Data fusion—the integration of signals from multiple sensors in real time—has become a rapidly evolving field with significant implications for biomedical signal processing. This chapter introduces a novel convolutional neural network (CNN)-based methodology for fusing multimodal, multi-sensor data in a fully automated and data-driven manner. Unlike traditional fusion approaches that require manual specification of the fusion level, the proposed model learns to determine the optimal level of information abstraction autonomously, integrating both feature extraction and fusion within a unified architecture. A key innovation of this approach is the inclusion of signal quality indicators (SQIs) as additional input streams, allowing the model to dynamically adjust the fusion process based on the reliability of each sensor input. The method is validated through the task of atrial fibrillation detection using multimodal data—electrocardiogram (ECG) and photoplethysmogram (PPG) signals—sourced from the MIMIC III database. The model, trained using an average loss function, achieves impressive performance metrics with an accuracy of 99.33% and sensitivity of 99.74%. Its robustness under noisy conditions further underscores the effectiveness of the SQI-informed, multi-level fusion strategy. This chapter highlights the potential of intelligent, quality-aware fusion techniques to advance the reliability and efficiency of health monitoring systems.