Optimizing neural networks for medical data sets: A case study on neonatal apnea prediction.

OBJECTIVE: The neonatal period of a child is considered the most crucial phase of its physical development and future health. As per the World Health Organization, India has the highest number of pre-term births [1], with over 3.5 million babies born prematurely, and up to 40% of them are babies with low birth weights, highly prone to a multitude of diseases such as Jaundice, Sepsis, Apnea, and other Metabolic disorders. Apnea is the primary concern for caretakers of neonates in intensive care units. The real-time medical data is known to be noisy and nonlinear and to address the resultant complexity in classification and prediction of diseases; there is a need for optimizing learning models to maximize predictive performance. Our study attempts to optimize neural network architectures to predict the occurrence of apneic episodes in neonates, after the first week of admission to Neonatal Intensive Care Unit (NICU). The primary contribution of this study is the formulation and description of a set of generic steps involved in selecting various model-specific, training and hyper-parametric optimization algorithms, as well as model architectures for optimal predictive performance on complex and noisy medical datasets. METHODS: The data used for the study being inherently complex and noisy, Kernel Principal Component Analysis (PCA) is used to reduce dataset dimensionality for the analysis such as interpretations and visualization of the dataset. Hyper-parametric and parametric optimization, in different categories, are considered, including learning rate updater algorithms, regularization methods, activation functions, gradient descent algorithms and depth of the network, based on their performance on the validation set, to obtain a holistically optimized neural network, that best model the given complex medical dataset. Deep Neural Network Architectures such as Deep Multilayer Perceptron's, Stacked Auto-encoders and Deep Belief Networks are employed to model the dataset, and their performance is compared to the optimized neural network obtained from the parametric exploration. Further, the results are compared with Support Vector Machine (SVM), K Nearest Neighbor, Decision Tree (DT) and Random Forest (RF) algorithms. RESULTS: The results indicate that the optimized eight layer Multilayer Perceptron (MLP) model, with Adam Decay and Stochastic Gradient Descent (AUC 0.82) can outperform the conventional machine learning models, and perform comparably to the Deep Auto-encoder model (AUC 0.83) in predicting the presence of apnea in neonates. CONCLUSION: The study shows that an MLP model can undergo significant improvements in predictive performance, by the proposed step-wise optimization. The optimized MLP is proved to be as accurate as deep neural network models such as Deep Belief Networks and Deep Auto-encoders for noisy and nonlinear data sets, and outperform all conventional models like Support Vector Machine (SVM), Decision Tree (DT), K Nearest Neighbor and Random Forest (RF) algorithms. The generic nature of the proposed step-wise optimization provides a framework to optimize neural networks on such complex nonlinear datasets. The investigated models can help neonatologists as a diagnostic tool.

Clinico-epidemiological study of low birth weight newborns in the Eastern part of Nepal.

Low birth weight (LBW) remains an important cause of newborn morbidity and mortality. A hospital based prospective and descriptive study was conducted at Paediatric wards, Nursery, Neonatal intensive unit (NICU) and Post natal ward during period of June to October 2010 to note the clinico-epidemiological profile of Low Birth Weight (LBW) newborns. Incidence of the LBW babies in our hospital was 14.45%; more than four fifth (82.2%) baby's mothers were primigravida. Eighty two percent mothers had unbooked pregnancies. Twenty and half percent LBW babies were twins. The mean duration of hospital stay of the subject was 7.4 (± 5.5) days. The mean birth weight of LBW babies was 1648 (± 344) grams. Clinical sepsis, non physiological jaundice and hypoglycaemia were the three most common complications of LBW babies. Antibiotics, oxygen and phototherapy were the three commonest modes of therapy. Majority of children (82.0%) improved and were discharged. Birth weight and gestational age were significantly different between survivors and babies who expired. Primigravida and lesser antenatal visits were important risk factors for Low birth weight babies. Birth weight, gestational age, apnoea and mechanical ventilation were the predictors of outcome.

Exposure to biological maternal sounds improves cardiorespiratory regulation in extremely preterm infants.

OBJECTIVE: Preterm infants experience frequent cardiorespiratory events (CREs) including multiple episodes of apnea and bradycardia per day. This physiological instability is due to their immature autonomic nervous system and limited capacity for self-regulation. This study examined whether systematic exposure to maternal sounds can reduce the frequency of CREs in NICU infants. METHODS: Fourteen preterm infants (26-32 weeks gestation) served as their own controls as we measured the frequency of adverse CREs during exposure to either Maternal Sound Stimulation (MSS) or Routine Hospital Sounds (RHS). MSS consisted of maternal voice and heartbeat sounds recorded individually for each infant. MSS was provided four times per 24-h period via a micro audio system installed in the infant's bed. Frequency of adverse CREs was determined based on monitor data and bedside documentation. RESULTS: There was an overall decreasing trend in CREs with age. Lower frequency of CREs was observed during exposure to MSS versus RHS. This effect was significantly evident in infants ≥ 33 weeks gestation (p=0.03), suggesting an effective therapeutic window for MSS when the infant's auditory brain development is most intact. CONCLUSION: This study provides preliminary evidence for short-term improvements in the physiological stability of NICU infants using MSS. Future studies are needed to investigate the potential of this non-pharmacological approach and its clinical relevance to the treatment of apnea of prematurity.

