DL-CADNet: A Deep Learning Framework for Multi-Class Diagnosis ofCoronary Artery Disease Using Resting-State SPECT MPI

Introduction: CAD is currently considered one of the leading causes of death worldwide; thus, this condition calls for an accurate and timely diagnosis. CAD is commonly diagnosed through myocardial perfusion imaging or MPI using SPECT. Method: The proposed framework makes use of EfficientNetB0 on preprocessed images, which undergo necessary steps such as cropping, scaling, and normalizing. A specially designed dense-layered architecture is used for the feature classification and extraction. Results: Accuracy, sensitivity, recall, F1-score, precision, and AUC of ischemia prediction were 97.00%, 97.83%, 96.30%, 97.00%, 98.11%, and 0.9214, respectively. Discussion: The present work investigates the detection of myocardial ischemia earlier by the systematic classification of resting-state SPECT MPI scans into normal and abnormal classes (such as local anemia and necrosis) employing an innovative hybrid deep learning architecture, DL-CADNet. This study is different because it considers both resting stress-state SPECT-MPI images for comprehensive categorization using a hybrid DL-CADNet architecture, whereas DenseNet-121, VGG-16, and some other approaches have been used for the purpose of accurate diagnosis. Conclusion: DL-CADNet demonstrates excellent medical interpretability, real-time application capability, and strong diagnostic performance. Future research will concentrate on adding multi-institutional validation and stress-rest datasets to improve generalizability. Together, the above findings enhance the study's diagnostic utility and show that it has the potential to be a trustworthy decision-making tool for cardiovascular nuclear imaging.