Reliable XAI Explanations in Sudden Cardiac Death Prediction for Chagas Cardiomyopathy

Vinícius P. Chagas; Luiz H. T. Viana; Mac M. da S. Carlos; João P. V. Madeiro; Roberto C. Pedrosa; Thiago Alves Rocha; Carlos H. L. Cavalcante

cs.LGFeb 25, 2026

Reliable XAI Explanations in Sudden Cardiac Death Prediction for Chagas Cardiomyopathy

Authors: Vinícius P. Chagas, Luiz H. T. Viana, Mac M. da S. Carlos, João P. V. Madeiro, Roberto C. Pedrosa, Thiago Alves Rocha, Carlos H. L. Cavalcante

Organizations: Federal Institute of Education and Technology of Ceara, Maracanau, Ceara, Brazil · Universidade Feferal do Ceará (UFC), CE · Edson Saad Heart Institute – Federal University of Rio de Janeiro, Rio de Janeiro, Brazil

Abstract

Sudden cardiac death (SCD) is unpredictable, and its prediction in Chagas cardiomyopathy (CC) remains a significant challenge, especially in patients not classified as high risk. While AI and machine learning models improve risk stratification, their adoption is hindered by a lack of transparency, as they are often perceived as black boxes with unclear decision-making processes. Some approaches apply heuristic explanations without correctness guarantees, leading to mistakes in the decision-making process. To address this, we apply a logic-based explainability method with correctness guarantees to the problem of SCD prediction in CC. This explainability method, applied to an AI classifier with over 95% accuracy and recall, demonstrated strong predictive performance and 100% explanation fidelity. When compared to state-of-the-art heuristic methods, it showed superior consistency and robustness. This approach enhances clinical trust, facilitates the integration of AI-driven tools into practice, and promotes large-scale deployment, particularly in endemic regions where it is most needed.

Figures & tables

Group	Qnt.	Features
Top 10	10	EVTotal; NSVT (mean duration); LVEDV; LA diameter; EF Teicholz; Rassi score (points); LVESV; TSH; Dyslipidemia (DLP); Amiodarone
Top 20	20	Top 10 + Holter Age; Stroke.1; ESV; NYHA Class I; HR; Rassi score_low; Sex; LVEF; Mean HR; Stroke
Positive Importance	48	Top 20 + Syncope; NYHA Class II; BMI; Diastolic Dysfunction_1.0; Classification_Moderate Dysfunction; Guideline 2005_B1; Type 2 Diabetes (DM2); Diastolic Dysfunction_2.0; Ventricular Ectopy (VE); Rassi score_intermediate; Hypertension (HTN); Sedentary lifestyle; Guideline 2005_B2; Coronary artery disease; Classification_Normal; Primary Alteration; Inactive Electrical Area; VE.1; Classification_Mild Dysfunction; ICD (Implantable Defibrillator); Heart Failure; Chronic Kidney Disease; NSVT; Segmental Deficit; AV Conduction Disturbance.1_0.0; Pulmonary Venous Doppler (PVD); Atrial Fibrillation/Flutter; Sinoatrial Node Dysfunction
Zero Importance	15	AFib/Atrial Flutter; Smoking; Pacemaker; Cancer; Pause > 3s; Guideline 2005_A; Ablations; AV Conduction Disturbance.1_1.0; AV Conduction Disturbance_3.0; AV Conduction Disturbance_2.0; AV Conduction Disturbance_1.0; AV Conduction Disturbance_0.0; Diastolic Dysfunction_0.0; Diastolic Dysfunction_3.0; NYHA Class III

Table 1: The 63 features, derived from one-hot encoding of 49 original input features, grouped according to their importance scores.

Features	Accuracy	Recall	Precision	F1-score	ROC AUC
All (63)	0.92 $\pm$ 0.04	0.85 $\pm$ 0.14	0.74 $\pm$ 0.16	0.78 $\pm$ 0.13	0.89 $\pm$ 0.08
Positive (48)	0.94 $\pm$ 0.037	0.90 $\pm$ 0.137	0.78 $\pm$ 0.14	0.83 $\pm$ 0.11	0.92 $\pm$ 0.07
Top 20	0.95 $\pm$ 0.03	0.95 $\pm$ 0.11	0.81 $\pm$ 0.18	0.86 $\pm$ 0.09	0.95 $\pm$ 0.05
Top 10	0.92 $\pm$ 0.05	0.78 $\pm$ 0.22	0.74 $\pm$ 0.18	0.76 $\pm$ 0.19	0.87 $\pm$ 0.12

