DOSE-I: A Multimodal Biosignal Dataset of Procedural Sedation for Endoscopy -- Technical Report
This technical report details the DOSE-I dataset, a multimodal biosignal resource comprising 78.5 hours of clinically annotated procedural sedation data from 281 endoscopic procedures, including consciousness transitions, sedation depth labels, and preprocessed pEEG features to support future research.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are trying to teach a robot how to drive a car, but instead of a road, the robot is navigating a patient's body during a medical procedure. To do this safely, the robot needs to know exactly how "awake" or "asleep" the patient is at every single second. This is the tricky world of procedural sedation. When doctors perform delicate internal exams like endoscopies (looking inside the stomach or lungs), they give patients medicine to make them relaxed and pain-free. But there's a fine line: too little medicine, and the patient feels pain or moves; too much, and they might stop breathing or have their heart rate drop dangerously.
For a long time, doctors have relied on their own eyes and ears to guess this level of sleep, checking if a patient can squeeze a hand when asked. It's a bit like trying to judge the temperature of a soup by sticking your finger in it—sometimes it works, but it's not always precise. Scientists have been trying to build better "thermometers" for the brain using biosignals. These are electrical whispers from the body: the ECG (the heart's electrical rhythm), the EEG (the brain's electrical chatter), and the PLETH (a light-based pulse check from the finger). The big question in this field is: Can we collect enough of these electrical whispers from real patients to teach an artificial intelligence to predict exactly how deep the sedation is, making these procedures safer for everyone?
This paper introduces a massive new toolbox for that very purpose, called the DOSE-I dataset. Think of it as a giant, high-definition library of "sleepy patient" recordings. The researchers gathered 171 separate recordings from real endoscopy procedures at a hospital in Germany, totaling 78.5 hours of data. Inside this library, they didn't just dump raw video; they carefully labeled every moment with what the patient was actually doing and feeling. They tracked 1,129 transitions between being awake and being unconscious, and they recorded over 7,300 specific checks of how sedated the patient was using a standard scale called MOAA/S.
What makes this dataset special is that it's a "multimodal" treasure chest. It doesn't just have one type of signal; it has the heart's rhythm, the brain's waves, the finger pulse, and even the blood pressure, all happening at the same time. To make it even more useful, the authors didn't just stop at raw data. They also provided a "processed" version of the brain signals (called pEEG) that turns the messy electrical waves into 40 different numbers that describe the brain's state, like how much "slow" or "fast" activity is happening. They even included a "map" of where the data might be messy or noisy (artifacts), so researchers know which parts to trust and which to ignore.
The paper itself is a technical manual for this dataset. It explains exactly how the data was collected, how the electrodes were stuck to the patients' heads, and how the doctors marked down every time they gave a dose of the sedative drug Propofol. It details the specific types of patients involved (ranging from 21 to 83 years old) and the different kinds of procedures they underwent. The authors aren't claiming to have solved the problem of sedation monitoring yet; instead, they are saying, "Here is the best, most detailed set of real-world data we have built so far. We hope other scientists will use this to train their own AI models to make these procedures safer." They provide the raw materials—the electrical whispers, the drug doses, and the clinical notes—so that the next generation of researchers can build the "autopilot" that keeps patients safe while they sleep.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.