**1. Introduction**

Parkinson's disease (PD) is a chronic neurodegenerative disease that affects over six million people worldwide. Because PD is most common in people over the age of 50, the number of PD patients is expected to double by 2040 due to the increase in life expectancy [1]. The loss of dopaminergic cells in the sustantia nigra region of the brain reduces the amount of dopamine in PD patients, causing dyscontrol in several areas of the brain. Some of the main symptoms of PD are motor symptoms, such as tremors, rigidity, and slow movement (bradykinesia). These symptoms, however, become apparent at an intermediate-advanced stage of the disease when the patient may have had the disease for over ten years [2].

The diagnosis of PD is considered one of the most challenging in the area of neurology. The autopsy of the brain of PD patients has shown that 35% of the

cases clinically diagnosed with PD were incorrect [3]. Usually, the diagnosis is done by a physician who looks for cardinal symptoms of the disease and starts dopaminergic therapy as a differential diagnosis. However, these symptoms appear at a late stage, and the patient may have lived with PD for years. Added to the similarity to other parkinsonian disorders that in some cases have the same motor symptoms as PD, may cost the patient crucial time and money, as inadequate treatments could be given by physicians. On the other hand, if detected in time, PD patients can improve their quality of life by taking the correct medication and therapy [2].

Great efforts are being made to find biomarkers that share some light into the causes and development of PD. Advances in technology provide alternatives to help physicians correctly diagnose PD patients at an early stage, and at the same time, obtain relevant information for understanding the disease. As shown in this article, non-invasive techniques such as medical images and voice recordings combined with machine learning and signal processing have proven to be adequate tools for solving the problem of PD detection with great accuracy.

The interest in using voice recordings for PD detection comes from the knowledge that voice disorders are prodromal symptoms present in over 90% of PD patients at an early stage. Some of the alterations include dysphonia (defective use of the voice), hypophonia (reduced vocal loudness), and imprecise hypokinetic articulation [4]. The advantages of this method are that voice recordings can be obtained without going to a hospital, and the economic cost is low, among others.

On the other hand, medical imaging techniques are important tools for understanding and helping diagnose PD. The images give information about neuroanatomical and pathophysiological processes related to the disease [5]. Some of the most used imaging techniques for neurological disease detection are DaTscan, *Magnetic Resonance Imaging* (MRI), and *Diffusion Tensor Imaging* (DTI). DaTscan images detect the concentration levels of dopamine in different regions of the brain, but the availability and cost of the studies may be prohibitive for patients. Structural Magnetic Resonance Imaging (sMRI) is a technique that provides structural information of the tissues and connectivity of the brain. It is available in most countries and is economically viable compared to other studies.

Work on voice-based and sMRI-based detection of PD and their clinical interpretation is reviewed in this article. The rest of the article is structured as follows: Section 2 presents voice-based analysis, including the database characteristics, classification results, and clinical interpretation of the extracted features. Section 3 introduces the analysis of sMRI of PD patients, the detection performance, and regions of interest for the diagnosis of both female and male patients. Section 4 gives conclusions and future work on the area of PD detection.
