Intensive Care Unit-Acquired Weakness (ICUAW) is one of the most common neuromuscular
complications in patients treated in intensive care. With increasing disease severity and
especially in analgosedated, ventilated and delirious patients with limited ability to
cooperate during the clinical examination, the detection and follow-up of ICUAW is
limited to impossible. The clinical diagnosis and severity assessment of ICUAW is usually
carried out with the help of established diagnostic methods (e.g. clinical-neurological
examination, Medical Research Council-Sum Score, electrophysiological examinations),
which, however, cannot be carried out regularly if the patient does not cooperate, thus
delaying the diagnosis of ICUAW and making follow-up more difficult. Neuromuscular
ultrasound (NMUS), on the other hand, is an easy-to-use, non-invasive examination option
that is largely independent of patient compliance and is increasingly being investigated
in patients with ICUAW. It was shown that NMUS can detect ICUAW and is helpful in
assessing the severity of muscular weakness. However, the standardized recording and
follow-up by means of scoring procedures (e.g. the 4-stage Heckmatt Scala) is assessed as
partially subjective by the examiner and each individual ultrasound image must be taken
with the human eye, taking into account various image parameters. To overcome these
diagnostic limitations, artificial intelligence (AI) could be a useful extension or even
an alternative.
AI is already being used in a variety of ways in medical diagnostics (e.g. in the
detection of tumors and organ assessment), and increasingly also in the analysis of
ultrasound images. In this study, the investigators aim to use AI, specifically
Convolutional Neural Networks (CNNs), to classify ultrasound images into different
categories based on muscle weakness. The main benefit of using AI for such tasks lies in
the automation it provides. Once an AI model has been trained on an initial set of
images, it can quickly categorize new, unseen images, significantly reducing the time and
human effort required for diagnosis. AI models can analyze large amounts of data quickly
and consistently, which is especially beneficial in a clinical intensive care setting. By
applying AI, this study aims to train the detection and classification of muscle weakness
in patients treated in intensive care. However, one challenge with AI models is their
"black box" nature, where the decision-making process is not transparent. To solve this
problem, the investigators will use explainable AI techniques (XAI) such as Grad-CAM
(Gradient-weighted Class Activation Mapping) to visualize the specific areas of the
ultrasound images that the AI model focuses on in its analysis. This not only helps
validate the AI decisions, but also provides insights into the morphological changes in
the muscles that come with different degrees of weakness.
By integrating AI and XAI, the study team aims to not only automate the detection and
categorization of muscle weakness, but also improve our understanding of the underlying
morphological changes in muscles. This dual approach could lead to more accurate and
reliable diagnostics and ultimately improve outcomes for patients in intensive care.
