130. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Li SN, Wang L, Dong JZ, Yu RH, Long DY, Tang RB, Sang CH, Jiang CX, Liu N, Bai R, Du X. Electrocardiographic left ventricular hypertrophy predicts recurrence of atrial arrhythmias after catheter ablation of paroxysmal atrial fibrillation. Clin Cardiol. 2018;41:797–802. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.D’Ascenzo F, Corleto A, Biondi-Zoccai G, Anselmino M, Ferraris F, di Biase LU, Natale A, Hunter RJ, Schilling RJ, Miyazaki S, et al. Which are the most reliable predictors of recurrence of atrial fibrillation after transcatheter ablation?: a meta-analysis. Int J Cardiol. 2013;167:1984–9. [DOI] [PubMed] [Google Scholar]
  • 10.Yang G, Zhuang X, Khan H, Haldar S, Nyktari E, Li L, Wage R, Ye X, Slabaugh G, Mohiaddin R, Wong T. Fully automatic segmentation and objective assessment of atrial scars for long-standing persistent atrial fibrillation patients using late gadolinium-enhanced MRI. Med Phys. 2018;45:1562–1576. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Ciuffo L, Nguyen H, Marques MD, Aronis KN, Sivasambu B, de Vasconcelos HD, Tao S, Spragg DD, Marine JE, Berger RD, Lima JA. Periatrial fat quality predicts atrial fibrillation ablation outcome. Circ Cardiovasc Imaging. 2019;12:e008764. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Nakamori S, Ngo LH, Tugal D, Manning WJ, Nezafat R. Incremental value of left atrial geometric remodeling in predicting late atrial fibrillation recurrence after pulmonary vein isolation: a cardiovascular magnetic resonance study. J Am Heart Assoc. 2018;7:e009793. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Olsen FJ, Bertelsen L, de Knegt MC, Christensen TE, Vejlstrup N, Svendsen JH, Jensen JS, Biering-Sørensen T. Multimodality cardiac imaging for the assessment of left atrial function and the association with atrial arrhythmias. Circ Cardiovasc Imaging. 2016;9:e004947. [DOI] [PubMed] [Google Scholar]
  • 14.Varela M, Bisbal F, Zacur E, Berruezo A, Aslanidi OV, Mont L, Lamata P. Novel computational analysis of left atrial anatomy improves prediction of atrial fibrillation recurrence after ablation. Front Physiol. 2017;8:68. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Whitaker J, Rajani R, Chubb H, Gabrawi M, Varela M, Wright M, Niederer S, O’Neill MD. The role of myocardial wall thickness in atrial arrhythmogenesis. Europace. 2016;18:1758–1772. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Bizzarri M, Giuliani A, Cucina A, D’Anselmi F, Soto AM, Sonnenschein C. Fractal analysis in a systems biology approach to cancer. Semin Cancer Biol. 2011;21:175–182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Rodríguez JO, Prieto SE, Correa C, Bernal PA, Puerta GE, Vitery S, Soracipa Y, Muñoz D. Theoretical generalization of normal and sick coronary arteries with fractal dimensions and the arterial intrinsic mathematical harmony. BMC Med Phys. 2010;10:1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Braverman B, Tambasco M. Scale-specific multifractal medical image analysis. Comput Math Methods Med. 2013;2013:262931. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Captur G, Karperien AL, Li C, Zemrak F, Tobon-Gomez C, Gao X, Bluemke DA, Elliott PM, Petersen SE, Moon JC. Fractal frontiers in cardiovascular magnetic resonance: towards clinical implementation. J Cardiovasc Magn Reson. 2015;17:80. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Park YW, Kim S, Ahn SS, Han K, Kang SG, Chang JH, Kim SH, Lee SK, Park SH. Magnetic resonance imaging-based 3-dimensional fractal dimension and lacunarity analyses may predict the meningioma grade. Eur Radiol. 2020;30:4615–4622. [DOI] [PubMed] [Google Scholar]
  • 21.Hussein AA, Lindsay B, Madden R, Martin D, Saliba WI, Tarakji KG, Saqi B, Rausch DJ, Dresing T, Callahan T, Chung MK. New model of automated patient-reported outcomes applied in atrial fibrillation. Circ Arrhythm Electrophysiol. 2019;12:e006986. [DOI] [PubMed] [Google Scholar]
  • 22.Abdechiri M, Faez K, Amindavar H. Visual object tracking with online weighted chaotic multiple instance learning. Neurocomputing. 2017;247:16–30. [Google Scholar]
  • 23.Al-Zainaldin S, Glover PW, Lorinczi P. Synthetic fractal modelling of heterogeneous and anisotropic reservoirs for use in simulation studies: implications on their hydrocarbon recovery prediction. Transp Porous Media. 2017;116:181–212. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Nayak SR, Mishra J, mohan Jena P. Fractal analysis of image sets using differential box counting techniques. Int J Inf Technol. 2018;10:39–47. [Google Scholar]
