Biomedical signals and images are usually characterized as highly noisy and nonstationary.To extract transient and tiny features from noisy and complicated biomedical signals or images often requires continuous involvement of human intervention,which is inefficient and impractical in many cases.Therefore,it is highly desirable to develop automated feature extraction techniques for applications where human intervention is unachievable or unfavorable.For instance,automatic extraction of evoked brain potentials on a single-trial basis is a key problem in modern brain-computer interface (BCI) and inter-operative monitoring,and simultaneous EEG/fMRI study.Another example where automated techniques are desirable is the segmentation and tracking of continuous image features from medical image/video sequences,which conventionally requires laborious manual operations.In this paper,we introduced two recently developed automated feature extraction techniques: 1) single-trial detection of evoked potential (EPs) for BCI and simultaneous EEG/fMRI fusion; 2) automated region tracking of ultrasonographic image sequences for multimodality integration of muscle behaviors.