Search

Patient-ventilator asynchronies occur during mechanical ventilation when there is a mismatch between the patient’s needs and the ventilator’s settings. Ineffective Efforts during Expiration (IEE) is such an asynchrony, which if undetected can cause patient stress and prolong the ventilation duration. Current approaches to detect IEEs only target adult patients and are not directly applicable to newborns. This work explores the use of Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models to detect IEEs in neonates. An attention mechanism is employed to additionally offer a visual explanation for the network’s classification. All of the tested attention RNN architectures yield an accuracy of over 90%, with LSTMs performing slightly better than simple RNNs. A user study is conducted to evaluate the usability of the employed explainability method. The results show that the used visualizations are intuitive to understand, however the network’s attention itself can be misleading in certain cases.

Research showed that the normothermic machine perfusion of kidneys can enable prolonged storage and improve conditions compared to traditional cold storage. For research in this area, there is a demand for a long-term in vitro perfusion setup. In this work, we present a fully automated normothermic machine perfusion (NMP) system as an experimental research platform. The perfusion system is intended as a tool for researching the effects of different perfusion strategies on the kidney. To enable the automation, the NMP system consists of a blood pressure control, a circulation volume level control and a pH-regulation component. The setup is realized as a medical cyber-physical system consisting of networked embedded microcontroller nodes.

In extracorporeal blood circulation intensive care treatments, the occurrence of gas within the circulation is one known major hazard. This gas volume can cause severe harm to the patient like infarctions. Consequently, within risk assessment for these treatments gas bubbles are usually addressed by either constructive or signal based approaches. All signal-based approaches do have in common that they need a sufficient amount of data to be parameterized. These data can only be acquired in animal trials or laboratory experiments, as they could result in harm to patients. Hence, we designed a mock loop, which is automatically able to create annotated data of gas bubbles injected into an extracorporeal circulation. We were able to run this setup with a periodicity of 15 seconds, which results in 240 annotated measurements per hour. For the evaluation, we created 1095 bubbles of varying sizes (0.3 to 0.5 ml). The elaborated setup enables us to produce a great amount of annotated data, which is shown to be comparable to manually generated data in a convenient and fully automated manner.