On the use of histograms of oriented gradients for tremor detection from sinusoidal and spiral handwritten drawings of people with Parkinson's disease.

Parkinson's disease (PD), whose cardinal signs are tremor, rigidity, bradykinesia, and postural instability, gradually reduces the quality of life of the patient, making early diagnosis and follow-up of the disorder essential. This study aims to contribute to the objective evaluation of tremor in PD by introducing and assessing histograms of oriented gradients (HOG) to the analysis of handwriting sinusoidal and spiral patterns. These patterns were digitized and collected from handwritten drawings of people with PD (n = 20) and control healthy individuals (n = 20). The HOG descriptor was employed to represent relevant information from the data classified by three distinct machine-learning methods (random forest, k-nearest neighbor, support vector machine) and a deep learning method (convolutional neural network) to identify tremor in participants with PD automatically. The HOG descriptor allowed for the highest discriminating rates (accuracy 83.1%, sensitivity 85.4%, specificity 80.8%, area under the curve 91%) on the test set of sinusoidal patterns by using the one-dimensional convolutional neural network. In addition, ANOVA and Tukey analysis showed that the sinusoidal drawing is more appropriate than the spiral pattern, which is the most common drawing used for tremor detection. This research introduces a novel and alternative way of quantifying and evaluating tremor by means of handwritten drawings.