Prashanthi Vemuri, Ph.D., a senior research fellow at the Mayo Clinic aging and dementia imaging research lab and the lead author of the study discusses using structural MRI to help accurately diagnose dementia in patients. This study was presented at the Alzheimer’s Association International Conference on Alzheimer’s Disease.
doneAbstract
BACKGROUND:
In this study, we developed a framework for MRI-based differential diagnosis of three common neurodegenerative disorders: Alzheimer disease(AD), Frontotemporal lobar degeneration(FTLD), Lewy body disease(LBD) using Structural MRI. The rationale is that if each neurodegenerative disorder is examined independently in pathologically confirmed “pure” dementia cases, they will be associated with a unique pattern of atrophy specific to the disease process which can then be applied prospectively for differential diagnosis of new incoming patients.
METHODS:
Pathologically confirmed subjects with only a single dementia pathology and MRI at the time of clinical diagnosis of dementia were identified from the Mayo Clinic database. Numbers of subjects were AD(44), LBD(18), FTLD(28) (FTLD subjects were restricted to behavioral variant clinical sub-type).
Step-wise description of the differential diagnosis system:
1) Computation of STAND-Maps: Overall pattern of gray matter atrophy in each subject based on their MRI scan is estimated and labeled “STructural Abnormality iNDex” or STAND-Map.
2) STAND-Maps specific to each dementia disease process: STAND-Maps of all the pathologically-confirmed subjects are computed. Regions of gray matter loss specific to each disease process when compared to pathology-confirmed cognitively normal subjects (pathological controls) are shown (figure).
3) Classifiers separating each dementia type from the other two dementia groups: Classifiers that detect each dementia type independently by separating it from the other two are constructed e.g. in the figure AD-related disease process is separated using a classifier-plane from LBD and FTLD. We used linear-support vector machine(SVM) classifiers with recursive feature reduction(SVM-RFE). Leave-one-out cross-validation was used to optimize classifier parameters and to compute area under the ROC(AUROC) for each classifier.
RESULTS: STAND-Maps for each dementia disorder when compared to pathological controls mirror the known anatomic distribution of pathological neurodegeneration in the literature for AD, LBD and FTLD. AUROC for discriminating AD, LBD and FTLD was 0.78, 0.80 and 0.90 respectively.
CONCLUSIONS: STAND-Map of each dementia syndrome in pathologically confirmed cases is unique and may be very useful for differential diagnosis of new incoming subjects. The proposed framework establishes a direct relationship between a structural abnormality biomarker (MRI) and the “gold standard” of pathology. This information is then incorporated to provide differential diagnosis in new incoming dementia patients.
ACKNOWLEDGEMENTS: This work was supported in part by NIH grants R01 AG11378, K23 AG030935, P50 AG16574, U01 AG06786, Robert H. Smith Family Foundation Research Fellowship, Alexander Family Alzheimer’s Disease Research Professorship and Opus building grant NIH C06 RR018898.
AUTHORS: Prashanthi Vemuri, Kejal Kantarci, Matthew L. Senjem, Jeffrey L. Gunter, Jennifer L. Whitwell, Keith A. Josephs, David S. Knopman, Bradley F. Boeve, Tanis J. Ferman, Dennis W. Dickson, Ronald C. Petersen, Clifford R. Jack Jr.
