Closed-loop systems such as Brain-Computer Interfaces (BCIs) represent a paradigm shift in neurotechnology, enabling real-time interaction between the brain and external devices for communication, control, and rehabilitation. However, their clinical and societal impact remains limited by BCI inefficiency, a phenomenon where up to 30% of users fail to achieve proficient control even after extensive training. While this concept highlights the challenges users face in interacting with BCI technologies, it also underscores the need for a fundamental rethinking of BCI design—one that moves beyond traditional approaches prioritizing decoder optimization alone. A more holistic framework is required, integrating neurophysiological insights, adaptive signal processing, and user-specific training strategies to unlock the full potential of these systems. This manuscript presents contributions to overcoming these limitations through a multidisciplinary approach bridging engineering and neuroscience. Leveraging multimodal and longitudinal BCI datasets, the neural mechanisms underlying the acquisition of BCI skills are explored, with a focus on identifying biomarkers of performance and training. To enhance the detection of mental states, the enrichment of information is investigated through multimodal integration and ensemble learning frameworks, resulting in the development of two open-source tools: HappyFeat, which facilitates feature selection, and FUCONE, which integrates functional connectivity estimators into ensemble methods. Building on this work, the utility of neuronal avalanches, bursts of enhanced neural activity, is evaluated as both a classification feature and a predictor of BCI performance. Finally, these findings are extended to demonstrate the versatility of BCI technologies beyond traditional communication and control paradigms, with applications in diagnosing and monitoring neurological disorders. All in all, this research aims to bridge the gap between technical innovation and clinical application, ultimately enabling the development of more adaptive, reliable, and user-centered BCI systems for real-world deployment, such as in the early diagnosis and rehabilitation of neurological disorders.