Assessment of Fourier methods and maximun entropy for automatic detection of Parkinson 's disease

Abstract

Parkinson's disease (PD) is the second most prevalent neurodegenerative clinical condition after Alzheimer's disease and for the global health system it is essential to identify early signs; however, nowadays it is still a new fi eld of study that needs further development. It has been shown that about 90% of Parkinson’s disease patients also develop defi ciencies in her voice, showing symptoms such as monotone speech, low-intensity tone, isolated
breaks, inaccurate pronunciation of consonants and prosody problems. Although such problems are identifi ed, only 3% to 4% are treated as a voice problem.
In the fi eld of research, time-frequency analysis has proven to be a powerful tool in the processing of acoustic signals, more specifi cally, the voice processing. With the aim of having clean representations of the spectrum to help with the removal of characteristics and mitigate other resulting problems when using estimated classical methods, it was decided to study the behavior of maximum entropy method, which is currently used in oceanography and astronomy in the study of voice signals. For these experiments, a database of patients with Parkinson's disease was used. Fifty records of pathological voices and the same amount of healthy voices were studied. Voiced and voiceless segments derived from the word `` PA-TA-KA '' produced by those patients were analyzed. Energies were estimated from both the Fourier transform and the maximum entropy method. In order to evaluate the performance of the classical methods and the method of maximum entropy, a K-nn classifi er was used and success rates were found in around 60% taking into account the maximum entropy method in both syllables and phonemes.