Sources of Risk of AI Systems.

Artificial intelligence can be used to realise new types of protective devices and assistance systems, so their importance for occupational safety and health is continuously increasing. However, established risk mitigation measures in software development are only partially suitable for applications in AI systems, which only create new sources of risk. Risk management for systems that for systems using AI must therefore be adapted to the new problems. This work objects to contribute hereto by identifying relevant sources of risk for AI systems. For this purpose, the differences between AI systems, especially those based on modern machine learning methods, and classical software were analysed, and the current research fields of trustworthy AI were evaluated. On this basis, a taxonomy could be created that provides an overview of various AI-specific sources of risk. These new sources of risk should be taken into account in the overall risk assessment of a system based on AI technologies, examined for their criticality and managed accordingly at an early stage to prevent a later system failure.

The Acute Phase of Experimental Subarachnoid Hemorrhage: Intracranial Pressure Dynamics and Their Effect on Cerebral Blood Flow and Autoregulation.

Clinical presentation and neurological outcome in subarachnoid hemorrhage (SAH) is highly variable. Aneurysmal SAH (aSAH) is hallmarked by sudden increase of intracranial pressure (ICP) and acute hypoperfusion contributing to early brain injury (EBI) and worse outcome, while milder or non-aneurysmal SAH with comparable amount of blood are associated with better neurological outcome, possibly due to less dramatic changes in ICP. Acute pressure dynamics may therefore be an important pathophysiological aspect determining neurological complications and outcome. We investigated the influence of ICP variability on acute changes after SAH by modulating injection velocity and composition in an experimental model of SAH. Five hundred microliters of arterial blood (AB) or normal saline (NS) were injected intracisternally over 1 (AB1, NS1), 10 (AB10, NS10), or 30 min (AB30) with monitoring for 6 h (n = 68). Rapid blood injection resulted in highest ICP peaks (AB1 median 142.7 mmHg [1.Q 116.7-3.Q 230.6], AB30 33.42 mmHg [18.8-38.3], p < 0.001) and most severe hypoperfusion (AB1 16.6% [11.3-30.6], AB30 44.2% [34.8-59.8]; p < 0.05). However, after 30 min, all blood groups showed comparable ICP elevation and prolonged hypoperfusion. Cerebral autoregulation was disrupted initially due to the immediate ICP increase in all groups except NS10; only AB1, however, resulted in sustained impairment of autoregulation, as well as early neuronal cell loss. Rapidity and composition of hemorrhage resulted in characteristic hyperacute hemodynamic changes, with comparable hypoperfusion despite different ICP ranges. Only rapid ICP increase was associated with pronounced and early, but sustained disruption of cerebral autoregulation, possibly contributing to EBI.

Analysis of NIRS-based muscle oxygenation parameters by inclusion of adipose tissue thickness.

The assessment of muscle oxygenation by non-invasive near-infrared spectroscopy generally assumes a homogeneous medium, and this is flawed for large adipose tissue layers underneath the skin. Here we summarize the influence of the adipose tissue thickness on the oxygenation data, show that the adipose layer can be measured by NIRS and indicate a possible correction algorithm. Spectroscopic evidence suggests the usefulness of this algorithm, however, not in all subjects.

Tissue oxygenation during exercise measured with NIRS: reproducibility and influence of wavelengths.

Near-infrared spectroscopy (NIRS) is widely used for the measurement of skeletal muscle oxygenation during exercise as it reflects muscle metabolism, and most studies report a large variability between subjects. Here we assess the data quality of tissue oxygen saturation (SO2) and oxygenated (oxyHb) and deoxygenated (deoxyHb) haemoglobin concentrations recorded during an incremental cycling protocol in nine healthy volunteers. The protocol was repeated three times on the same day and a fourth session on a different day to estimate the reproducibility of the method with a broadband, spatially resolved spectroscopy (SRS) system. We found that the inter-subject variation in SO2 (standard deviation ≈ 6 %) was considerably larger than the reproducibility (≈ 1.5 %) both for the same-day and different-day tests. The reproducibility of changes in SO2 was better than 1 %.

Wavelength selection for the improvement of the signal-to-noise ratio for imaging of haemoglobin oxygenation with RGB reflectometry.

We demonstrate the optimisation of wavelengths for the imaging of cortical haemoglobin oxygenation with broadband RGB reflectometry. Wavelengths were chosen in order to minimise the likely crosstalk and optimise the signal-to-noise ratio by simulating effects of different combinations of wavelengths on the condition number of the resulting extinction coefficient matrices. The results obtained were evaluated experimentally for four combinations of commercially available LED combinations and compared with data from the literature.

Simultaneous imaging of cortical blood flow and haemoglobin concentration with LASCA and RGB reflectometry.

We demonstrate a system for the simultaneous imaging of cortical blood flow and haemoglobin oxygenation by laser speckle contrast analysis (LASCA) and RGB reflectometry. The sensitivity of the system was tested by observing changes of haemoglobin oxygenation and blood flow in rats in response to ischaemic stroke, hypercapnia, hyperoxia, hypoxia, cortical spreading depression and cortical activation following forepaw stimulation.