Table 5A.

One-way ANOVA test on the classification performances across different numbers of PCs for nine tractography algorithms in three diagnostic tasks.

		Degrees of freedom	AD vs. NC		AD vs. MCI		MCI vs. NC
			F	Sig.	F	Sig.	F	Sig.
Tensor-FACT	Between groups	11	1.312	0.219	1.912	0.039	0.600	0.828
	Within groups	228
Tensor-RK2	Between groups	11	3.388	0.000	2.065	0.024	0.299	0.986
	Within groups	228
Tensor-SL	Between groups	11	0.348	0.973	4.128	0.000	0.826	0.614
	Within groups	228
Tensor-TL	Between groups	11	0.770	0.670	2.886	0.001	0.649	0.786
	Within groups	228
ODF-FACT	Between groups	11	0.620	0.811	2.250	0.013	0.331	0.978
	Within groups	228
ODF-RK2	Between groups	11	0.508	0.897	1.142	0.330	2.083	0.022
	Within groups	228
Probtrackx	Between groups	11	0.260	0.992	4.908	0.000	2.641	0.003
	Within groups	228
PICo	Between groups	11	0.053	1.000	0.836	0.604	1.074	0.383
	Within groups	228
Hough	Between groups	11	0.541	0.874	2.417	0.007	0.653	0.782
	Within groups	228

Since we have 12 possible numbers of PCs (10∼150), the number of degrees of freedom for the Between Groups comparison is 12-1 = 11. Moreover, since we have 20 splits for each number of PCs, the number of degrees of freedom for the Within Groups comparison is 20 x 12-12 = 228. Thus, the critical F-value = 1.8308 at the α = 0.05 level with degrees of freedom = (11,228). Our null hypothesis, H0, is that there is no significant difference among different numbers of PCs. We can only reject H0 when our computed F-value > 1.8308, these situations are been marked in red. The corresponding post hoc comparison results in these situations are shown in (B).