When will you die? Computer prediction helps doctors

17 มกราคม 2560

3D virtual heart built using Artificial intelligence (AI) & machine learning used to predict time of heart failure so aggressive treatment can be chosen if necessary.

MEDICINE, SCIENCE & COMPUTER SCIENCE

When will you die? Computer prediction helps doctors

A computer program tells you when you will die.

This might seem a little scary, like something straight out of a science fiction movie.

This is exactly what doctors in UK have done in a recent study published in the medical journal Radiology (read here).

WHEN YOU WILL DIE PREDICTIONS: WHY ARE THEY USEFUL?

High blood pressure in the lungs (pulmonary hypertension) damages the heart. Within five years of diagnosis, one third of people diagnosed with this medical condition will be dead.

Knowing patient life expectancy is critical for doctors to choose the right treatment which might run from a regimen of drugs to more aggressive injections straight into the blood stream or even a lung transplant.

USING ARTIFICIAL INTELLIGENCE & MACHINE LEARNING TO PREDICT LIFE EXPECTANCY

(Source: Wikipedia)

Machine learning is an artificial intelligence (AI) technique in computer science that consists of computer programs that learn from large amounts of data (big data). The programs are given large amounts of data and from this patterns are identified and data- driven predictions and decisions are made (see here).

The machine learning program is given the MRI scans of hearts and blood tests from 256 patients along with eight years of patient medical records. In the MRI scans the movement of 30,000 different points in the beating heart are measured during each a heartbeat.

A 3D virtual heart that replicate the motions of the heart as it contracts is created for each patient from the MRI scans.

Looking five years into the future, the program learns to predict the heart abnormalities that lead to the death of patients.

Whereas doctor's successfully predict the one year survival rates of patients with about 60% accuracy, the machine learning program can do better than doctors and predict one year survival with 80% accuracy.

Those most in danger fo dieing soon are given more aggressive and intensive treatments such as a lung transplant.

Before the study results can be employed more widely in hospitals, more tests will be made in different hospitals. There are also plans to extend the technique to other forms of heart failure such as cardiomyopathy to determine which patients need a pacemaker.