SEGMENTATION OF CARDIAC MRI IMAGES USING FULLY CONVOLUTIONAL DENSE NETWORKS (FC-DENSENET): AN ISLAMIC PERSPECTIVE ON MULTI- CLASS SEGMENTATION OF RIGHT VENTRICLE (RV), MYOCARDIUM (MYO), AND LEFT VENTRICLE (LV)

Abstract

This study developed an automatic multi-class segmentation model for cardiac MRI images using the FC-DenseNet architecture. The model was trained on a dataset of 1,322 images from patients with various cardiovascular conditions and healthy subjects. Data augmentation techniques, including rotation, shifting, zooming, and flipping, were employed to enhance generalization. The model achieved Dice Coefficient scores of 0.83 for the right ventricle (RV),

0.78 for the myocardium (Myo), and 0.71 for the left ventricle (LV), demonstrating satisfactory segmentation performance. High training accuracy and decreasing loss over 30 epochs indicate effective learning, highlighting the model's promise for clinical application in  cardiac MRI analysis.  This  research  resonates  with  the  teachings of  Surah  Al-Maidah, verse  32,  which emphasizes the significance of saving and protecting life. By improving diagnostic capabilities in cardiology, this model not only contributes to better treatment planning for heart diseases but also embodies the moral responsibility to use technology for the benefit of humanity. To further enhance the model’s accuracy and reliability, strategies such as advanced data augmentation, ensemble modeling, hyperparameter optimization, and attention mechanisms could be explored, ultimately supporting improved diagnosis and treatment planning in cardiology.