Table 2: Metric Results for Each Subset of Features using Optuna with

k

-fold

Method	Class	Samples	Synthetic	Exp. Size	Fidelity(%)	Time(s)
Anchors	0	101	100	$1.99\pm 0.36$	$84.68\pm 15.32$	$0.88\pm 0.24$
Anchors	1	19	100	$3.58\pm 0.44$	$74.26\pm 25.74$	$0.88\pm 0.20$
Lime	0	101	100	$10\pm 0$	$98.78\pm 11.00$	$0.05\pm 0.01$
Lime	1	19	100	$10\pm 0$	$97.79\pm 14.71$	$0.06\pm 0.01$
Logic-based	0	101	100	$6.33\pm 0.69$	$\mathbf{100.00\pm 0.00}$	$0.38\pm 0.11$
Logic-based	1	19	100	$3.89\pm 1.52$	$\mathbf{100.00\pm 0.00}$	$0.32\pm 0.09$

Table 3: Comparison of Lime, Anchors and Logic-Based Explanations with respect to Fidelity

Sample	Class	Explanation
1	High risk	LA diameter = 4.5, LVEDV = 5.8, EVTotal = 11094
2	High risk	LVEDV = 5.8, EVTotal = 13812, NSVT (mean duration) = 1
3	Low risk	LVEDV = 4.4, EF Teicholz = 0.64, Rassi score (points) = 3, Stroke = 0, EVTotal = 1589, NSVT (mean duration) = 0
4	Low risk	LVEDV = 4.3, EF Teicholz = 0.68, Rassi score (points) = 2, Stroke = 0, EVTotal = 236, NSVT (mean duration) = 0

Table 4: Examples of Explanations to Four Samples of the Chagas Disease Dataset

Rank	Feature	Count in Explanations	Feature Importance
1	EVTotal	119	0.2488
2	LVEDV	114	0.0413
3	NSVT Holter	113	0.2756
4	Stroke	103	0.0608
5	Rassi Score(points)	97	0.0492
6	TSH	87	0.0488

Table 5: Comparison Between Feature Presence Count in Explanations and Feature Importance, Ordered by Count.

Clinical Risk Prediction

Clinical risk prediction estimates patients’ future health outcomes from medical records, physiological signals, imaging, and other health data.

See all 214 papers

ExECG: An Explainable AI Framework for ECG models
May 19, 2026Jong-Hwan Jang, Yong-yeon JoExplainable Artificial Intelligence Healthcare
Medical AI Co. Ltd., Seoul, Republic of Korea
Enhancing Framingham Cardiovascular Risk Score Transparency through Logic-Based XAI
Feb 25, 2026Emannuel L. de A. Bezerra, Luiz H. T. Viana, Vinícius P. Chagas +3Explainable Artificial Intelligence Clinical Decision Support
Instituto Federal do Ceará (IFCE), Maracanaú, Ceará Brazil
HeartAgent: An Autonomous Agent System for Explainable Differential Diagnosis in Cardiology
Mar 11, 2026Shuang Zhou, Kai Yu, Song Wang +11Clinical Decision Support Differential Diagnosis
Division of Computational Health Sciences, Department of Surgery, University of Minnesota, Minneapolis, MN, USA · Department of Computer Science, University of Central Florida, Orlando, FL, USA · College of Science and Engineering, University of Minnesota, Minneapolis, MN, USA +6

May 19, 2026J/K move · Enter open · S save

ExECG: An Explainable AI Framework for ECG models

Jong-Hwan Jang, Yong-yeon Jo

Medical AI Co. Ltd., Seoul, Republic of Korea

Explainable Artificial Intelligence Healthcare Electrocardiography ML Reproducibility Electrocardiogram Interpretation

Open paper AskPDF arXiv

Deep learning has enabled ECG diagnostic models with strong performance in tasks such as arrhythmia classification and abnormality detection. However, accuracy alone is insufficient for clinical deployment because it does not explain why a specific output was produced, limiting justification, error analysis, and trust. Although ECG XAI has been extensively investigated and steadily improved, practical pipelines and reporting conventions vary across studies, hindering reuse and reproducibility. To address these issues, we present Explainable AI framework for ECG models (ExECG), a Python framework that provides a three-stage pipeline: Wrapper standardizes access across heterogeneous ECG formats and intermediate representations, Explainer unifies diverse XAI methods under a shared execution protocol, and Visualizer supports consistent cross-method comparison within a unified interface. We demonstrate end-to-end usage with concise examples and two case studies, highlighting interoperable and reproducible ECG explainability.

Abstract

Clinical Risk Prediction

Explore similar work

ExECG: An Explainable AI Framework for ECG models

Enhancing Framingham Cardiovascular Risk Score Transparency through Logic-Based XAI

HeartAgent: An Autonomous Agent System for Explainable Differential Diagnosis in Cardiology