Computerized Eyes To Read Fuzzy Mr Is

A Picture could be worth a lot more than a thousand words. With assistance from a computer, it might even save a life. Two research groups are teaching computers to analyze and interpret magnetic resonance imaging (MRI) scans, so the electronic systems can help discover better ways to diagnose-- and treat--lung disease, strokes, and other illnesses.

At Purdue University, Carla E. Brodley, an electrical engineer, has designed a visual database devoted to lung ailments. A new patient's MRI is compared with images in the database. The computer tags the best match plus four others as a way of helping the doctor diagnose the type of lung disease and rate its severity. The physician can also call up information on the treatments and outcomes in similar cases. Additional databases are being assembled for liver, knee, and brain diseases.

At the University of Massachusetts and Baystate Medical Center, researchers are working on a similar system for treating stroke victims. They hope to train their computer to spot clogged or ruptured arteries in the brain, which can cause one kind of stroke, and to estimate the extent of the trauma.

If machine-learning techniques can be used to teach computers to become progressively better at understanding the hard-to-interpret images of MRI brain scans, electronic systems could become even more valuable in health care. The machines might turn up insights that could lead to improved procedures for minimizing stroke damage, according to Dr. A. Bernard Pleet, Baystate's chairman of neurology.