Recent advances in machine learning based data analytics are opening opportunities for designing effective clinical decision support systems (CDSS) which can become the “third-eye” in the current clinical procedures and diagnosis. However, common patient movements in hospital wards may lead to faulty measurements in physiological sensor readings, and training a CDSS from such noisy data can cause misleading predictions, directly leading to potentially dangerous clinical decisions. In this work, we present MediSense, a system to sense, classify, and identify noise-causing motions and activities that affect physiological signal when made by patients on their hospital beds. Essentially, such a system can be considered as “glasses" for the clinical third eye in correctly observing medical data. MediSense combines wirelessly connected embedded platforms for motion detection with physiological signal data collected from patients to identify faulty physiological signal measurements and filters such noisy data from being used in CDSS training or testing datasets. We deploy our system in real intensive care units (ICUs), and evaluate its performance from real patient traces collected at these ICUs through a 4-month pilot study at the Ajou University Hospital Trauma Center, a major hospital facility located in Suwon, South Korea. Our results show that MediSense successfully classifies patient motions on the bed with >90% accuracy, shows 100% reliability in determining the locations of beds within the ICU, and each bed-attached sensor achieves a lifetime of more than 33 days, which satisfies the application-level requirements suggested by our clinical partners. Furthermore, a simple case-study with arrhythmia patient data shows that MediSense can help improve the clinical diagnosis accuracy.