A Novel Diagnostic Rule for Parkinson's Disease Based on a Hybrid Extraction Method

Parkinson's disease has become an increasing public health issue that its symptoms become more severe over time. Early diagnosis and treatment of this disease leads to improving the skills, abilities and performance of patients in daily life. In order to diagnose the disease early, it is necessary to produce clinical decision-making assistance systems that are able to detect the diagnostic rules of the disease.

This study provides an automatic way to extract novel diagnostic rules for Parkinson's disease. The proposed method is based on logic regression and simulated annealing algorithm. To evaluate the method, the Oxford Parkinson's data set was used, which contains 22 biomedical voice measurements from 31 people, 23 with Parkinson's disease. The dataset has 195 voice recording from these individuals.

The results include two diagnostic rules; If high accuracy was the main concern, a new rule has been proposed that includes 21 logical statements that have an accuracy of 92.31%, a sensitivity of 85.42%, and a specificity of 94.56%. However, for real-time systems and clinical decision-making assistance with high interpretability, a rule consisting of 3 logical statements has been proposed, which has an accuracy of 78.97%, a sensitivity of 77.08% and a feature of 79.59%.

The results show the high power of interpretability and reliability of the proposed rules in the diagnosis of Parkinson's disease, which can be used in the implementation of remote diagnostic systems.