Fig. 3

Comparison of regional specificity profiles between methods 2–4. The statistically most conservative approach (method 2, “z-min: per data point, averaged across labels”, red dashed line) yielded ideal specificity for all brain regions, i.e. it correctly assigns “no atrophy” in 100% of cases. The less conservative method 3 (“z-min: per label”, purple dashed line) also showed specificity of 100% for many brain regions, but had some drops, e.g. for the right lingual gyrus. The most liberal approach, method 4 (“z-score: per label”, golden dashed line) yielded lower specificity for all brain regions. Note that method 1 (“z-min: per data point”) is not shown here because it does not allow for labelwise assessment. See also Table 1 for the cumulative specificity values for each method