Healthcare organizations around the world are now investing in electronic health records (EHR). From simple prescriptions to complex structural images, every aspect of patients walk through the hospital is recorded. The EHR revolution started about a two decades ago in the US and has rapidly been adopted both in urban as well as rural areas (http://dashboard.healthit.gov/quickstats/quickstats.php). Other countries in Europe and Asia are in the process of adopting EHR as well. With access to such massive patient data, one can ask – How can we use this massive amount of data for improving patient outcome?
This is where health care can borrow methodologies from Machine Learning (ML). ML has been widely used in many other domains, from financial systems to prediction of faults in manufacturing to modern automotive industry. However, the application of ML to healthcare is still in its nascent state. Multiple research groups from various studies have applied ML techniques for segmentation, diagnosis to text analytics in radiology (Wang S, 2012). For cancer, ML is finding its use in prediction of outcomes of therapy as well as in mining large genomics data to find specific gene signatures.
Most of the above studies have been accomplished using supervised learning – A method where annotated training examples are available for the specific question at hand. At the heart of any ML algorithm is to learn the specific pattern of the training data and create an effective separation between the training classes such it would be useful in predicting other similar patterns (a concept known as generalizability). Furthermore, it also identifies key variables that contribute to the separation. Such a ranking could be useful as a potential biomarker for the specific question at hand.
However, there are many cases where specific annotation cannot be achieved or are not accurate. Typically, data scientist resort to unsupervised learning – A method to learn internal data structure. At HealthNextGen, we built a platform for narrating the patient story utilizing an amalgamation of supervised and unsupervised/other deep learning techniques. We continue to evolve, applying machine learning for clinical research!
More to follow...