A student researcher wins an award for software that can diagnose ear, nose and throat disease and hopefully cut those long waiting lists, writes Dick Ahlstrom.
A brief chat with a computer may become a quick way to pre-diagnose throat and speech problems. A University College Dublin research team has developed a computerised system that analyses speech sounds as a way to identify disease.
A doctor's expert skills would be needed to confirm a pathology but the automated system may be a way to shorten waiting lists for ear, nose and throat (ENT) specialists, says PhD student Rosalyn Moran.
Working with Dr Richard Reilly in the Digital Signal Processing Group at UCD's department of electronic and electrical engineering, Moran put together a "proof of concept" for a dial-up, telephone based automated diagnostic system for laryngeal disorders.
The work is so promising that Moran last month (November 12th) received the Irish Software Association's inaugural Student of the Year Award for the most commercially viable software package. NovaUCD, the innovation and technology transfer centre at UCD, proposed her for the award.
The UCD team has been working on the "back end" voice analysis software for some time, says Moran. Her work was on the "front end", the initial telephone-based voice signal processing necessary to make the system as simple to use as possible. The user simply dials up a number that could connect to anywhere in the world and answers prompts from a recording.
"The patient dials up and then gives a set of responses to an automated system. The voice responses are then sent to a web site for off-line analysis," she says. The eventual goal is to be able to do this real-time with a near immediate response confirming the probable presence or absence of a pathology.
She worked in conjunction with Voxpilot Ltd a specialist company in the area of voice analysis and recognition systems.
The challenge was to find a way to get a computer to recognise the subtle changes to speech caused by an underlying pathology, she says. "We started with advanced signal analysis, using voice features that are familiar to speech and language therapists for recognising voice pathologies."
She wrote analysis algorithms, software that compared the patient's own speech patterns with examples of voice box pathologies from 650 speech files from the Massachusetts Eye and Ear Infirmary in the US. These provided "typical" voice pattern changes, things as subtle as disturbances to voice sound frequencies.
There are three main classifications for laryngeal problems, she says, neuro-muscular disorders; physiological disorders caused for example by growths, cysts or nodules; and a mix of these two. The new algorithms helped push diagnostic accuracy to 87 per cent for neuro-muscular disorders, Moran says. Accuracy for physiological problems reached 79 per cent and they achieved 63 per cent accuracy for the third classification.
There is still a long way to go but her efforts proved that the system has great potential. It is also able to run on a conventional computer. "We are trying to focus on the core features of the voice," Moran says. "We want this to be trusted by doctors and people working in the voice therapy area."
A working system would be a great advantage to ENT patients and doctors, she believes. Patients typically wait three months to see an ENT specialist, but with the UCD system a simple phone call could quickly identify those in need of urgent treatment.
The long waiting lists relate to the need for endoscopic examination, allowing the doctor to look directly at the voice box. The system could help eliminate those who do not require this uncomfortable procedure.
The computerised system would also benefit post-operative patients, Moran believes. Patients are often given voice exercises to strengthen the larynx, with improvement confirmed using the endoscope. The voice analysis software could also do this or pinpoint those whose voices are not improving as expected.