Elementary for Watson

IBM's supercomputer is transforming medical diagnostics

Japanese oncologists were recently stumped by the case of a 60-year-old woman, whose cancer refused to respond to treatment. In desperation, they consulted an American specialist. The American took only 10 minutes to diagnose a very rare type of leukaemia, which responded to very specific treatment. That American cancer specialist also consults with the Manipal Comprehensive Cancer Centre, Bangalore, and with 21 Chinese hospitals.

The specialist is the IBM Watson Supercomputer. Its cloud-based deployment has had a transformative effect on diagnostics. In the case of the Japanese woman, it mapped the data of the patient to 20 million clinical oncology case studies within 10 minutes. The Watson Oncology system is also used by the US government and in Canada.

Watson can absorb and correlate huge data. But the killer app is that it understands and responds in natural language. So it can be asked questions, or given information by a layperson. This is via a self-learning system that uses data mining, pattern recognition and natural language processing to mimic the way the brain works. Such learning systems get smarter as they process more knowledge, much as humans get smarter by learning more. IBM has worked with the Memorial Sloan-Kettering Cancer Center and with US health insurance major, Wellpoint, to "teach" Watson about cancer. Consider the logistics. A physician gets 10-15 minutes to talk to a patient, mentally shuffle through symptoms and make a diagnosis. Now, the sum of medical knowledge doubles every few months and, by 2020, medical knowledge will double every 30 days.

There are about 15 million pages of relevant research related to cancer. About 122 new oncology papers are published every day. Human learning ability is outpaced by that knowledge explosion. In contrast, Watson takes just a few minutes to upload new data and it quickly absorbs and correlates new information. It can match tumour images from scans to its databases, which contain billions of tumour images. It can match blood reports and genetic data from a patient to millions of genetic profiles and clinical studies.

In effect, Watson can access the sum of medical knowledge and reference the accumulated experiences of many doctors and researchers. Given infallible memory and good pattern-recognition, it does faster, more accurate diagnoses and suggests high probability lines of treatments. IBM claims that it will also make health care cheaper and more efficient as it learns more. The speed of data-mining augments human intuition. For example, Watson analysed 70,000 papers on P53, a protein related to many cancers and suggested areas of research to a team at The Baylor College of Medicine, Houston (USA). The researchers then identified proteins that modify P53 in a few weeks.

About 680,000 Indians die every year from cancer, which is a leading cause of mortality. About one million new cancer cases are diagnosed every year. Similar artificial intelligence systems can be developed and deployed to tackle other diseases such as diabetes, heart disease, etc. However, before such artificial intelligence systems are extensively deployed, psychological resistance from the medical profession and patients will have to be overcome. There will also be questions about the liability for possible mistreatment. Privacy issues must also be tackled to prevent misuse of patient data.

The Watson system is proprietary but there is surely reason to work with it on a larger scale as China is doing. Indian researchers should also look to marry indigenous know-how in medicine, and computing, to create similar applications in other health care domains. This could be a force multiplier for the creaky health care system and a great way to leverage the Digital India initiative.