Computer aid to ayurveda diagnosis
New Delhi, Oct. 7: A team of biologists, computer scientists and doctors has used artificial intelligence to classify the physiological makeup of individuals, the feature called prakriti in ayurveda that practitioners say influences health and disease.
The scientists, from the Council of Scientific and Industrial Research (CSIR) in New Delhi and institutions in Calcutta and Pune, say their computational approach seeks to eliminate the subjectivity that clouds public perceptions about ayurveda.
According to ayurveda, every human's physiological makeup belongs to one of the seven types of prakriti: vata (V), pitta (P), kapha (K), or their combinations VP, PK, VK or VPK.
Ayurveda practitioners believe that prakriti influences one's predisposition to illnesses and response to therapy. They seek to classify patients' types on the basis of their body size, skin appearance, behaviour, bowel habits and even food preferences.
Scientists say that genome studies over the past decade have broadly seemed to support these claims.
Now, ayurveda specialist Bhavana Prasher at the CSIR Institute of Genomics and Integrative Biology, New Delhi, and her colleagues have shown that machine learning, a ubiquitous tool in the current quest for intelligent machines, can accurately label people as V, P or K types.
Prasher and her colleagues have through a series of studies since 2008 shown that people classified as V, P or K have subtle genetic differences that might influence their risk for bleeding, clot formation, obesity and heart attacks, and capacity to tolerate low-oxygen conditions.
"While those earlier studies helped reveal a link between the traditional concept of prakriti and genomics, we know that identifying prakriti can be tricky," said Mitali Mukherji, a genome biologist and colleague of Prasher for more than a decade.
She said that practitioners' experience and skills determine whether they are able to assign the correct prakriti.
In their new work, the scientists "trained" a computer to classify prakriti using a sample of prakriti labels assigned to 147 healthy individuals by qualified ayurveda practitioners. The practitioners assigned a prakriti to each person on the basis of information extracted from 133 questions relating to physiology and lifestyle.
Mukherji says the work wouldn't have been possible without the unusual backgrounds of the team members, who included statistics specialist Pradeep Tiwari, computational biologist Rintu Kutum, and Tavpritesh Sethi, a doctor trained in computer science research.
"It required exchanges between ayurveda, genomics, medicine and computer science," Mukherji said.
Saurabh Ghosh from the Indian Statistical Institute in Calcutta and ayurvedic practitioners and health researchers from the King Edward Memorial Hospital in Pune contributed to the research.
The researchers found that the machine learning technique could classify the prakritis of a different set of individuals reasonably well, the accuracy levels ranging from 79 per cent to 100 per cent, compared with the prakriti types assigned independently by two sets of qualified practitioners. The findings have been published in the journal PLOS One.
"Such a formalised method of assigning prakriti will help clinical practice and research," said Prasher. "A computer could help assign the correct prakriti and traditional practitioners could use that information for diagnosis and therapeutic decisions."
A rapid computer-generated classification of prakriti types may also help scientists explore further the connections between illnesses and the human genetic and physiological makeup, Mukherji said.