Predicting clinical scores from magnetic resonance scans in Alzheimer's disease

Abstract

Machine learning and pattern recognition methods have been used to diagnose Alzheimer's disease (AD) and mild cognitive impairment (MCI) from individual MRI scans. Another application of such methods is to predict clinical scores from individual scans. Using relevance vector regression (RVR), we predicted individuals' performances on established tests from their MRI T1 weighted image in two independent data sets. From Mayo Clinic, 73 probable AD patients and 91 cognitively normal (CN) controls completed the Mini-Mental State Examination (MMSE), Dementia Rating Scale (DRS), and Auditory Verbal Learning Test (AVLT) within 3months of their scan. Baseline MRI's from the Alzheimer's disease Neuroimaging Initiative (ADNI) comprised the other data set; 113 AD, 351 MCI, and 122 CN subjects completed the MMSE and Alzheimer's Disease Assessment Scale-Cognitive subtest (ADAS-cog) and 39 AD, 92 MCI, and 32 CN ADNI subjects completed MMSE, ADAS-cog, and AVLT. Predicted and actual clinical scores were highly correlated for the MMSE, DRS, and ADAS-cog tests (P<0.0001). Training with one data set and testing with another demonstrated stability between data sets. DRS, MMSE, and ADAS-Cog correlated better than AVLT with whole brain grey matter changes associated with AD. This result underscores their utility for screening and tracking disease. RVR offers a novel way to measure interactions between structural changes and neuropsychological tests beyond that of univariate methods. In clinical practice, we envision using RVR to aid in diagnosis and predict clinical outcome.

Fig. 1

Flow diagram showing pre-processing steps. Abbr.: GM = grey matter; WM = white matter; CSF = cerebral spinal fluid; DARTEL = Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra.

Fig. 2

Illustration of Relevance Vector Regression with hypothetical 2D training data. The numbers in the squares are the training targets, i.e., actual scores, and the coordinates show the value of 2 different voxel intensities (feature 1 and feature 2). The goal is to find the features most predictive of the clinical score. All subjects are projected into a 1D line in such a way that minimizes the differences between the actual scores and the scores after projection, i.e., predicted score. For example, box 31 is projected into a value less than 31, which is the error that needs to be minimized in the learning algorithm.

Fig. 3

Whole brain grey matter plots of predicted versus actual scores for 4 different clinical ratings. Abbr: AD = Alzheimer's disease; MCI = Mild Cognitive Impairment; CN = Cognitively Normal; AVLT = Rey's Auditory Verbal Learning Test; DRS = Dementia Rating Scale; MMSE = Mini-Mental State Exam; ADAS-Cog = Alzheimer's Disease Assessment Scale—Cognitive Subtest.

Fig. 4

Weight maps for whole brain images, reflecting areas of the brain most vital in determining each RVR score. The red areas indicate where more grey matter adds to the accuracy of predicted test score, whereas blue areas indicate areas where more grey matter subtracts from the score. Note that dementia does not necessarily cause increased volumes of grey matter in these blue areas, but simply that data from these regions may help to adjust for anatomical variability (c.f. the contribution made by height when computing body mass index). In contrast to the MMSE and DRS, a lower score indicates better performance for the ADAS-Cog; therefore, these weight maps are mirror images. AVLT = Rey's Auditory Verbal Learning Test; DRS = Dementia Rating Scale; MMSE = Mini-Mental State Exam; ADAS-Cog = Alzheimer's Disease Assessment Scale—Cognitive Subtest.