Fig. 1. Flowchart of ELM classifier.

Fig. 2. The number of significant brain regions when changing scale factor τ, embedding dimension m and similar factor r in the MSE model: (a) τ = 1; (b) τ = 2; (c) τ = 3; (d) τ = 4; (e) τ = 5; (f) τ = 6; (g) average number of significant brain regions over the scale factor τ (p < 0.05). Here, the similarity factor r changed from 0.05 to 0.6 with a step of 0.05 and parameter of m = 1 (redline) was fixed and m = 2 (blueline) respectively (p < 0.05).

Fig. 3. Sorting effects of similarity factor rby ROC and AUC value in a single brain region when the similarity factor ris setfrom 0.05 to 0.6 with a step of 0.05 and parameters of m = 1, τ = 5 fixed in the MSE model: (a) PCG.L:left posterior cingulate gyrus; (b)STG.R: right superior temporal gyrus; (c) MOG.R: right middle occipital gyrus; (d) PoCG.R: right postcentral gyrus. In above two planes such as (a) and (b), a single sensitive brain area to cognitive testing score could be characted by both ROC beyond the reference line and great AUC value, therefore, be employed as a functional biomarker in this study. In reverse, a single insensitive brain area could be characted by both ROC around the reference line and small AUC value in below two planes such as (c) and (d).

Fig. 4. Sorting effects of scale factor τby ROC and AUC value in a single brain region when the scale factor τis set from 1 to 6 with a step of 1 and the optimization parameters of m = 1 and r =0.5 fixed in the MSE model: (a) PCG.L: left posterior cingulate gyrus; (b) STG.R: right superior temporal gyrus; (c) MOG.R: rightmiddle occipital gyrus; (d) PoCG.R: rightpostcentral gyrus.In above two planes such as (a) and (b), a single sensitive brain area to the cognitive testing score could be characted by both ROC beyond the reference line and great AUC value, therefore, be employed as a functional biomaker in this study. In reverse, a single insensitive brain area to the cognitive testing score could be characted by both ROC around the reference line and small AUC value in below two planes such as (c) and (d).

Fig. 5. Respective ROC and AUC value of a single indicative brain region, a single non-indicative brain regions and a total of 9 indicative brain regions at the optimization parameters of m = 1, r = 0.5 and τ = 5 in the MSE model: (a) A single indicative brain region. A total of 9 indicative brain regions. (b)a single of non-indicative brain region. A total of 9 non-indicative brain regions are randomly chosen; (c) a total of 9 indicative brain regions all together.

Fig. 6. Inter-group MSE values change with the parameter of scale factor τ (from 1 to 5 with a step of 1) in a total of 9 indicative brain regions: (a) CAL.R; (b) SFGmed.L; (c) PCG.L; (d) ITG.L; (e) STG.R; (f) CUN.R; (g) PUT.R; (h) HIP.R; (i)TPOmid.R. (*p < 0.05).

Fig. 7. Classification accuracy tested by ELM. Two groups of samples with excellent cognitive scores (Category 1) and poor cognitive scores (Category 0) could be classified at a sorting rate of about 80%.