Machine Learning Detects Intraventricular Haemorrhage in Extremely Preterm Infants.

OBJECTIVE: To test the potential utility of applying machine learning methods to regional cerebral (rcSO2) and peripheral oxygen saturation (SpO2) signals to detect brain injury in extremely preterm infants. STUDY DESIGN: A subset of infants enrolled in the Management of Hypotension in Preterm infants (HIP) trial were analysed (n = 46). All eligible infants were <28 weeks' gestational age and had continuous rcSO2 measurements performed over the first 72 h and cranial ultrasounds performed during the first week after birth. SpO2 data were available for 32 infants. The rcSO2 and SpO2 signals were preprocessed, and prolonged relative desaturations (PRDs; data-driven desaturation in the 2-to-15-min range) were extracted. Numerous quantitative features were extracted from the biosignals before and after the exclusion of the PRDs within the signals. PRDs were also evaluated as a stand-alone feature. A machine learning model was used to detect brain injury (intraventricular haemorrhage-IVH grade II-IV) using a leave-one-out cross-validation approach. RESULTS: The area under the receiver operating characteristic curve (AUC) for the PRD rcSO2 was 0.846 (95% CI: 0.720-0.948), outperforming the rcSO2 threshold approach (AUC 0.593 95% CI 0.399-0.775). Neither the clinical model nor any of the SpO2 models were significantly associated with brain injury. CONCLUSION: There was a significant association between the data-driven definition of PRDs in rcSO2 and brain injury. Automated analysis of PRDs of the cerebral NIRS signal in extremely preterm infants may aid in better prediction of IVH compared with a threshold-based approach. Further investigation of the definition of the extracted PRDs and an understanding of the physiology underlying these events are required.

Sparse-Denoising Methods for Extracting Desaturation Transients in Cerebral Oxygenation Signals of Preterm Infants.

Preterm infants are at high risk of developing brain injury in the first days of life as a consequence of poor cerebral oxygen delivery. Near-infrared spectroscopy (NIRS) is an established technology developed to monitor regional tissue oxygenation. Detailed waveform analysis of the cerebral NIRS signal could improve the clinical utility of this method in accurately predicting brain injury. Frequent transient cerebral oxygen desaturations are commonly observed in extremely preterm infants, yet their clinical significance remains unclear. The aim of this study was to examine and compare the performance of two distinct approaches in isolating and extracting transient deflections within NIRS signals. We optimized three different simultaneous low-pass filtering and total variation denoising (LPF-TVD) methods and compared their performance with a recently proposed method that uses singular-spectrum analysis and the discrete cosine transform (SSA-DCT). Parameters for the LPF-TVD methods were optimized over a grid search using synthetic NIRS-like signals. The SSA-DCT method was modified with a post-processing procedure to increase sparsity in the extracted components. Our analysis, using a synthetic NIRS-like dataset, showed that a LPF-TVD method outperformed the modified SSA-DCT method: median mean-squared error of 0.97 (95% CI: 0.86 to 1.07) was lower for the LPF-TVD method compared to the modified SSA-DCT method of 1.48 (95% CI: 1.33 to 1.63), P<0.001. The dual low-pass filter and total variation denoising methods are considerably more computational efficient, by 3 to 4 orders of magnitude, than the SSA-DCT method. More research is needed to examine the efficacy of these methods in extracting oxygen desaturation in real NIRS signals.Clinical relevance- Early and precise identification of abnormal brain oxygenation in premature infants would enable clinicians to employ therapeutic strategies that seek to prevent brain injury and long-term morbidity in this vulnerable population.

The impact of unpredictability of dyspnea offset on dyspnea perception, fear, and respiratory neural gating.

Dyspnea is a debilitating and threatening symptom in various diseases. Affected patients often report the unpredictability of dyspnea episodes being particularly anxiety-provoking and amplifying the perception of dyspnea. Experimental studies testing dyspnea unpredictability together with related neural processes, physiological fear responses, and dyspnea-related personality traits are sparse. Therefore, we investigated the impact of unpredictability of dyspnea offset on dyspnea perception and fear ratings, respiratory neural gating and physiological fear indices, as well as the influence of interindividual differences in fear of suffocation (FoS). Forty healthy participants underwent a task manipulating the offset predictability of resistive load-induced dyspnea including one unloaded safety condition. Respiratory variables, self-reports of dyspnea intensity, dyspnea unpleasantness, and fear were recorded. Moreover, respiratory neural gating was measured in a paired inspiratory occlusion paradigm using electroencephalography, while electrodermal activity, startle eyeblink, and startle probe N100 were assessed as physiological fear indices. Participants reported higher dyspnea unpleasantness and fear when dyspnea offset was unpredictable compared to being predictable. Individuals with high levels of FoS showed the greatest increase in fear and overall higher levels of fear and physiological arousal across all conditions. Respiratory neural gating, startle eyeblink, and startle probe N100 showed general reductions during dyspnea conditions but no difference between unpredictable and predictable dyspnea conditions. Together, the current results suggest that the unpredictable offset of dyspnea amplifies dyspnea perception and fear, especially in individuals with high levels of FoS. These effects were unrelated to respiratory neural gating or physiological fear responses, requiring future studies on underlying mechanisms.