Researchers at the Vellore Institute of Technology have developed an innovative hybrid artificial intelligence framework aimed at improving early detection of heart conditions. The Predator Crow search optimization Explainable algorithm (PCSO-XAI) combines optimization algorithms and interpretable machine learning as a single model that addresses existing challenges within clinical cardiology.
Heart diseases remains the leading cause of worldwide deaths, according to current mortality data, affecting numerous patients annually through myocardial infarctions, cerebrovascular accidents, and sudden fainting episodes. Medical professionals must detect these conditions appropriately, but they should achieve this diagnosis promptly before treatments can begin. Progress in wearable monitors and smart sensors remains hampered by the absence of dependable and interpretable AI solutions until this new advancement.
The introduction of the PCSO-XAI model marks a breakthrough in this research domain. The integration of PCSO as a nature-inspired metaheuristic with XAI achieves enhanced prediction accuracy while maintaining transparency levels. Lead researcher Dr. S. Rajalakshmi expressed the main limitation in predictive systems stems from their mysterious internal functioning. Medical practitioners require clear insight into the operation of predictive models, so XAI stands as a fundamental requirement for obtaining their trust.
A revolutionary technology fusion combines various systems that resolve problems across different stages, including image segmentation, feature selection, and diagnostic accuracy. The standard machine learning algorithms demonstrate poor performance when identifying predictive health indicators because they operate with high complexity and dataset imbalance. PCSO uses crow foraging behavior to optimize feature selection, thus allowing the model to concentrate on significant data points while reducing computational load.
This study both enhances classification quality and develops an advanced U-Net network for left ventricle segmentation in cardiac MRI examinations since this segmentation enables detailed heart structure and functional analysis. The process of segmenting cardiac images presents major challenges to medical professionals who normally perform manual corrections. The adapted approach lightens the processing work without losing segmentation accuracy.
The research draws its information from the Automated Cardiac Diagnosis Challenge (ACDC) dataset alongside MRI scans acquired through the Continuous Non-Invasive Atrial Fibrillation Diagnosis using imaging (imATFIB) study, which received ethical clearance. The MRI acquisition sequence included FIESTA together with LGE while obtaining detailed anatomical information about healthy subjects and patients with atrial fibrillation.
A comprehensive preprocessing improved image quality and established consistent intensity measurement standards across the dataset. The Synthetic Minority Over-Sampling Technique (SMOTE) algorithm serves as the solution to address the class imbalance by generating synthetic instances for balancing positive and negative sample distributions, which commonly affect cardiac datasets.
The hybrid framework demonstrated exceptional performance in extensive testing, with results reaching 99.72% accuracy, 96.47% precision, and 98.6% recall combined with a 94.6% F1 Score.
The introduced system outperformed current randomness forests and support vector machine (SVM) methods together with simple deep-learning approaches. The system presents both forecasting results and their supporting explanations to users because of its interpretability capability.
Available evidence suggests this model would establish a new standard for AI utilization in cardiology practice. The outstanding element of this work derives from its dedication to clinical applicability. For clinicians to adopt AI models effectively, they require diagnostic tools that both perform well and explain their reasoning to doctors, according to Dr. Anjali Sharma, who is a cardiologist outside of this study.
Still, challenges remain. The system demands considerable computing resources, which prevents it from achieving effective deployment at resource-limited facilities. Fresh versions of the system will work to decrease hardware specifications needed for integration between cloud and edge environments. The researchers intend to expand the framework for identifying arrhythmias and cardiac insufficiency manifestations.
The PCSO-XAI model showcases major progress in diagnostic science through its combination of optimization procedures, transparency features, and medical imaging technologies in a single integrated system. Healthcare’s advancement toward personalized intelligent solutions depends on breakthroughs such as these, which enable early, accurate, and dependable patient management practices.
References: Asha MM, Ramya G. Predator crow search optimization with explainable AI for cardiac vascular disease classification. Sci Rep. 2025;15:11692. doi:10.1038/s41598-025-96003-9