Manejo del SAHS mediante dispositivos de avance mandibular. Estudio preliminar / Sleep apnea-hypopnea syndrome management using a mandibular advancement device. Preliminary study

El tratamiento que ha demostrado mayor eficacia en los pacientes con síndrome de apneas-hipopneas del sueño (SAHS) es la presión positiva continua de la vía aérea (CPAP). Los mayores inconvenientes son la incomodidad y la sensación de claustrofobia, que en algunos pacientes provoca rechazo o intolerancia. Una alternativa son los dispositivos de avance mandibular (DAM), que insertados en las arcadas dentarias producen el avance de la mandíbula y de la lengua, aumentando el volumen de la vía aérea. Objetivo: Presentar nuestra experiencia en el tratamiento del SAHS mediante dispositivos de avance mandibular tipo Herbst. Metodologia: Estudio de seguimiento prospectivo desde junio de 2006 hasta enero de 2009 de 7 pacientes del Área Hospitalaria Virgen Macarena con SAHS que rechazan el tratamiento con CPAP y a los que se ofrece tratamiento con DAM. Las variables analizadas son: índice de apneas-hipopneas por hora, índice de desaturaciones por hora, intensidad subjetiva del ronquido y el test de somnolencia de Epworth, antes del tratamiento y al menos 6 meses después desde el inicio de su uso. Utilizamos el test de Wilcoxon para detectar diferencias estadísticas significativas (p<0,05). Resultados: Se observó una reducción estadísticamente significativa del índice de apneahipopneas por hora (p<0,018) y del índice de desaturaciones por hora (p<0,018), así como una reducción no significativa del ronquido y de la somnolencia. Conclusiones: El uso de DAM tipo Herbst en pacientes afectos de SAHS que rechazan el uso de la CPAP podría ser útil, mejorando clínica y funcionalmente su situación(AU)

The most effective treatment in patients with sleep apnea-hypopnea syndrome (SAHS) is CPAP (continuous positive airway pressure). The main drawback of CPAP is the discomfort and claustrophobic sensation that it causes, which elicits rejection or intolerance by some patients. A non-surgical alternative to CPAP is the mandibular advancement device (MAD), which consists of a plastic splint inserted between the dental arches to shift the jaw and tongue forward and thus increase airway volume. Objective: Report our experience with the treatment of SAHS using the Herbst mandibular advancement device. Material and method: A prospective follow-up study was carried out from June 2006 until January 2009 at the Virgen Macarena University Hospital with 7 patients with SAHS who refused treatment with CPAP and were treated with the mandibular advancement device. The outcome variables analyzed were: apnea-hypopnea disruptions per hour index, desaturations per hour index, subjective intensity of snoring, and the Epworth Sleepiness Scale. Variables were evaluated pre-treatment and at least once 6 months after initiation MAD use. The Wilcoxon test for paired samples was used to detect statistically significant differences (p<0.05). Results: A statistically significant reduction in the hourly indices of apnea-hypopnea disruptions (p<0.018) and desaturations (p<0.018) was observed, as well as a statistically nonsignificant reduction in snoring and sleepiness. Conclusions: Use of the Herbst mandibular advancement device in patients with obstructive sleep apnea syndrome who refuse CPAP may be helpful as it improves the clinical and functional parameters of the condition(AU)

Anesthesia and apnea. Perioperative considerations in the former preterm infant.

Former preterm infants younger than 44 weeks postconceptual age are at increased risk for developing postoperative apnea and PB. When surgery cannot be deferred until the infant is developmentally more mature, several measures should be taken to minimize the risk of ventilatory dysfunction. First, outpatient surgery is not advisable for infants younger than 44 weeks postconceptual age. All infants should be admitted to the hospital and monitored for apnea and bradycardia for at least 12 to 18 hours after surgery. Second, we recommend the use of intravenous caffeine base 10 mg/kg in all infants at risk for postoperative apnea following general anesthesia. Preliminary studies of a small number of patients indicate that spinal anesthesia without sedation is associated with less apnea than is general anesthesia or spinal anesthesia with ketamine sedation. This option warrants further consideration. Infants with anemia of prematurity, generally a benign condition, are at increased risk for postoperative apnea. It is therefore preferable to delay elective surgery and supplement the feeds with iron until the Hct is above 30%. When surgery cannot be deferred, anemic infants must be observed and monitored carefully in the postoperative period.