A new computer vision-based approach to aid the diagnosis of Parkinson's disease

TLDR: It is observed that the main problem in detecting PD is the patients in the early stages, who can draw near-perfect objects, which are very similar to the ones made by control patients, leading to higher accuracies than spiral images.

About: This article is published in Computer Methods and Programs in Biomedicine.The article was published on 2016-11-01 and is currently open access. It has received 98 citations till now. The article focuses on the topics: Feature extraction & Image processing.

Figures

Fig. 5 – Thinning of HT and ET using Zhang–Suen algorithm.

Fig. 6 – Sampling process: (a) a certain region with discontinuities, and (b) the proposed fair sampling process. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 7 – Some random points and the straight lines representing their connections with the spiral’s and meander’s center point. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 1 – Handwriting exams (a) filled out by a 56-year-old PD patient, and (b) an empty exam with the templates.

Fig. 2 – Image processing steps concerning ET extraction.

Fig. 8 – Spirals from the control group (a, b) and from the patient group (c, d).

Frequently asked questions

Q1. How did Hariharan and his colleagues develop a new feature weighting method?

Hariharan et al. [20] developed a new feature weighting method using Model-based clustering (Gaussian mixture model) in order to enrich the discriminative ability of the dysphonia-based features, thus achieving 100% of classification accuracy. 

Q2. What is the common method used for the diagnosis of Parkinson’s disease?

Since one of the first manifestation of Parkinson’s Disease is the deterioration of handwriting, the micrography (a writing exam) is another approach widely used for the diagnosis of Parkinson’s disease [11]. 

Q3. How did Zhao and his colleagues use the voice analysis to identify PD?

In the work conducted by Zhao et al. [8], five patients and seven healthy individuals were used to recognize Parkinson’s disease by means of voice analysis. 

Q4. What are the three techniques used in this experiment?

the authors evaluate three pattern recognition techniques: Naïve Bayes (NB), Optimum-Path Forest (OPF), and Support Vector Machines with Radial Basis Function (SVMRBF). 

Q5. What is the importance of the ray tracing phase?

This phase is crucial, since it has a considerable influence in the feature extraction step, which may affect the learning process as well. 

Q6. How did Pan et al. analyze the performance of Support Vector Machines with Radial?

Pan et al. [19] analyzed the performance of Support Vector Machines with Radial Basis Function in order to compare the onset of tremor in patients with Parkinson’s disease. 

Q7. How did the authors evaluate the recognition of PD?

The authors presented very good recognition rates, with 97.5% of the participants classified correctly (100% of the control individuals, and 95% of PD patients). 

Q8. What is the main contribution to the study?

The main contributions are related to the design of a new dataset that contains images from both spirals and meanders, which are cropped out from digitized handwritten exams, and the authors proposed a pipeline that can deal with the problem of learning from non-registered images. 

Q9. How did Pereira and his colleagues achieve this?

Very recently, Pereira et al. [22] proposed to extract features from writing exams using image processing techniques, achieving around 79% of recognition rates, which is considered very reasonable. 

Q10. How many people are affected by Parkinson’s disease?

It is composed of images extracted from handwriting exams of 92 individuals, divided in two groups: (i) the first one contains 18 exams of healthy people, named control group, with 6 male subjects and 12 female individuals; (ii) the second group contains 74 exams of people affected with Parkinson’s disease, named patient group, having 59 male and 15 female subjects. 

Q11. Did the authors use robust statistical evaluation in this experiment?

as the authors have only four accuracy values to compute the mean recognition rates and their standard deviation, the authors did not employ any robust statistical evaluation in this experiment. 

Q12. How did the authors demonstrate that in-air hand movements have a higher impact than the pure?

The authors demonstrated the assessment of in-air hand movements during sentence handwriting has a higher impact than the pure evaluation of on surface movements, leading to classification accuracies of 84% and 78%, respectively.