Identification of asphyxia in newborns using gpu for deep learning

With the rapid advancement in technology, we still observe a significant amount of deaths of children under the age of five years. Majority of these deaths worldwide can be attributed to various medical conditions out of which three are very significant: birth asphyxia, preterm and infections. Birth asphyxia (perinatal asphyxia) is a medical condition which is characterised by abnormal breathing patterns in a newborn child which may eventually lead to irrevocable damage to the brain or if neglected can prove to be fatal. In most cases, the condition is diagnosed after the newborn has suffered considerable damage. This is because birth asphyxia can be conclusively determined only by medical examinations by skilled people who are not easily available in poor areas. In this paper, we aimed at such sections of the society and have tried to build up on a machine learning approach by which asphyxia can be determined at its early stages. It involves diagnosing the condition by observing patterns in the child's cry and subjecting it through different layers of a neural network built on a database of previously recorded samples of affected children. The software used is NVIDIA DIGITS and the highest sustained accuracy achieved by us is 94%. It is an economic method which does not require highly skilled labour and can be adopted in almost all sections of the society if adequate resources are available.