TY - GEN
T1 - Explainable AI-Driven Optimized LSTM for Heart Sound Classification and Diagnostic Support
AU - Khan, Faiq Ahmad
AU - Hassan, Arshad
AU - Anwar, Lia
AU - Bermak, Amine
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2027.
PY - 2027
Y1 - 2027
N2 - Heart disease mortality necessitates accurate and interpretable diagnostic tools. This study optimizes a Long Short-Term Memory (LSTM) network for heart sound classification via GridSearchCV and Bayesian optimization. GridSearchCV achieved a balanced clinical profile (89.04% accuracy, 84.09% sensitivity), while Bayesian optimization maximized specificity (92.68%) at the expense of sensitivity (71.59%). To ensure clinical transparency, SHAP and LIME were integrated. SHAP identified global critical features (S1 and S2 means), while LIME provided patient-specific, local interpretability. The integration of rigorous optimization with Explainable AI validates network behavior and enhances clinical trust in automated diagnostic support systems.
AB - Heart disease mortality necessitates accurate and interpretable diagnostic tools. This study optimizes a Long Short-Term Memory (LSTM) network for heart sound classification via GridSearchCV and Bayesian optimization. GridSearchCV achieved a balanced clinical profile (89.04% accuracy, 84.09% sensitivity), while Bayesian optimization maximized specificity (92.68%) at the expense of sensitivity (71.59%). To ensure clinical transparency, SHAP and LIME were integrated. SHAP identified global critical features (S1 and S2 means), while LIME provided patient-specific, local interpretability. The integration of rigorous optimization with Explainable AI validates network behavior and enhances clinical trust in automated diagnostic support systems.
KW - Cardiac Diagnostics
KW - Explainable AI (XAI)
KW - Heart Sound Classification
KW - Hyperparameter Optimization
KW - Phonocardiogram (PCG)
UR - https://www.scopus.com/pages/publications/105045705271
U2 - 10.1007/978-3-032-30813-9_5
DO - 10.1007/978-3-032-30813-9_5
M3 - Conference contribution
AN - SCOPUS:105045705271
SN - 9783032308122
T3 - Lecture Notes in Computer Science
SP - 24
EP - 28
BT - Artificial Intelligence in Medicine - 24th International Conference, AIME 2026, Proceedings
A2 - Andreev, Pavel
A2 - Van Woensel, William
A2 - Sauré, Antoine
A2 - Holmes, John
PB - Springer Science and Business Media Deutschland GmbH
T2 - 24th International Conference on Artificial Intelligence in Medicine, AIME 2026
Y2 - 7 July 2026 through 10 July 2026
ER -