Abstract
Background
Structural neuro-imaging is seen as one possible surrogate biomarker for diagnosing and predicting AD. However, image processing techniques have so far not been able to accurately predict conversion to AD in patients with MCI[1]. In this study we investigated the possibility of using patterns of cortical thinning for predicting AD in a group of subjects with MCI.
Methods
1.5T T1w MRI baseline data were selected from the ADNI database (AD = 150, MCI = 325, controls = 190). Cortical thickness was automatically calculated using FACE[2] and mapped to an average cortical surface of 100 AD patients (template surface) [3]. Four MCI subgroups were constructed: MCIs that converted to AD within 12 months (MCI12 = 56), 24 months (MCI24 = 105), and 36 months (MCI36 = 117) from baseline, and MCIs that did not progress to AD during three years (MCI non-converters = 208). A statistical parametric map of differences in cortical thickness between MCI12 and MCI non-converters was constructed, thresholded at p = 0.01, and filtered (figure). The resulting thinning pattern on the template cortical surface was used as a mask to sample cortical thickness measurements with the purpose of classifying MCI subjects into converters and non-converters using linear discriminant analysis. The converters and non-converters were divided randomly into training and test sets of equal sizes and used in the classifier. McNemar's chi-square test was used to assess whether the classifier performed better than a random classifier.
Results
The table shows correct classification rates, sensitivity, and specificity for the classifier along with results of McNemar's test (a = 0.05). The correct classification rate of converters and non-converters was above 61%. However, none of the classifications performed significantly better than a random classifier. Classification of AD, MCI, and controls performed significantly better than a random classifier (p = 0.03).
Conclusions
Using patterns of characteristic cortical thinning in MCI converters compared to MCI non-converters demonstrated promising results for the prediction of patients with prodromal AD progressing towards clinically definite AD. The classification results are better than results of comparable methods recently published[1]. Still, the sensitivity of the method needs to be increased to be clinically applicable.
References:
[1] Cuingnet et al., NeuroImage; In Press. [2] Eskildsen et al., NeuroImage 2009; 45(3):713-721. [3] Fonov et al., NeuroImage 2011, 54(1):313-327.
Structural neuro-imaging is seen as one possible surrogate biomarker for diagnosing and predicting AD. However, image processing techniques have so far not been able to accurately predict conversion to AD in patients with MCI[1]. In this study we investigated the possibility of using patterns of cortical thinning for predicting AD in a group of subjects with MCI.
Methods
1.5T T1w MRI baseline data were selected from the ADNI database (AD = 150, MCI = 325, controls = 190). Cortical thickness was automatically calculated using FACE[2] and mapped to an average cortical surface of 100 AD patients (template surface) [3]. Four MCI subgroups were constructed: MCIs that converted to AD within 12 months (MCI12 = 56), 24 months (MCI24 = 105), and 36 months (MCI36 = 117) from baseline, and MCIs that did not progress to AD during three years (MCI non-converters = 208). A statistical parametric map of differences in cortical thickness between MCI12 and MCI non-converters was constructed, thresholded at p = 0.01, and filtered (figure). The resulting thinning pattern on the template cortical surface was used as a mask to sample cortical thickness measurements with the purpose of classifying MCI subjects into converters and non-converters using linear discriminant analysis. The converters and non-converters were divided randomly into training and test sets of equal sizes and used in the classifier. McNemar's chi-square test was used to assess whether the classifier performed better than a random classifier.
Results
The table shows correct classification rates, sensitivity, and specificity for the classifier along with results of McNemar's test (a = 0.05). The correct classification rate of converters and non-converters was above 61%. However, none of the classifications performed significantly better than a random classifier. Classification of AD, MCI, and controls performed significantly better than a random classifier (p = 0.03).
Conclusions
Using patterns of characteristic cortical thinning in MCI converters compared to MCI non-converters demonstrated promising results for the prediction of patients with prodromal AD progressing towards clinically definite AD. The classification results are better than results of comparable methods recently published[1]. Still, the sensitivity of the method needs to be increased to be clinically applicable.
References:
[1] Cuingnet et al., NeuroImage; In Press. [2] Eskildsen et al., NeuroImage 2009; 45(3):713-721. [3] Fonov et al., NeuroImage 2011, 54(1):313-327.
| Original language | English |
|---|---|
| Journal | Alzheimer's & Dementia |
| Volume | 7 |
| Issue | 4, Suppl. |
| Pages (from-to) | S25, No. IC-P-042 |
| ISSN | 1552-5260 |
| DOIs | |
| Publication status | Published - Jul 2011 |
| Externally published | Yes |
| Event | Alzheimer's Association International Conference - Paris, France Duration: 16 Jul 2011 → 21 Jul 2011 |
Conference
| Conference | Alzheimer's Association International Conference |
|---|---|
| Country/Territory | France |
| City | Paris |
| Period | 16/07/2011 → 21/07/2011 |