Brain hypometabolism, as evaluated by PET with 18F-fluorodeoxyglucose, is considered a sensitive biomarker of synaptic dysfunction associated with neurodegeneration in Alzheimer disease. Statistical tools for image analysis showed promising capabilities in detecting and evaluating Alzheimer-related hypometabolism. The extended application of such tools requires their consolidation by proving generalizability and reproducibility. The aim of this study was to verify the reliability of an automatic tool for the detection of Alzheimer-related hypometabolic patterns based on a Support-Vector-Machine model. The model processed metabolic data from anatomical volumes of interest also considering interhemispheric asymmetrie. The model was developed and trained on a homogeneous dataset from a memory clinic center and then tested on an independent multicentric dataset drawn from the Alzheimer's Disease Neuroimaging Initiative. The accuracy of the discrimination between patients with Alzheimer disease, in either prodromal or dementia stage, and normal aging subjects was estimated at 95.8%, after cross-validation, in the training set. The accuracy of the same model in the testing set was 86.5%. The role of the two datasets was then reversed and the accuracy in the multicentric training set, after cross-validation, was 89.8% whereas the accuracy of the same model in the independent monocentric testing set was 88.0%. The classification rate was also evaluated in all subgroups in which the samples were partitioned, including non-converter mild cognitive impairment (MCI) patients, subjects with MCI reverted to normal conditions and subjects with significant memory concern (SMC) not confirmed with neuropsychological tests. The percent of positive detections was at the level of healthy controls (around or below 10%) for SMC and reverted MCI patients. An increasing rate of positive tests was found in patients from early prodromal Alzheimer disease (77%) to dementia (91%). A low rate of positivity was found in non-converted MCI patients. The two datasets exhibited similar trends of classification rate and related score through the different subgroups. The present findings show a good level of reproducibility and generalizability of a model for detecting hypometabolic pattern in Alzheimer disease and encourage further training to deepen the characterization of metabolic patterns in the early stage of the disease.
Accuracy and generalization capability of an automatic method for the detection of typical brain hypometabolism in prodromal Alzheimer disease
De Carli F;Pagani M;
2019
Abstract
Brain hypometabolism, as evaluated by PET with 18F-fluorodeoxyglucose, is considered a sensitive biomarker of synaptic dysfunction associated with neurodegeneration in Alzheimer disease. Statistical tools for image analysis showed promising capabilities in detecting and evaluating Alzheimer-related hypometabolism. The extended application of such tools requires their consolidation by proving generalizability and reproducibility. The aim of this study was to verify the reliability of an automatic tool for the detection of Alzheimer-related hypometabolic patterns based on a Support-Vector-Machine model. The model processed metabolic data from anatomical volumes of interest also considering interhemispheric asymmetrie. The model was developed and trained on a homogeneous dataset from a memory clinic center and then tested on an independent multicentric dataset drawn from the Alzheimer's Disease Neuroimaging Initiative. The accuracy of the discrimination between patients with Alzheimer disease, in either prodromal or dementia stage, and normal aging subjects was estimated at 95.8%, after cross-validation, in the training set. The accuracy of the same model in the testing set was 86.5%. The role of the two datasets was then reversed and the accuracy in the multicentric training set, after cross-validation, was 89.8% whereas the accuracy of the same model in the independent monocentric testing set was 88.0%. The classification rate was also evaluated in all subgroups in which the samples were partitioned, including non-converter mild cognitive impairment (MCI) patients, subjects with MCI reverted to normal conditions and subjects with significant memory concern (SMC) not confirmed with neuropsychological tests. The percent of positive detections was at the level of healthy controls (around or below 10%) for SMC and reverted MCI patients. An increasing rate of positive tests was found in patients from early prodromal Alzheimer disease (77%) to dementia (91%). A low rate of positivity was found in non-converted MCI patients. The two datasets exhibited similar trends of classification rate and related score through the different subgroups. The present findings show a good level of reproducibility and generalizability of a model for detecting hypometabolic pattern in Alzheimer disease and encourage further training to deepen the characterization of metabolic patterns in the early stage of the disease.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.