Skip to main content
. 2016 Jan 30;58:475–485. doi: 10.1007/s00234-016-1648-3

Table 2.

Tests to evaluate the WMH change assessment methods’ performance on the sample and the effect of BFC on the winner method. Description, rationale and expected outcome

Purpose Tests’ description Rationale and expected outcome
Evaluate the output and performance of the computational methods for calculating WMH volume change. (1) Annotate the performance of each method (without BFC) on each dataset on the brain regions specified by the Prins scale [29], brainstem and cerebellum and summarise the results of the visual inspection. The best method should be robust against artefacts and accurately highlight zones of increase/decrease in WMH.
(2) Calculate the correlation between the volume of WMH change by each method (without BFC) and the Prins visual rating scale. Cross-sectional results from MCMxxxVI are also evaluated against Fazekas scores as per [18]. The output from the best method should correlate highly and significantly with the output from the visual rating.
Evaluate the influence that the BFC has on the output of the winning computational method (1) Calculate the correlation between the volume of WMH change obtained by the winning method with and without BFC (the latter done also with the winning method) and the Prins visual rating scale. If the winning method is MCMxxxVI, cross-sectional results are also evaluated against Fazekas scores as per [18]. If the application of BFC is beneficial, the correlation between the output of the WMH volume change measurements when this is applied and the visual ratings should be higher and stronger than when the BFC is not applied.
(2) Calculate the correlation between the volume of WMH change obtained by the winning method with and without BFC (the latter done also with the winning method) and age. If the application of BFC is beneficial, the correlation between the output of the WMH volume change measurements when this is applied and age should be higher and stronger than when the BFC is not applied.
(3) Visually inspect the performance and results of the winning computational method when BFC images are used vs. those obtained without the previous application of this step (i.e. BFC). If the application of BFC is beneficial, the results should not differ significantly from those obtained when the original images are used, and the manual correction to the automatically obtained results should be minimal.