Abstract #1604

Dyrby T, Rostrup E, van Straaten E, Barkhof F, Ropele S, Hansen L, Waldemar G

Since age-related white matter changes are being recognised as a marker of vascular pathology, and are associated with motor and cognitive deficits, their automated quantification may be of great importance in multicentre studies. The robustness of an artificial neural network (ANN) segmentation is investigated and compared to manual delineation in a large dataset. When scan quality was optimal high correlation coefficients were found (avg. of 0.89), demonstrating the cross-centre generalisability of ANN methods. Anatomical and pathophysiological expert knowledge was a minor source of discrepancy, and suboptimal data quality a major cause of discrepancy. Centre-specific training sets may give further improvement.