Hoshik Lee

A new algorithm for detecting central apnea in neonates

Apnea of prematurity is an important and common clinical problem, and is often the rate-limiting process in NICU discharge. Accurate detection of episodes of clinically important neonatal apnea using existing chest impedance (CI) monitoring is a clinical imperative. The technique relies on changes in impedance as the lungs fill with air, a high impedance substance. A potential confounder, however, is blood coursing through the heart. Thus, the cardiac signal during apnea might be mistaken for breathing. We report here a new filter to remove the cardiac signal from the CI that employs a novel resampling technique optimally suited to remove the heart rate signal, allowing improved apnea detection. We also develop an apnea detection method that employs the CI after cardiac filtering. The method has been applied to a large database of physiological signals, and we prove that, compared to the presently used monitors, the new method gives substantial improvement in apnea detection.

Anemia, Apnea of Prematurity, and Blood Transfusions

OBJECTIVE To compare the frequency and severity of apneic events in very low birth weight (VLBW) infants before and after blood transfusions using continuous electronic waveform analysis. STUDY DESIGN We continuously collected waveform, heart rate, and oxygen saturation data from patients in all 45 neonatal intensive care unit beds at the University of Virginia for 120 weeks. Central apneas were detected using continuous computer processing of chest impedance, electrocardiographic, and oximetry signals. Apnea was defined as respiratory pauses of >10, >20, and >30 seconds when accompanied by bradycardia (<100 beats per minute) and hypoxemia (<80% oxyhemoglobin saturation as detected by pulse oximetry). Times of packed red blood cell transfusions were determined from bedside charts. Two cohorts were analyzed. In the transfusion cohort, waveforms were analyzed for 3 days before and after the transfusion for all VLBW infants who received a blood transfusion while also breathing spontaneously. Mean apnea rates for the previous 12 hours were quantified and differences for 12 hours before and after transfusion were compared. In the hematocrit cohort, 1453 hematocrit values from all VLBW infants admitted and breathing spontaneously during the time period were retrieved, and the association of hematocrit and apnea in the next 12 hours was tested using logistic regression. RESULTS Sixty-seven infants had 110 blood transfusions during times when complete monitoring data were available. Transfusion was associated with fewer computer-detected apneic events (P < .01). Probability of future apnea occurring within 12 hours increased with decreasing hematocrit values (P < .001). CONCLUSIONS Blood transfusions are associated with decreased apnea in VLBW infants, and apneas are less frequent at higher hematocrits.

RNNDROP: A novel dropout for RNNS in ASR

Recently, recurrent neural networks (RNN) have achieved the state-of-the-art performance in several applications that deal with temporal data, e.g., speech recognition, handwriting recognition and machine translation. While the ability of handling long-term dependency in data is the key for the success of RNN, combating over-fitting in training the models is a critical issue for achieving the cutting-edge performance particularly when the depth and size of the network increase. To that end, there have been some attempts to apply the dropout, a popular regularization scheme for the feed-forward neural networks, to RNNs, but they do not perform as well as other regularization scheme such as weight noise injection. In this paper, we propose rnnDrop, a novel variant of the dropout tailored for RNNs. Unlike the existing methods where dropout is applied only to the non-recurrent connections, the proposed method applies dropout to the recurrent connections as well in such a way that RNNs generalize well. Our experiments show that rnnDrop is a better regularization method than others including weight noise injection. Namely, when deep bidirectional long short-term memory (LSTM) RNNs were trained with rnnDrop as acoustic models for phoneme and speech recognition, they significantly outperformed the current state-of-the-arts; we achieved the phoneme error rate of 16.29% on the TIMIT core test set for phoneme recognition and the word error rate of 5.53% on the Wall Street Journal (WSJ) dataset, dev93, for speech recognition, which are the best reported results on both of the datasets.

Accurate automated apnea analysis in preterm infants.

OBJECTIVE In 2006 the apnea of prematurity (AOP) consensus group identified inaccurate counting of apnea episodes as a major barrier to progress in AOP research. We compare nursing records of AOP to events detected by a clinically validated computer algorithm that detects apnea from standard bedside monitors. STUDY DESIGN Waveform, vital sign, and alarm data were collected continuously from all very low-birth-weight infants admitted over a 25-month period, analyzed for central apnea, bradycardia, and desaturation (ABD) events, and compared with nursing documentation collected from charts. Our algorithm defined apnea as > 10 seconds if accompanied by bradycardia and desaturation. RESULTS Of the 3,019 nurse-recorded events, only 68% had any algorithm-detected ABD event. Of the 5,275 algorithm-detected prolonged apnea events > 30 seconds, only 26% had nurse-recorded documentation within 1 hour. Monitor alarms sounded in only 74% of events of algorithm-detected prolonged apnea events > 10 seconds. There were 8,190,418 monitor alarms of any description throughout the neonatal intensive care unit during the 747 days analyzed, or one alarm every 2 to 3 minutes per nurse. CONCLUSION An automated computer algorithm for continuous ABD quantitation is a far more reliable tool than the medical record to address the important research questions identified by the 2006 AOP consensus group.