  • 25.Janitza S, Strobl C, Boulesteix AL. An AUC-based permutation variable importance measure for random forests. BMC Bioinformatics. 2013;14:119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Karim R, Blake LE, Inoue J, Tao Q, Jia S, Housden RJ, Bhagirath P, Duval JL, Varela M, Behar JM, Cadour L. Algorithms for left atrial wall segmentation and thickness–Evaluation on an open-source CT and MRI image database. Med Image Anal. 2018;50:36–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Ashburner John, Barnes Gareth, Chen C, Daunizeau Jean, Flandin Guillaume, Friston Karl, Kiebel Stefan et al. , SPM12 manual. Wellcome Trust Centre for Neuroimaging, London, UK, p.2464, 2014. [Google Scholar]
  • 28.Chopra HK, Chandra P, Wander GS, Kumar V. Atrial fibrillation update: a textbook of cardiology. JP Medical Ltd; 2017. [Google Scholar]
  • 29.Den Uijl DW, Tops LF, Delgado V, Schuijf JD, Kroft LJ, de Roos A, Boersma E, Trines SA, Zeppenfeld K, Schalij MJ, Bax JJ. Effect of pulmonary vein anatomy and left atrial dimensions on outcome of circumferential radiofrequency catheter ablation for atrial fibrillation. Am J Cardiol. 2011;107:243–9. [DOI] [PubMed] [Google Scholar]
  • 30.Tian X, Zhang XJ, Yuan YF, Li CY, Zhou LX, Gao BL. Morphological and functional parameters of left atrial appendage play a greater role in atrial fibrillation relapse after radiofrequency ablation. Sci Rep. 2020;10:8072. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Miyama H, Takatsuki S, Hashimoto K, Yamashita T, Fujisawa T, Katsumata Y, Kimura T, Fukuda K. Anatomical changes in the pulmonary veins and left atrium after cryoballoon ablation. Pacing Clin Electrophysiol. 2020;43:1289–1294. [DOI] [PubMed] [Google Scholar]
  • 32.Atwater BD, Wallace TW, Kim HW, Hranitzky PM, Bahnson TD, Hegland DD, Daubert JP. Pulmonary vein contraction before and after radiofrequency ablation for atrial fibrillation. J Cardiovasc Electrophysiol. 2011;22:169–74. [DOI] [PubMed] [Google Scholar]
  • 33.Syed MA, Peters DC, Rashid H, Arai AE. Pulmonary vein imaging: comparison of 3D magnetic resonance angiography with 2D cine MRI for characterizing anatomy and size. J Cardiovasc Magn Reson. 2005;7:355–60.15881514 [Google Scholar]
  • 34.Wei W, Ge JB, Zou Y, Lin L, Cai Y, Liu XB, Zhu WQ. Anatomical characteristics of pulmonary veins for the prediction of postoperative recurrence after radiofrequency catheter ablation of atrial fibrillation. PLoS One. 2014;9:e93817. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Csécs I, Yamaguchi T, Kheirkhahan M, Czimbalmos C, Fochler F, Kholmovski EG, Morris AK, Kaur G, Vago H, Merkely B, Chelu MG. Left atrial functional and structural changes associated with ablation of atrial fibrillation-Cardiac magnetic resonance study. Int J Cardiol. 2020;305:154–160. [DOI] [PubMed] [Google Scholar]
  • 36.Donnellan E, Aagaard P, Kanj M, Jaber W, Elshazly M, Hoosien M, Baranowski B, Hussein A, Saliba W, Wazni O. Association between pre-ablation glycemic control and outcomes among patients with diabetes undergoing atrial fibrillation ablation. JACC Clin Electrophysiol. 2019;5:897–903. [DOI] [PubMed] [Google Scholar]
  • 37.Hussein AA, Saliba WI, Barakat A, Bassiouny M, Chamsi-Pasha M, Al-Bawardy R, Hakim A, Tarakji K, Baranowski B, Cantillon D, Dresing T. Radiofrequency ablation of persistent atrial fibrillation: diagnosis-to-ablation time, markers of pathways of atrial remodeling, and outcomes. Circ Arrhythm Electrophysiol. 2016;9:e003669. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Kanj MH, Wazni OM, Natale A. How to do circular mapping catheter-guided pulmonary vein antrum isolation: the Cleveland Clinic approach. Heart Rhythm. 2006;3:866–9. [DOI] [PubMed] [Google Scholar]
  • 39.Hussein AA, Martin DO, Saliba W, Patel D, Karim S, Batal O, Banna M, Williams-Andrews M, Sherman M, Kanj M, Bhargava M. Radiofrequency ablation of atrial fibrillation under therapeutic international normalized ratio: a safe and efficacious periprocedural anticoagulation strategy. Heart Rhythm. 2009;6:1425–9. [DOI] [PubMed] [Google Scholar]
  • 40.Bassiouny M, Saliba W, Rickard J, Shao M, Sey A, Diab M, Martin DO, Hussein A, Khoury M, Abi-Saleh B et al. Use of dabigatran for periprocedural anticoagulation in patients undergoing catheter ablation for atrial fibrillation. Circ Arrhythm Electrophysiol. 2013;6:460–466. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Canny J. A computational approach to edge detection. IEEE Trans Pattern Anal Mach Intell. 1986;8:679–98. [PubMed] [Google Scholar]
  • 42.Al-Zainaldin S, Glover PW, Lorinczi P. Synthetic fractal modelling of heterogeneous and anisotropic reservoirs for use in simulation studies: implications on their hydrocarbon recovery prediction. Transp Porous Media. 2017;116:181–212. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Material
NIHMS1673743-supplement-Supplemental_Material.pdf (204KB, pdf)

RESOURCES