Table 4 Performance of our framework for NC vs. AD, subcortical vascular disease with no cognitive impairment (NCI) vs. subcortical vascular mild cognitive impairment (svMCI), and methylated MGMT (MGMT+) vs. unmethylated MGMT (MGMT-) classifications

Methods	AUC	ACC	Sensitivity	Specificity	F1-score
NC vs. AD on OASIS-3 (pretrained on ADNI)
3D CNN (SI)	0.852 ± 0.020	0.772 ± 0.014	0.710 ± 0.091	0.826 ± 0.103	0.743 ± 0.014
3D CNN (MI)	0.885 ± 0.003	0.821 ± 0.007	0.719 ± 0.023	0.912 ± 0.013	0.790 ± 0.011
Ours_3DCNN (SI+MmFE)	0.890 ± 0.005	0.831 ± 0.004	0.752 ± 0.024	0.900 ± 0.018	0.806 ± 0.008
NC vs. AD on HS Hospital (pretrained on ADNI)
3D CNN (SI)	0.751 ± 0.081	0.718 ± 0.038	0.639 ± 0.067	0.774 ± 0.043	0.639 ± 0.055
3D CNN (MI)	0.863 ± 0.076	0.839 ± 0.043	0.754 ± 0.086	0.912 ± 0.043	0.778 ± 0.046
Ours_3DCNN (SI+MmFE)	0.849 ± 0.098	0.783 ± 0.078	0.734 ± 0.085	0.810 ± 0.106	0.730 ± 0.080
NCI vs. svMCI on RJ Hospital
3D CNN (SI)	0.677 ± 0.036	0.649 ± 0.034	0.634 ± 0.079	0.676 ± 0.099	0.670 ± 0.057
3D CNN (MI)	0.713 ± 0.041	0.676 ± 0.064	0.697 ± 0.066	0.680 ± 0.149	0.708 ± 0.066
Ours_3DCNN (SI+MmFE)	0.705 ± 0.040	0.660 ± 0.051	0.684 ± 0.105	0.662 ± 0.095	0.693 ± 0.057
MGMT+ vs. MGMT- on BraTS 2021
3D CNN (SI)	0.574 ± 0.036	0.564 ± 0.022	0.539 ± 0.035	0.591 ± 0.033	0.562 ± 0.025
3D CNN (MI)	0.594 ± 0.030	0.581 ± 0.034	0.602 ± 0.057	0.556 ± 0.029	0.598 ± 0.052
Ours_3DCNN (SI+MmFE)	0.600 ± 0.023	0.593 ± 0.025	0.634 ± 0.060	0.544 ± 0.074	0.616 ± 0.048

We show the mean value and standard deviation of the 5-fold results. SI Single-modal input, MI Multi-modal input, MmFE Multi-modal feature enhanced.

Quick links

Search