Aybüke Handan Şişman, Turkey
Hacettepe University Faculty of Medicine PsychiatryAuthor Of 1 Presentation
MR MORPHOMETRIC MEASURES MAY PREDICT THE SEVERITY OF NEUROPSYCHIATRIC SYMPTOMS IN THE ELDERLY
Abstract
Aims
Neuropsychiatric symptoms are frequent in dementia. Occurrence of neuropychiatric symptoms and their severity might be associated with certain morphological changes. MRI measures might be used to demonstrate the risk of neuropsychiatric symptoms.
Methods
Neuropsychiatric features as assessed with Neuropsychiatric Inventory, Modified Mini Mental State (3MS) and MRI data were utilized. We evaluated classification performance of volumetric data in patients with no/mild vs. moderate/severe neuropsychiatric features. We applied both random forest (RF) using hold-out validation approach and SVM. Then we examined the predictive performance of the model using performance measures. Lastly, using residuals approach we determined the variables with the highest predictive value. The study was funded by TUBITAK 214S048, Psychiatry Associaton of Turkey and Hacettepe University.
Results
Data of a total of 175 patients over the age of 55 were analyzed. PCA revealed volume, thickness and ventricle sizes as three components. When these were combined with the clinical data including age, gender, education and 3MS scores in order to predict neuropsychiatric symptom severity, performance of RF was proved to be higher than SVM with an accuracy of 0.93 (0.84-0.98). 3MS scores, volumes, thickness, age and education predicted the symptom severity. Potentially predictive MR parameters as determined via RF were volumes of left parahippocampus, amygdala, insula and entorhinal cortex as well as thickness of left and right parahippocampus, left and right medial orbitofrontal cortices, right entorhinal cortex and left temporal pole.
Conclusions
Certain MRI measures may be useful in predicting the development of neuropsychiatric symptoms.