Deep transfer learning algorithms applied to synthetic drawing images as a tool for supporting Alzheimer's disease prediction

Neurodegenerative diseases, such as Alzheimer's Disease or Parkinson's disease, are unfortunately still incurable, although there are many therapies that can slow down the progression of the disease and improve patients' lives. An essential condition, however, is the early diagnosis of these disorders to begin therapies as soon as possible: In fact, when the signs of the disease become evident, damages may be already significant and irreversible. In this context, it is generally agreed that handwriting is one of the first skills affected by the onset of cognitive disorders. For this reason, in a preliminary study, we considered a database of handwriting and drawing specimens and proposed a method for selecting the most relevant information for diagnosing neurodegenerative disorders. The basic idea was to generate, for each handwriting sample, a color image to exploit the ability of convolutional neural network to automatically extract features from raw images. In the generated images, the color of each elementary trait encodes, in the three RGB channels, the dynamic information associated with that trait. Starting from the very encouraging obtained results, the aim of this study is twofold: On the one hand, we have tried to improve the feature extraction phase, associating further dynamic information with each handwritten trait. On the other hand, we have expanded the database of handwriting samples by adding specimen derived from more complex drawing tasks. Finally, we carried out a large set of experiments for comparing the results obtained by using standard online features with those obtained with our feature extraction approach.

Automated summary

Early diagnosis of neurodegenerative diseases is crucial for effective treatment, with handwriting being one of the first skills affected. Researchers proposed a method using color images and convolutional neural networks to extract features and improve the diagnosis. They also expanded the database by adding more complex drawing samples and conducted experiments comparing the results with standard feature methods.