Postoperative IMC
Multimodal AI-based pain measurement for intermediate care patients in the postoperative phase
Motivation
Previous projects at the NIT research group resulted in advacements in the area of pain monitoring through the creation of the datasets BioVid and X-ITE. These datasets are recorded in a laboratory environment with healthy patients that are fully oriented. This however may lead to a reality gap when AI-based models are trained on these and then applied to real world settings. Additionally, since the effective communication is alredy problematic with fully oriented patients, the problem gets even more severe with not fully oriented or non communicative patients. These can include sleeping or comatose patients, as well as patients with mental impairments such as Alzheimer’s. Therefore, it is imperative to include such patients as a focus group to ensure that the pain monitoring system continues to work reliably and robustly even for these cases.
Goals
The research project consists of three main phases. First, a new database must be created based on a real-world clinical scenario. To achieve this, a recording setup was installed in the intermediate care unit of the University Hospital in Ulm, Germany. Since September 2024, patients in the immediate recovery phase following abdominal surgery have been recorded over a 24-hour period, capturing both facial video and bio-signals. This dataset will then be used to evaluate the reliability of all developed AI-based models in a realistic environment. In a second recording phase, the focus will shift exclusively to non-communicative and disoriented patients.
