Medical image reconstruction and enhancement play a critical role in improving the reliability of computer assisted disease analysis. In this study, a simulation-based hybrid framework is proposed that integrates a deep neural network (DNN) with a cubature Kalman filter (CKF) to combine nonlinear state estimation with learning-based refinement for improved image reconstruction and organ-wise disease classification. The framework is evaluated in a controlled simulation environment using synthetic multimodal radiology-pathology images representing four organs (liver, kidney, lung, and heart) and three disease severity levels (normal, mild, and severe). The proposed hybrid approach consistently demonstrates higher reconstruction fidelity and classification performance than standalone CKF and DNN models. Quantitative evaluation using the peak signal-to-noise ratio and the structural similarity index measure indicates improved structural preservation and noise reduction. In addition, classification analysis using confusion matrices and derived performance metrics demonstrates reliable discrimination between disease severity levels. Across the simulated cases, improvements of approximately 5–10% in classification accuracy and higher reconstruction quality are achieved relative to baseline methods. These findings suggest that the integration of nonlinear filtering with deep learning can provide a robust computational framework for multimodal medical image reconstruction and organ-wise disease analysis. Although the results are based on simulation experiments, the proposed approach demonstrates potential for future validation using real clinical imaging datasets.
Saad Arif (Wed,) studied this question.