Related posts:> Modeling the performance of an engine.
> Kaggle higgs challenge.
> Telemonitoring of patients with parkinsons disease.
Parkinson's disease is a degenerative disorder of the central nervous system. There is no cure for this disease, but medications, surgery and multidisciplinary management can provide relief from the symptoms.
One of the biggest difficulties for patients with Parkinson’s disease is the requisite physical visit to the clinic in order for monitoring and treatment.
The main goal of this study is to predict the unified Parkinson’s disease rating scale (UPDRS) for classifying the stage of the disease, by remote telemonitoring of the patients. This reduces the costs and the inconvenience of physical visits.
Research has shown that approximately 90% of people with Parkinson’s disease exhibit some form of vocal impairment such as dysphonia, when normal sounds production is affected, or dysarthria, when problems with the normal articulation of speech appear.
Therefore, vocal impairment may be one of the earliest indicators of the beginning of the illness.
The data used for this study consist of a range of biomedical voice measurements from 42 people with early-stage Parkinson’s disease, recruited for a six-month trial of a telemonitoring device for remote symptom progression monitoring. The recordings were automatically captured in the patient’s homes.
The number of recordings in the data set is 5.875 and there is a total of 22 variables (explained below) in each recording that are used to predict the total and motor UPDRS scale.
All this information is related to the main traditional measurement methods including F0 (the fundamental frequency of vocal oscillation), absolute sound pressure level (indicating the relative loudness of speech), jitter (the extent of variation in speech F0 from vocal cycle to cycle), shimmer (the extent of variation in speech amplitude from cycle to cycle) and noise-to-harmonics ratios (the amplitude of noise relative to tonal components in the speech). As some studies have proved that variations in these parameters may be useful in assessing the extent of dysphonia.
In this case, there are 3525 instances for training (60%), 1175 instances for generalization (20%), 1175 instances for testing (20%) and 0 unused instances (0%).
In order to discover the relationships among those voice measurement and the scores on the UPDRS scale, deep learning was used here.
In particular, the professional tool Neural Designer has been applied for building the medical diagnosis model.
The following image illustrates the deep learning architecture used to solve this problem. The yellow circles represent the scaling neurons for the inputs, the green circles represent the perceptron neurons and blue circles represents the unscaling neurons for every output in the neural network.
After performing the analysis and creating the predictive model, this software also provides us some tools to test it.
The most common method for testing the accuracy of a medical diagnosis system is to perform a linear regression analysis. Here the correlation for both of the output variables (motor UPDRS and total UPDRS) is good, since the corresponding coefficient is around 87,5%.
Moreover, the designed system yields a mean error of around 7% for both the motor and the total UPDRS, which can be considered very good for this type of application.
The neural network is now ready to estimate the motor_UPDRS and the toal_UPDRS with satisfactory quality over the same range of data.
This analysis shows that these voice measurements, being non-invasive and simple to administer, could be a good indicator in assessing the extent of Parkinson’s disease.