Table 1.
Recent Procedures for Diagnosing an Individual as having Alzheimer’s Disorder or not using anatomical MRI data. Two studies used PET imaging data in addition to MRI data. Features extracted from these imaging data typically consisted of voxel-wise maps intensities across the entire brain, therefore yielding a high dimensional feature space. In all studies, the decision boundary within the feature space was learned using support vector machine (SVM), and the classification accuracy was typically assessed using leave-one-out (LOO) cross-validation. SN=sensitivity; SP=specificity; MCI=mild cognitive impairment; MCIc=MCI participants who converted to AD; MCInc=MCI participants who did not convert to AD; GM=gray matter; WM=white matter; CSF=cerebrospinal fluid; ROI=region of interest; PET=positron emission tomography; FDG-PET=18fluorodeoxyglucose-PET; HC=healthy controls; SPHARM=spherical harmonics; SPHARM-PDM=spherical harmonics-point distribution model; LOO=leave-one-out.
| Study | Participants | Features | Performance |
|---|---|---|---|
| Fan, 2008(Fan et al., 2008) |
56 with AD, 88 with MCI, and 66 healthy controls |
Voxel-wise maps of localized volumes in (1) GM, (2) WM, and (3) CSF. |
LOO cross-validation. (1) AD vs HC: Accuracy=94.3%; (2) MCI vs HC: Accuracy=81.8%; (3) AD vs MCI: Accuracy=74.3%. |
| Klöppel, 2008(Klöppel et al., 2008b) |
Group 1: 20 with AD and 20 healthy; Group 2: 14 with AD and 14 healthy. with histopathological confirmation; and Group 3: 33 with mild AD and 57 healthy. |
Voxel-wise measure of gray matter intensity |
LOO cross-validation. Group 1: SN=95%; SP=95%; Group 2: SN=100%; SP=85.7%; Group 3: SN=60.6%; SP=93%. |
| Gerardin, 2009(Gerardin et al., 2009) |
23 with AD; 23 with MCI; 25 healthy controls |
SPHARM coefficients for the shape of hippocampus |
LOO cross-validation. (1) AD vs HC: SN=96%; SP=92%; (2) MCI vs HC: SN=83%, SP=84%. |
| Hinrichs, 2009(Hinrichs et al., 2009) |
77 with AD and 82 healthy controls | Voxel-wise maps for (1) gray matter probability, and (2) FDG-PET intensity |
10-fold cross-validation. SN=78.5%; SP=81.8%. |
| Magnin, 2009(Magnin et al., 2009) |
16 with AD, and 22 healthy controls | Percent gray matter within 90 ROIs across the brain. |
Bootstrap validation. SN=91.5%, SP=96.6%. |
| Misra, 2009(Misra et al., 2009) |
103 with MCI who converted to AD (MCIc); and 76 MCI who did not convert to AD (MCInc). |
Voxel-wise maps of variations in localized volumes within (1) GM, (2) WM, and (3) CSF. |
LOO cross-validation. Accuracy (number of correctly classified) >75%. |
| Cuingnet, 2011(Cuingnet et al., 2011) |
162 elderly controls (HC); 137 with AD; 76 with MCI who converted to AD (MCIc); 134 with MCI who did not convert to AD (MCInc) |
(1) voxel-wise probability maps of GM, WM, and CSF; (2) Voxel-wise measure of cortical thickness; (3) Hippocampal volume and shape using spherical armonics |
LOO cross-validation. SN and SP varied across 28 different methods tested: (1) HC vs AD: SN>63%; SP>77%. The best SN=81% and SP=95% (Table 4(Cuingnet et al., 2011)); (2) HC vs MCIc: SN>22%; SP>73%. The best SN=65% and SP=94% (Table 5(Cuingnet et al., 2011)); (3) MCInv vs MCIc: SN>0%, SP>61%. The best SN=57% and SP=78% (Table 6(Cuingnet et al., 2011)). |
| Oliveira, 2010(Oliveira et al., 2010) |
14 with AD, and 20 healthy controls | Volume of 45 brain regions delineated automatically using FreeSurfer |
LOO cross-validation. SN=92.8%; SP=85%. |
| Zhang, 2011(Zhang et al., 2011) |
51 with AD, 99 with MCI, and 52 healthy controls |
For 93 ROIs, average measures of GM volume, and intensity in PET; and three proteins in CSF. |
10-fold cross-validation. (1) AD vs HC: SN=93%, SP=93.3%;(2) MCI vs HC: SN=81.8%, SP=66%. |