#CLOCK DRAWING CODE#

Furthermore, correlations were highly significant (P <. 001), and there was relatively little overlap between ratings for Alzheimer patients and normal controls. Inter-rater reliability for the clocks drawn by Alzheimer patients was highly significant (r = 0.86 P <. The mean performance score of Alzheimer subjects was 4.9 ± 2.7 compared to 8.7 ± 1.1 for normal controls (P <. Six independent observers blindly evaluated the results with ratings from 10 (best) to 1 (worst). To determine the usefulness of this measure, we asked 67 Alzheimer patients and 83 normal controls to draw the face of a clock reading the time of 2:45. Training is performed over a total of 50 epochs.We have tested a simple and reliable measure of visuospatial ability in Alzheimer patients - the Clock Drawing Test. The learning rate is reduced by a factor of ten after the 25th and 40th epochs. The learning rate is optimized using the Adam optimizer with an initial learning rate of 0.001, betas=(0.9, 0.999), and an L2 regularization parameter 0.0005. The neural network is trained using cross-entropy loss. One densely-connected output node (Linear activation).2D convolution (n_filters=2, filter_size=(1, 1), stride=1, padding=0).Addition layer summing the block input with the output of the previous BatchNorm layer.It consists of an input head, two residual blocks with skip connections, one 1x1 convolution, and one fully-connected output node. The two-dimensional neural network is constructed in PyTorch. The mean and standard deviation parameters obtained from the training dataset are then used to similarly scale the testing dataset.įinally, for each patient, the remaining command and copy features are stacked on top of each other, to create a feature map. Over the entire augmented training set, each numeric feature is standardized to have zero mean and unit variance. The training set is then augmented by oversampling the underrepresented class(es) with SMOTENC, a variant of the synthetic minority over-sampling technique that accounts for both numerical and categorical features in its generation of synthetic data. The data is first split into a training set and a testing set, with the holdout ratio being 0.3. These two features are each the fifth feature in their respective sets). Except for two, all other features are shared between the two feature sets (the remaining features are 'ClockFaceNonClockFaceLatency' in the command set and 'copy_PostClockFaceLatencyNoNoise_A' in the copy set. The features 'comm_ClockFaceNonClockFaceNoNoiseLatency' and 'CFNonCFNoNoiseTerminator' are removed from the dataset, as they appear in the 'command' set of features but not the 'copy' set of features.

Where the labels stand for Subjective Cognitive Impairment (SCI), Amnestic Mild Cognitive Impairment (AMCI), Mixed Mild Cognitive Impairment (MMCI), Alzheimer's Disease (AD), and Vascular Dementia (VAD).

#CLOCK DRAWING CODE#

Patients are classified as one of five levels of cognitive impairment: Code The 'copy' test asks the patient to draw a copy of an analog clock set to ten minutes after eleven.

The 'command' test asks the patient to draw, from memory, an analog clock set to ten minutes after eleven. The features are extracted from two tests: a 'command' test and a 'copy' test.

For each patient, there are 352 features, comprising 68 categorical features and 284 numerical features. The Alzheimer's disease clock drawing dataset consists of 196 patients who have taken the clock drawing test. This repository reports the results of constructing a two-dimensional convolutional neural network to classify the RowanSOM Alzheimer's disease clock drawing dataset.

Clock drawing