Breath-by-breath analysis of cardiorespiratory interaction for quantifying developmental maturity in premature infants

In healthy neonates, connections between the heart and lungs through brain stem chemosensory pathways and the autonomic nervous system result in cardiorespiratory synchronization. This interdependence between cardiac and respiratory dynamics can be difficult to measure because of intermittent signal quality in intensive care settings and variability of heart and breathing rates. We employed a phase-based measure suggested by Schäfer and coworkers (Schäfer C, Rosenblum MG, Kurths J, Abel HH. Nature 392: 239-240, 1998) to obtain a breath-by-breath analysis of cardiorespiratory interaction. This measure of cardiorespiratory interaction does not distinguish between cardiac control of respiration associated with cardioventilatory coupling and respiratory influences on the heart rate associated with respiratory sinus arrhythmia. We calculated, in sliding 4-min windows, the probability density of heartbeats as a function of the concurrent phase of the respiratory cycle. Probability density functions whose Shannon entropy had a <0.1% chance of occurring from random numbers were classified as exhibiting interaction. In this way, we analyzed 18 infant-years of data from 1,202 patients in the Neonatal Intensive Care Unit at University of Virginia. We found evidence of interaction in 3.3 patient-years of data (18%). Cardiorespiratory interaction increased several-fold with postnatal development, but, surprisingly, the rate of increase was not affected by gestational age at birth. We find evidence for moderate correspondence between this measure of cardiorespiratory interaction and cardioventilatory coupling and no evidence for respiratory sinus arrhythmia, leading to the need for further investigation of the underlying mechanism. Such continuous measures of physiological interaction may serve to gauge developmental maturity in neonatal intensive care patients and prove useful in decisions about incipient illness and about hospital discharge.

Very long apnea events in preterm infants

Apnea is nearly universal among very low birth weight (VLBW) infants, and the associated bradycardia and desaturation may have detrimental consequences. We describe here very long (>60 s) central apnea events (VLAs) with bradycardia and desaturation, discovered using a computerized detection system applied to our database of over 100 infant years of electronic signals. Eighty-six VLAs occurred in 29 out of 335 VLBW infants. Eighteen of the 29 infants had a clinical event or condition possibly related to the VLA. Most VLAs occurred while infants were on nasal continuous positive airway pressure, supplemental oxygen, and caffeine. Apnea alarms on the bedside monitor activated in 66% of events, on average 28 s after cessation of breathing. Bradycardia alarms activated late, on average 64 s after cessation of breathing. Before VLAs oxygen saturation was unusually high, and during VLAs oxygen saturation and heart rate fell unusually slowly. We give measures of the relative severity of VLAs and theoretical calculations that describe the rate of decrease of oxygen saturation. A clinical conclusion is that very long apnea (VLA) events with bradycardia and desaturation are not rare. Apnea alarms failed to activate for about one-third of VLAs. It appears that neonatal intensive care unit (NICU) personnel respond quickly to bradycardia alarms but not consistently to apnea alarms. We speculate that more reliable apnea detection systems would improve patient safety in the NICU. A physiological conclusion is that the slow decrease of oxygen saturation is consistent with a physiological model based on assumed high values of initial oxygen saturation.

REGULARIZATION OF TUNNELING RATES WITH QUANTUM CHAOS

We study tunneling in various shaped, closed, two-dimensional, flat-potential, double wells by calculating the energy splitting between symmetric and antisymmetric state pairs. For shapes that have regular or nearly regular classical behavior (e.g. rectangular or circular) the tunneling rates vary greatly over wide ranges often by several orders of magnitude. However, for well shapes that admit more classically chaotic behavior (e.g. the stadium, the Sinai billiard) the range of tunneling rates narrows, often by orders of magnitude. This dramatic narrowing appears to come from destabilization of periodic orbits in the regular wells that produce the largest and smallest tunneling rates and causes the splitting versus energy relation to take on a possibly universal shape. It is in this sense that we say the quantum chaos regularizes the tunneling rates.

Stochastic modeling of central apnea events in preterm infants.

A near-ubiquitous pathology in very low birth weight infants is neonatal apnea, breathing pauses with slowing of the heart and falling blood oxygen. Events of substantial duration occasionally occur after an infant is discharged from the neonatal intensive care unit (NICU). It is not known whether apneas result from a predictable process or from a stochastic process, but the observation that they occur in seemingly random clusters justifies the use of stochastic models. We use a hidden-Markov model to analyze the distribution of durations of apneas and the distribution of times between apneas. The model suggests the presence of four breathing states, ranging from very stable (with an average lifetime of 12 h) to very unstable (with an average lifetime of 10 s). Although the states themselves are not visible, the mathematical analysis gives estimates of the transition rates among these states. We have obtained these transition rates, and shown how they change with post-menstrual age; as expected, the residence time in the more stable breathing states increases with age. We also extrapolated the model to predict the frequency of very prolonged apnea during the first year of life. This paradigm-stochastic modeling of cardiorespiratory control in neonatal infants to estimate risk for severe clinical events-may be a first step toward personalized risk assessment for life threatening apnea events after NICU discharge.

Heart Rate Characteristics Monitoring and Central Neonatal Apnea

Background: Heart rate characteristics monitoring using the HeRO score detects reduced variability and transient decelerations that occur prior to clinical signs of neonatal sepsis, and reduces VLBW mortality when displayed to clinicians. Neonatal apnea causes HR decelerations.Aim: We tested the hypothesis that central apneas are the cause of high HeRO scores.Methods: We fashioned an algorithm to detect central neonatal apnea that efficiently removes the cardiac component of the chest impedance signal. We identified 0.9 episodes per day lasting more than 30 sec and accompanied by bradycardia and desaturation in 1837 non-ventilated days in 105 VLBW infants in the University of Virginia NICU.Results: Severe central apneas accounted for only 17% of the variance in HeRO scores but were highly significant predictors of high HeRO score even after birth weight, Apgar, and hospital day were taken into account. Individual records (Figure), though, showed inconsistent temporal association of HeRO score (shade) and the number of severe apneas (line), here in an infant who died of lung disease - some but not all, synchronized with HeRO.FigureConclusion: Frequent and severe central neonatal apneas can correlate with elevated HeRO scores, and may be partly causative. We speculate that abnormal heart rate control in sepsis may have mechanisms in addition to hypoxemia caused by central apnea.