Intraparenchymal near-infrared spectroscopy for detection of delayed cerebral ischemia in poor-grade aneurysmal subarachnoid hemorrhage.

OBJECTIVE: Detection of delayed cerebral ischemia (DCI) is challenging in comatose patients with poor-grade aneurysmal subarachnoid hemorrhage (aSAH). Brain tissue oxygen pressure (PbtO2) monitoring may allow early detection of its occurrence. Recently, a probe for combined measurement of intracranial pressure (ICP) and intraparenchymal near-infrared spectroscopy (NIRS) has become available. In this pilot study, the parameters PbtO2, Hboxy, Hbdeoxy, Hbtotal and rSO2 were measured in parallel and evaluated for their potential to detect perfusion deficits or cerebral infarction. METHODS: In patients undergoing multimodal neuromonitoring due to poor neurological condition after aSAH, Clark oxygen probes, microdialysis and NIRS-ICP probes were applied. DCI was suspected when the measured parameters in neuromonitoring deteriorated. Thus, perfusion CT scan was performed as follow up, and DCI was confirmed as perfusion deficit. Median values for PbtO2, Hboxy, Hbdeoxy, Hbtotal and rSO2 in patients with perfusion deficit (Tmax > 6 s in at least 1 vascular territory) and/or already demarked infarcts were compared in 24- and 48-hour time frames before imaging. RESULTS: Data from 19 patients (14 University Hospital Zurich, 5 Charité Universitätsmedizin Berlin) were prospectively collected and analyzed. In patients with perfusion deficits, the median values for Hbtotal and Hboxy in both time frames were significantly lower. With perfusion deficits, the median values for Hboxy and Hbtotal in the 24 h time frame were 46,3 [39.6, 51.8] µmol/l (no perfusion deficits 53 [45.9, 55.4] µmol/l, p = 0.019) and 69,3 [61.9, 73.6] µmol/l (no perfusion deficits 74,6 [70.1, 79.6] µmol/l, p = 0.010), in the 48 h time frame 45,9 [39.4, 51.5] µmol/l (no perfusion deficits 52,9 [48.1, 55.1] µmol/l, p = 0.011) and 69,5 [62.4, 74.3] µmol/l (no perfusion deficits 75 [70,80] µmol/l, p = 0.008), respectively. In patients with perfusion deficits, PbtO2 showed no differences in both time frames. PbtO2 was significantly lower in patients with infarctions in both time frames. The median PbtO2 was 17,3 [8,25] mmHg (with no infarctions 29 [22.5, 36] mmHg, p = 0.006) in the 24 h time frame and 21,6 [11.1, 26.4] mmHg (with no infarctions 31 [22,35] mmHg, p = 0.042) in the 48 h time frame. In patients with infarctions, the median values of parameters measured by NIRS showed no significant differences. CONCLUSIONS: The combined NIRS-ICP probe may be useful for early detection of cerebral perfusion deficits and impending DCI. Validation in larger patient collectives is needed.

A supervised, externally validated machine learning model for artifact and drainage detection in high-resolution intracranial pressure monitoring data.

OBJECTIVE: In neurocritical care, data from multiple biosensors are continuously measured, but only sporadically acknowledged by the attending physicians. In contrast, machine learning (ML) tools can analyze large amounts of data continuously, taking advantage of underlying information. However, the performance of such ML-based solutions is limited by different factors, for example, by patient motion, manipulation, or, as in the case of external ventricular drains (EVDs), the drainage of CSF to control intracranial pressure (ICP). The authors aimed to develop an ML-based algorithm that automatically classifies normal signals, artifacts, and drainages in high-resolution ICP monitoring data from EVDs, making the data suitable for real-time artifact removal and for future ML applications. METHODS: In their 2-center retrospective cohort study, the authors used labeled ICP data from 40 patients in the first neurocritical care unit (University Hospital Zurich) for model development. The authors created 94 descriptive features that were used to train the model. They compared histogram-based gradient boosting with extremely randomized trees after building pipelines with principal component analysis, hyperparameter optimization via grid search, and sequential feature selection. Performance was measured with nested 5-fold cross-validation and multiclass area under the receiver operating characteristic curve (AUROC). Data from 20 patients in a second, independent neurocritical care unit (Charité - Universitätsmedizin Berlin) were used for external validation with bootstrapping technique and AUROC. RESULTS: In cross-validation, the best-performing model achieved a mean AUROC of 0.945 (95% CI 0.92-0.969) on the development dataset. On the external validation dataset, the model performed with a mean AUROC of 0.928 (95% CI 0.908-0.946) in 100 bootstrapping validation cycles to classify normal signals, artifacts, and drainages. CONCLUSIONS: Here, the authors developed a well-performing supervised model with external validation that can detect normal signals, artifacts, and drainages in ICP signals from patients in neurocritical care units. For future analyses, this is a powerful tool to discard artifacts or to detect drainage events in ICP monitoring signals.

Evaluation of machine learning algorithms for noninvasive intracranial pressure estimation using near infrared spectroscopy as a covariate.

BACKGROUND: Intracranial pressure (ICP) is a vital parameter that is continuously monitored in patients with severe brain injury and imminent intracranial hypertension. OBJECTIVE: To estimate intracranial pressure without intracranial probes based on transcutaneous near infrared spectroscopy (NIRS). METHODS: We developed machine learning based approaches for noninvasive intracranial pressure (ICP) estimation using signals from transcutaneous near infrared spectroscopy (NIRS) as well as other cardiovascular and artificial ventilation parameters. RESULTS: In a patient cohort of 25 patients, with 22 used for model development and 3 for model testing, the best performing models were Fourier transform based Transformer ICP waveform estimation which produced a mean absolute error of 4.68 mm Hg (SD = 5.4) in estimation. CONCLUSION: We did not find a significant improvement in ICP estimation accuracy by including signals measured by transcutaneous NIRS. We expect that with higher quality and greater volume of data, noninvasive estimation of ICP will improve.

