The recent advancements in the field of biomedical signal processing and brain-computer interfaces (BCIs) have shown a significant shift towards leveraging deep learning and self-supervised learning techniques. A notable trend is the development of general-purpose models that can be applied across various clinical practices, such as echocardiography and EEG analysis, demonstrating improved scalability and performance over traditional methods. These models often incorporate novel architectural elements, such as graph neural networks for audio identification and splittable frameworks for single-channel EEG representation learning, which enhance robustness and adaptability. Additionally, there is a growing focus on topology-preserving image registration methods for cardiac imaging and the integration of multi-concept generative adversarial networks for complex signal detection, such as non-suicidal self-injury. These innovations not only advance the accuracy and efficiency of diagnostic tools but also broaden their applicability in real-world scenarios. Notably, the introduction of end-to-end deep learning models for auditory attention decoding represents a significant leap in the development of neuro-steered hearing devices, offering improved generalization across subjects and potential for future assistive technologies.