12 January 2009

Comparison of diagnositic techniques

Accuracy of dementia diagnosis — a direct comparison between radiologists and a computerized method
Dementia News (Alzheimer's Australia): 8 January 2008
There is considerable interest in developing machine-learning techniques to allow for neuroimaging-based diagnosis. Such methods would allow fully automated, standard computer-based clinical decisions that are unbiased by varying levels of radiological expertise. Support vector machines are able to accurately classify brain images and have been used to separate Alzheimer's disease from normal ageing and from frontotemporal lobar degeneration.

Stefan Klöppel and colleagues* report on their study in which six radiologists from various facilities around the world and with different levels of experience were given the same brain scans and information that had been used by support vector machines to make diagnoses. The diagnoses made by the radiologists were compared with those made by the support vector machines.

In one set of cases the support vector machines correctly classified 95% of sporadic Alzheimer's disease and controls into their respective groups; the radiologists correctly classified 65–95% of scans. In another set of cases of sporadic Alzheimer's disease support vector machines correctly classified 93% of cases, whereas the radiologists correctly classified between 80% and 90%. In separating patients with sporadic Alzheimer's disease from those with frontotemporal lobar degeneration support vector machines accurately classifyied 89% of cases compared to 63% to 83% correctly classified by the radiologists. Notably, when the radiologists reported a high degree of diagnostic confidence they were always accurate.

These results indicate that the capacity for support vector machines to accurately classify typical Alzheimer's disease-associated scans compares favourably with that of well-trained neuroradiologists. Importantly, support vector machines require no expert knowledge and can be exchanged between centres for use in diagnostic classification. Computerised diagnostic methods could be especially valuable in clinical medicine in areas where neuroradiologists are not available to examine scans.

The study is published in Brain 2008: 131(11):2969-2974

*The authors of this study are from University Clinic Freiburg, Germany; University College London; Mayo Clinics (Arizona and Rochester, USA); Institute of Neurology, UK; Brighton & Sussex Universities Hospital NHS Trust; Austin Health, Heidelberg; University of Melbourne, Australia; Ecole Normale Supérieure, France; and the Laboratory of Neuroimaging, IRCCS Santa Lucia, Italy.

No comments:

Post a Comment

Alzheimer's Australia NSW

Alzheimer's Australia NSW
Alzheimer's Australia NSW

Latest headlines from Library News

Library News

Total Pageviews