Table 7A.

One-way ANOVA testing for differences in the classification performances across different GLRAM dimension parameters 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	6	1.790	0.106	0.662	0.681	1.457	0.198
	Within groups	133
Tensor-RK2	Between groups	6	1.299	0.262	0.564	0.759	2.438	0.029
	Within groups	133
Tensor-SL	Between groups	6	1.445	0.202	0.403	0.876	3.010	0.009
	Within groups	133
Tensor-TL	Between groups	6	2.169	0.049	1.094	0.369	1.590	0.155
	Within groups	133
ODF-FACT	Between groups	6	0.384	0.888	2.540	0.023	1.398	0.220
	Within groups	133
ODF-RK2	Between groups	6	1.824	0.099	0.523	0.790	1.930	0.080
	Within groups	133
Probtrackx	Between groups	6	1.385	0.225	3.800	0.002	0.817	0.559
	Within groups	133
PICo	Between groups	6	0.178	0.982	0.448	0.845	0.314	0.929
	Within groups	133
Hough	Between groups	6	0.536	0.780	0.661	0.681	1.083	0.376
	Within groups	133

Since we have seven possible dimension parameters (5–35), the degrees of freedom for Between Groups is 7-1 = 6. Moreover, since we have 20 splits for each setting of the dimension parameter, the number of degrees of freedom for the Within Groups comparison is 20x7-7 = 133. Thus, the critical F-value = 2.1674 at the α = 0.05 level, when the number of degrees of freedom = (6,133). Our H0 is that there is no significant difference among different dimension parameters. We were able to reject H0 at the nominal significance level when our computed F-value > 2.1674; this included the cases where tensor-TL was used for the task AD vs. NC; ODF-FACT and Probtrackx in task AD vs. MCI; tensor-RK2, SL in task MCI vs. NC. However, three of these cases – including tensor-TL in the task AD vs. NC, ODF-FACT for the task AD vs. MCI, and tensor-RK2 for the task MCI vs. NC – did not pass the Bonferroni correction in the post hoc comparisons. There are therefore no significant differences in the classification performance when changing the dimension parameters in these three cases. We marked the other two cases in red and the corresponding post hoc results for these situations are shown in (B).