Solving the explainable AI conundrum by bridging clinicians' needs and developers' goals.

Explainable artificial intelligence (XAI) has emerged as a promising solution for addressing the implementation challenges of AI/ML in healthcare. However, little is known about how developers and clinicians interpret XAI and what conflicting goals and requirements they may have. This paper presents the findings of a longitudinal multi-method study involving 112 developers and clinicians co-designing an XAI solution for a clinical decision support system. Our study identifies three key differences between developer and clinician mental models of XAI, including opposing goals (model interpretability vs. clinical plausibility), different sources of truth (data vs. patient), and the role of exploring new vs. exploiting old knowledge. Based on our findings, we propose design solutions that can help address the XAI conundrum in healthcare, including the use of causal inference models, personalized explanations, and ambidexterity between exploration and exploitation mindsets. Our study highlights the importance of considering the perspectives of both developers and clinicians in the design of XAI systems and provides practical recommendations for improving the effectiveness and usability of XAI in healthcare.

Distance measurements between 5 nanometer diamonds - single particle magnetic resonance or optical super-resolution imaging?

5 nanometer sized detonation nanodiamonds (DNDs) are studied as potential single-particle labels for distance measurements in biomolecules. Nitrogen-vacancy (NV) defects in the crystal lattice can be addressed through their fluorescence and optically-detected magnetic resonance (ODMR) of a single particle can be recorded. To achieve single-particle distance measurements, we propose two complementary approaches based on spin-spin coupling or optical super-resolution imaging. As a first approach, we try to measure the mutual magnetic dipole-dipole coupling between two NV centers in close DNDs using a pulse ODMR sequence (DEER). The electron spin coherence time, a key parameter to reach long distance DEER measurements, was prolonged using dynamical decoupling reaching T 2,DD ≈ 20 µs, extending the Hahn echo decay time T 2 by one order of magnitude. Nevertheless, an inter-particle NV-NV dipole coupling could not be measured. As a second approach, we successfully localize the NV centers in DNDs using STORM super-resolution imaging, achieving a localization precision of down to 15 nm, enabling optical nanometer-scale single-particle distance measurements.

ICU Cockpit: a platform for collecting multimodal waveform data, AI-based computational disease modeling and real-time decision support in the intensive care unit.

ICU Cockpit: a secure, fast, and scalable platform for collecting multimodal waveform data, online and historical data visualization, and online validation of algorithms in the intensive care unit. We present a network of software services that continuously stream waveforms from ICU beds to databases and a web-based user interface. Machine learning algorithms process the data streams and send outputs to the user interface. The architecture and capabilities of the platform are described. Since 2016, the platform has processed over 89 billion data points (N = 979 patients) from 200 signals (0.5-500 Hz) and laboratory analyses (once a day). We present an infrastructure-based framework for deploying and validating algorithms for critical care. The ICU Cockpit is a Big Data platform for critical care medicine, especially for multimodal waveform data. Uniquely, it allows algorithms to seamlessly integrate into the live data stream to produce clinical decision support and predictions in clinical practice.

Blood Pressure Variability Indices for Outcome Prediction After Thrombectomy in Stroke by Using High-Resolution Data.

BACKGROUND: Blood pressure variability (BPV) is associated with outcome after endovascular thrombectomy in acute large vessel occlusion stroke. We aimed to provide the optimal sampling frequency and BPV index for outcome prediction by using high-resolution blood pressure (BP) data. METHODS: Patient characteristics, 3-month outcome, and BP values measured intraarterially at 1 Hz for up to 24 h were extracted from 34 patients treated at a tertiary care center neurocritical care unit. Outcome was dichotomized (modified Rankin Scale 0-2, favorable, and 3-6, unfavorable) and associated with systolic BPV (as calculated by using standard deviation, coefficient of variation, averaged real variability, successive variation, number of trend changes, and a spectral approach using the power of specific BP frequencies). BP values were downsampled by either averaging or omitting all BP values within each prespecified time bin to compare the different sampling rates. RESULTS: Out of 34 patients (age 72 ± 12.7 years, 67.6% men), 10 (29.4%) achieved a favorable functional outcome and 24 (70.6%) had an unfavorable functional outcome at 3 months. No group differences were found in mean absolute systolic BP (SBP) (130 ± 18 mm Hg, p = 0.82) and diastolic BP (DBP) (59 ± 10 mm Hg, p = 1.00) during the monitoring time. BPV only reached predictive significance when using successive variation extracted from downsampled (averaged over 5 min) SBP data (median 4.8 mm Hg [range 3.8-7.1]) in patients with favorable versus 7.1 mmHg [range 5.5-9.7] in those with unfavorable outcome, area under the curve = 0.74 [confidence interval (CI) 0.57-0.85; p = 0.031], or the power of midrange frequencies between 1/20 and 1/5 min [area under the curve = 0.75 (CI 0.59-0.86), p = 0.020]. CONCLUSIONS: Using high-resolution BP data of 1 Hz, downsampling by averaging all BP values within 5-min intervals is essential to find relevant differences in systolic BPV, as noise can be avoided (confirmed by the significance of the power of midrange frequencies). These results demonstrate how high-resolution BP data can be processed for effective outcome prediction.