Diagnosis tool for doctors on the way

Scientists and businessmen at Karishma Software have been working for three years trying to develop an artificial intelligence tool which when ready will provide doctors a "decision support system" to make diagnosis fail-safe.

A similar attempt in the U.S. took over 25 years, $70 million and nearly 30 Ph.Ds to complete it. Impressed by their work, the Technology Information, Forecasting and Assessment Council (TIFAC) of the federal government's department of science and technology (DST), decided to collaborate with the firm to help scale up the project in which other leading federal government scientific institutions are also collaborating.

The head of the Home Grown Technology division of DST, Sajid Mubashir, said Karishma's work was "risky but exciting" and its involvement would be to the extent of helping the project become commercially viable. According to Mubashir it will take three more years for the product to be marketed, adding it had potential to change the way doctors work. Each time a doctor deals with a patient, he has to figure out the diagnosis working through symptoms the disease expresses in the patient. There are over 19,000 symptoms which have over 56,000 discrete expressions for the 620 internal medicine diseases known as of now. For an ordinary person, this means every patient's symptoms can indicate many different diseases.

The researchers said for example, combined symptoms of fever, cough and body ache could indicate over 100 diseases. A collection or group of geographical, epidemological and social factors which, along with a doctor's famous "intuition," guide or lead the physician to suspect one particular disease over another. This must be verified further through tests.

Karishma's artificial intelligence (AI) tool promises to help physicians by eliminating the guesswork involved. The software being developed is based on a mathematical tool called the Bayesian Probabilistic Belief Networks, developed over the last few decades at MIT, Stanford, UCLA and Berkeley in the US. It has powered advanced AI equipment like the Mars Rover as well as simple things like the predictive 'Help' of MS Word.

The computer science department of the University of Hyderabad, called the Bayesian framework "elegant and robust" for tackling such problems and said it was impressed by the work done by the researchers. The two main problems for developing a decision support system for medical diagnosis, were data mining and adapting the mathematical tool into a practical software. This would require extensive medical knowledge combined with smart software skills, Bapi Raju Surampudi of the department said here.

If successful, the final product will not only be able to list out all known diseases which match the given group of symptoms, but also provide the percentage likelihood of its occurrence along with the reasons for that. It will not only tell the doctor what the probable disease is, but will also explain why one disease has 28 per cent probability of occurrence and another only two per cent.