Robotic Refactoring the Workplace
The
Role of Technology in Robotics
Position Sensors for Robotics
Robot-Proof: Higher Education in the Age of Artificial Intelligence
Joseph E. Aoun, President, Northeastern University
From Robotics to Big Data, What the Future Holds for Manufacturing
Karl Rosenblum, Head of Global Capacity and Risk Strategy, Alcon
Robots: A New Source of Business Data
John Santagate, Research Director, Service Robotics, IDC
Machine Learning and Artificial Intelligence: Revolutionizing the...
Kali Durgampudi, VP of Innovation, Mobile Architecture, R&D, Healthcare Solutions, Nuance Communications
Thank you for Subscribing to CIO Applications Weekly Brief

Deep Learning and Future of Healthcare
Sanjib Basak, Former Director of Data Science & Artificial Intelligence, Carlson Wagonlit Travel


Sanjib Basak, Former Director of Data Science & Artificial Intelligence, Carlson Wagonlit Travel
In this context, it is important to mention the relationship of deep learning and AI. Deep learning is a specialized area within AI which incorporates machine learning algorithms, neural network architectures and large volume of data to deliver some powerful results. We, human, perceive AI as an unnatural superpower and most often set a very high expectation about AI. Deep learning separates “facts” of AI from “fictions” and shows us what is possible in reality and what is not with its ingredients—data and algorithms.
In the medical field, although we are capturing massive amount of patient data for past couple of years, deep learning is mostly being used for image or text data analysis, so far.
In the study, the researchers used de-identified EHR data from two US academic medical centers– University of California, San Francisco (UCSF) and the University of Chicago Medicine (UCM). They used data like patient demographics, provider orders, diagnoses, procedures, medications, laboratory values, vital signs, and nursing flowsheet data from all inpatient and outpatient encounters, free-text medical notes, 30-day unplanned readmission, prolonged length of stay, and patient’s final discharge diagnoses data. Data consisted of 216,221 adults for 4-7 years patient history data representing 46B+ data points.
Researchers used three deep learning neural network model architectures— Recurrent neural networks (long short-term memory (LSTM)), Attention based TANN, and a neural network with boosted time-based decision methods. Results were compared against traditional statistical predictive models which were set as baseline. Model accuracies for inpatient mortality prediction increased by 10 percent, unexpected readmissions within 30 days prediction increased by 8 percent, long length of stay increased by 10 percent, and discharge diagnose code prediction at the time of admission increased by 10 percent. In most of the cases the new model reached about of 90 percent accuracies. With better prediction of inpatient mortality, the false alerts for death decreased by 50 percent. By better prediction of unexpected readmission or long length of stay cases, doctors can recommend preventive actions and patient can follow that to prevent disaster and thereby improving quality of life and decreasing cost of care.
Deep learning has a tremendous future ahead in the field of medicine. We are barely scratching the surface, making some rudimentary improvements in various areas within healthcare. With standardization of data, better access of all data like EHRs, claims, and medication details, researchers and doctors will be able to make better prediction about human diseases and patients can take preventive measures much ahead. It won’t be surprising to see, in near future, that average human life expectancy has increased by 20 years which been made possible by advanced AI techniques and deep learning.
I agree We use cookies on this website to enhance your user experience. By clicking any link on this page you are giving your consent for us to set cookies. More info