IT Helps Save Babies
To lSU MAT 120 students–welcome to this term’s blog! To all our other readers thanks for checking us out again. We’re back, and will be posting until late November. So, with that said, let’s get started with something that might interest you.
I was surprised to learn that more than 500,000 babies are born prematurely every year in the US. That probably means some of you who are reading this post were born prematurely, and survived, and some have siblings who were premature. Being born prematurely raises the risk of major health problems for the baby even death. So, this can be a big deal.
In most US hospitals, there is a special unit that evaluates and monitors premature babies. The continuous monitoring uses sensors attached to the babies, to measure things like heart rate, respiration rate, and blood pressure. For years, this data has been quickly scanned by doctors, courses of treatment were determined, and this process helped to save the health of the babies. But the process is time-consuming, and it is easily possible that subtle signs of impending health issues get missed.
Lately, computer scientists like Dr. Suchi Saria have been developing a more accurate and more helpful way to do this monitoring, using IT. They have been feeding massive amounts of old monitoring data from many premature babies into computers for analysis. Along with the monitoring data, they also feed data about how well the babies did, and what problems the babies encountered. Then, they are using the principles of machine learning to analyze the data, letting the computers “discover” patterns of association between the various pieces of data.
The amazing thing is that they are not telling the computers what specific patterns to look for. So, these computers are not programmed in the way that people usually think computers are programmed. Instead, the machines are programmed to search on their own through the mountains of data looking for any significant patterns that relate the monitoring data to the health situations of the babies. In this way the machines learn what patterns of sensor data are significant for predicting health problems of the babies. Since the computers know what the health outcomes of each baby were, they can look for any hidden patterns in the monitoring data that would have predicted those health outcomes. Doing things this way, the machines might discover patterns that were previously not known to be significant.
Everyone knows that computers can store and keep in memory huge amounts of data, but it is not generally known that they also can also be designed to analyze large amounts of data so as to learn things about that data on their own. Most people don’t know computers are capable of that kind of learning. Of course, at the present time, it all depends on someone telling the computer how to go about learning significant things.
Once the computers have learned what patterns predict important possible health problems for the babies, knowledge about those patterns can be transferred to any computer that is hooked up to the sensors on newly born premature babies. Once that computer is programmed to look for those patterns, it is constantly fed live data from the sensors, and it will look for any occurrence of the significant patterns that were discovered previously by machine learning. Then, if any of those patterns occur, the computer issues alerts to let doctors know how the babies are doing, and what problems are beginning to develop. These alerts may well come earlier than they would have if the monitoring were done in the old way, and it is less likely that something bad will be missed.
This new approach has been shown to be a more accurate and efficient way of handling the care of premature babies, saving some from serious problems by catching the hidden development of health problems early on, before they would have been detected by the old methods in which doctors and nurses would periodically check the sensor data, looking for clues about how the baby is doing.
Machine learning has many applications in other fields, from marketing to crime fighting. But I found this particular application in the field of health to be especially interesting. It is probably not too hard to imagine other applications of machine learning in the field of nursing and health, or other fields — any time it would be useful to be able to find hidden patterns in mountains of data. Roughly speaking, when you don’t know what patterns to look for, let the machine figure it out for you.
If you want to find out how to make a computer learn things on its own, study computer science. If you want to know what machine learning is like, and get some sense of how it might be used in a field like marketing, nursing, manufacturing, or whatever, get a background in some of the more technical aspects of computing.
ISU MAT 120 students, don’t forget to take your quiz about this blog post on ReggieNet. That’s the only way you can show your interest in the blog and have your evaluation of it recorded.
Non-spam, relevant comments from anyone are welcome, below.
PS: Student bloggers will begin posting here later this week.
Learn about Machine Learning at: http://robotics.stanford.edu/~nilsson/mlbook.html.
Or you might want to take a look at the scientific paper regarding the PhysiScore premature baby monitoring at: http://stm.sciencemag.org/content/2/48/48ra65.abstract.
Entry filed under: Uncategorized.