Hybrid Fusion of Time-Intensity Curve, Deep Learning, and Fractional Zernike-Caputo Features for Accurate Liver Lesion Classification in DCE-MRI.

Early and accurate detection of liver masses is essential for effective clinical management and improved patient outcomes, as treatment strategies differ significantly between benign and malignant lesions. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is widely used for liver lesion evaluation due to its ability to capture temporal enhancement behavior. However, reliable interpretation remains challenging because many lesions exhibit similar enhancement patterns, leading to diagnostic uncertainty and potential misclassification. This study proposes a hybrid diagnostic framework that integrates multiple complementary feature representations for automated liver mass classification. Specifically, the model extracts time-intensity curve (TIC) characteristics to capture contrast dynamics, deep learning features to represent complex spatial patterns, and fractional Zernike-Caputo descriptors to encode advanced mathematical shape and texture information. These heterogeneous feature sets are subsequently fused to form a unified and discriminative representation of each lesion. The proposed study aims to enhance the differentiation between benign and malignant liver masses by leveraging the strengths of kinetic, data-driven, and fractional mathematical descriptors. Experimental results demonstrate that the model achieves superior diagnostic performance, reaching an accuracy of 94.05% on a dataset comprising 645 dynamic MRI cases. Overall, the proposed approach provides a robust and efficient tool for liver lesion characterization, with potential to support clinical decision-making and reduce diagnostic ambiguity in medical imaging practice.