Pixel Map-Based Hybrid AI Framework for Early and Accurate Cardiovascular Disease Diagnosis

• Published in: Journal of computational and applied mathematics Vol. 476; p. 117095
• Main Authors: Emin, Berkay, Alaca, Yusuf, Akmeşe, Ömer Faruk, Karaca, Yeliz, Akgül, Akif
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.04.2026
• Subjects: Cardiovascular Disease Diagnosis; Classification Algorithms; Improved Salp Swarm Algorithm; Machine Learning Algorithms; Medical Artificial Intelligence; Multi- Modal Medical Image Analysis; Optimized Transfer Learning; Physiological Signal Analysis; Pixel Map Medical Imaging; Improved Salp Swarm Algorithm; Classification Algorithms; Machine Learning Algorithms; Optimized Transfer Learning; Medical Artificial Intelligence; Pixel Map Medical Imaging; Physiological Signal Analysis; Multi- Modal Medical Image Analysis; Cardiovascular Disease Diagnosis
• ISSN: 0377-0427
• DOI: 10.1016/j.cam.2025.117095

•The proposed PMMI- NN model involves the conversion of numerical clinical data into pixel map images, which facilitates the execution of visual-based artificial intelligence analysis for the detection of cardiovascular disease.•The unified hybrid AI regime integrates advanced transfer learning models, metaheuristic feature selection and machine learning classifiers for multi-modal medical image analysis.•The improved Salp Swarm Algorithm utilized optimizes feature selection, enhancing diagnostic efficiency and reducing computational burden.•The medical AI technique developed facilitates early and reliable identification of cardiovascular disease, providing alternative visual representation enhancing feature extraction for deep learning models.•The proposed PMMI-NN model established mitigates the risk of overfitting inherent in a single decision tree in the case of medical and clinical datasets comprised of different features, which ensures more stable and accurate predictions.. Pixel mapping is a computational approach that involves the analysis of individual pixels within digital images to extract clinically relevant information. This technique offers significant advantages in medical imaging applications, particularly in enhancing diagnostic accuracy and optimizing treatment planning. By reducing the need for complex manual programming, pixel mapping also contributes to workflow efficiency. The incorporation of machine learning strategies, such as transfer learning and other advanced algorithms, has further extended the capabilities of pixel mapping by enabling the segmentation of intricate anatomical structures and the detection of pathological anomalies in medical scans. Given the global prevalence of cardiovascular diseases (CVDs) which account for approximately 17.9 million deaths annually early diagnosis and effective clinical management remain paramount. This figure constitutes approximately 32% of all global deaths, with 85% of the deaths being because of heart attacks and strokes. In addition, 38% of deaths due to cardiovascular diseases occur at an early age in individuals under 70. Among these conditions, heart failure represents a major complication, underscoring the need for innovative diagnostic technologies like pixel-based analysis. This paper proposes an innovative hybrid model, which is Pixel Mapping Multi-Modal Medical Image -based Neural Networks (PMMI- NNs), by combining transfer and machine learning for diagnosing cardiovascular disease using pixel map images. Heart failure is also a common consequence of these diseases, which makes early diagnosis and management critical. This paper proposes an innovative hybrid model, which is Pixel Mapping Multi-Modal Medical Image -based Neural Networks (PMMI- NNs), by combining transfer and machine learning for diagnosing cardiovascular disease using pixel map images. Six different pre-trained convolutional neural network (CNN) models extract the significant features from the images: DenseNet121, VGG16, VGG19, InceptionV3, InceptionResNetV2 and MobileNetV2. The Improved Salp Swarm Algorithm is applied to improve classification accuracy and select the most discriminative features. The selected features are then classified by three machine learning algorithms, which are the k-nearest Neighbor, Random Forest and Support Vector Machine. The findings obtained by the analyses demonstrate that integration of transfer learning, feature selection, and machine learning proves to be effective in improving CVD diagnosis in medical image analysis, providing a scalable and applicable framework for future clinical applications. The primary objective of the study whose datasets include clinical and physiological signal elements is to compare different machine learning and deep learning approaches to identify the most effective methods for early diagnostic process of cardiovascular diseases. [Display omitted] .