Dr., Dr. rer. medic.
Diese Website verwendet Cookies, die Ihre Zustimmung brauchen.
In this paper, we investigate how distribution shift affects model generalizability in medical time-series from intensive care. We define clinically relevant domain adaptation scenarios for predicting hypotension and hypoxemia using the MIMIC-III Matched Waveform Database v1.0. Distribution shift is quantified via the Kolmogorov–Smirnov test (KS) applied to an interpretable, lower-dimensional data representation. Our results highlight the challenges of measuring covariate, prior probability, and semantic shifts in non-stationary multivariate time series. We observed linear correlation (0.74 and 0.99) between KS(source train, target) - KS(source train, source test) and RMSE difference when comparing in-distribution and out-of-distribution model performance. We argue that developing and benchmarking data-driven models should explicitly account for modality-specific data shift characteristics.
Der menschliche Körper sendet kontinuierlich Biosignale aus, die wertvolle Einblicke in physiologische Prozesse liefern. In der Medizin werden diese Signale sowohl zu Forschungszwecken genutzt als auch um Diagnose und Monitoring von Krankheiten und Patienten zu unterstützen. Diese Veranstaltung bietet eine praxisnahe Einführung in die computergestützte Biosignalanalyse. Nach einer kurzen theoretischen Einführung zu den Grundlagen der Signalverarbeitung, einschließlich Definition, Erfassung und Anwendungsmöglichkeiten, erfolgt eine praktische Einführung in die Datenanalyse. Anhand eines realistischen Beispiels aus dem Patientenmonitoring im Intensivmedizin-Setting werden essentielle Schritte vermittelt: Daten-Vorbereitung, Feature-Engineering und die Vorhersage des Signals mit modernen Machine Learning Methoden durchgeführt. Es wird von Teilnehmenden der Besuch der vorangegangenen Veranstaltung ¿Coding Basics¿ oder ein äquivalentes Vorwissen in der Programmierung in Python vorausgesetzt.
In KLIPS anzeigen[Diese Lehrveranstaltung wird nur auf Englisch angeboten] Heart rate variability (HRV) is widely used in clinical settings as a non-invasive autonomic nervous system function marker. It helps assess cardiovascular health, stress levels, and overall well-being. Clinicians use HRV to monitor conditions like heart disease, hypertension, and diabetes, as well as to evaluate recovery in post-surgical and critically ill patients. HRV also plays a role in mental health, aiding in the diagnosis and management of anxiety, depression, or PTSD. Additionally, it is used in sports medicine and rehabilitation to track recovery and optimize training. Its broad applications make it a valuable tool in personalized medicine. The development of AI methods allows to make more complex predictions using multiple HRV parameters simultaneously. This complexity enabled successful decision support in the domains where HRV was not previously prominent for clinical use, such as epileptology. In this lecture block, we will discuss the technical aspects of HRV assessment, such as different sensors, data quality control or different HRV measures. We will review various types of clinical applications, but also the ¿citizen science¿ approach and sports coaching. For the practical part we take a dataset with precomputed R-to-R intervals and different labels (e.g. RR Interval Time Series Modeling: The PhysioNet/Computing in Cardiology Challenge 2002 v1.0.0 ). We will test different machine learning approaches to classify the data, e.g. to detect whether the data was recorded in a stressed or relaxed phase. This block lecture does not require any previous coding experience, we will use the graphical low-code platform KNIME. Students are required to bring their own laptop.
In KLIPS anzeigen