Abstract
Pathologists can diagnose diseases more quickly and accurately due to advancements in Computer-Aided Design (CAD) systems, leading to increased focus on deep learning-based CAD models. In this research, the HealthLNK and ICIAR datasets are used for implementation. The proposed “MultiNet” architecture is based on inductive transmission for classifying different cardiac vascular types. It is designed to deliver fast and precise cardiovascular diagnostics through both binary and multi-class classification. The approach integrates traditional and deep learning models, including DenseNet201 and VGG19, for feature extraction from microscopic images. Compared to conventional methods, transfer learning improves accuracy significantly. Extracted features are combined in a fusion layer to generate a merged feature vector for classification. The proposed method achieves accuracy rates of 99% and 95% on the HealthLNK and ICIAR cardiac MR datasets, respectively. The “CardiacMultiNet” framework demonstrates reliability and suitability for deployment in healthcare institutions. It outperforms existing approaches, achieving 94.2 MCC and 99.4% F1-score. Additionally, benign prediction accuracy is 98.9%, malignancy prediction is 99% and the average precision reaches 99%, highlighting its effectiveness for clinical decision support.
Keywords: Computer-aided Diagnosis, Deep Learning Models, Magnetic Resonance Imaging, Multi-class Classification, Multi-net Framework.