Explosive technological advances lead to current and future exponential growth of massive data-sets in medicine and health-related fields. Of particular importance is an innovative and adaptive acquisition of intrinsic features and metric structure hidden in the massive data-sets. For example, the hidden low dimensional physiological dynamics often expresses itself as the time-varying periodicity and trend in the observed dataset. Furthermore, from the practical viewpoint, the robustness of the algorithm to heteroscedastic noise/stochasticity and computational efficiency cannot be ignored. In this talk, I will discuss how to combine two modern adaptive signal processing techniques, empirical intrinsic geometry (EIG) and concentration of frequency and time (ConceFT), to meet these needs. In addition to the theoretical justification, a direct application to the sleep-depth detection problem, ventilator weaning prediction problem and the anesthesia depth problem will be demonstrated. If time permits, more applications like photoplethysmography and electrocardiography signal analysis will be discussed.