Review of Parkinson’s disease detection with deep‑learning machines

Parkinson's disease is the second most prevalent disease in the world. Many technologies have been applied as an aid to diagnosis, treatment monitoring, and to reduce the symptoms of this disease. In this paper we review technologies to assess symptoms and diagnose the disease in the early stages. Choosing the most appropriate method for a specific scenario remains a difficult decision, due to the existence of many approaches described in the literature for this purpose. Therefore, a systematic review was performed to identify the most used techniques in the diagnosis of Parkinson's disease. A total of 7079 studies were identified, from which the 47 most relevant were selected for this review. A PRISMA flow chart was used to rationalize the selection of the 47 papers. Focusing on papers published from 2018 to the present, a subset was selected for inclusion in this review. Papers with abstracts indicating that their scope was not aligned with the topic of the review were excluded. The papers selected based on abstract review were then analyzed in depth to identify those most relevant to the review. Five categories or technologies have been found to be used for Parkinson’s disease diagnosis: imaging, analysis of movement, handwriting, speech, and electroencephalogram signals. The best results are obtained with imaging technology, providing accuracies as high as 98%. The analysis of movement provides a best accuracy of 95%, while handwriting and speech analysis provide accuracies of 97.62% and 97.6%, respectively. Finally, as for EEG signals analysis, the best accuracy reached in the studies is 99.9%. In addition, the study analyzes the use of deep learning, types of sensors and images, increasing the information frequently provided in this type of analysis or assessment. The cost associated with obtaining the necessary data is also discussed, concluding that MRI is expensive, but speech, hand-drawn pictures or EEG signals can be obtained with low-cost equipment. The review has been performed using the web resources provided by the library of the University of Alcalá (Spain), which provides access to most of the specialized publications in this research area. The review is intended as a guide for specialists to support them in choosing the most appropriate method for a particular scenario.