[Quantitative analysis of pathological voice and identification with artificial neural network]

Area of Science:

• Otolaryngology
• Speech-Language Pathology
• Biomedical Engineering

Background:

• Pathological voice, often caused by vocal cord polyps, presents distinct acoustic characteristics.
• Objective voice analysis offers a potential method for identifying these pathological voice changes.

Purpose of the Study:

• To characterize the acoustic features of pathological voices in patients with vocal cord polyps.
• To evaluate the feasibility of using computer-based artificial neural networks for automatic pathological voice identification.

Main Methods:

• Acoustic parameters (Jitter, Shimmer, HNR, SNR, NNE) were analyzed using Praat software in 129 patients with vocal cord polyps and 125 controls.
• An artificial neural network model was trained and tested using these acoustic parameters, with a separate validation set of 140 cases.
• SPSS Modeler was employed for neural network reconstruction and identification rate calculation.

Main Results:

• Pathological voices exhibited significantly higher Jitter, Shimmer, and Normalized Noise Energy (NNE), and lower Harmonic to Noise Ratio (HNR) and Signal to Noise Ratio (SNR) compared to normal voices (P<0.05).
• The artificial neural network model achieved a 75.7% recognition rate for pathological voices.
• Gender-specific calculations confirmed these acoustic differences.

Conclusions:

• Objective voice analysis, utilizing acoustic parameters, is effective in identifying pathological voices.
• Artificial neural networks demonstrate high accuracy and clinical utility for the automated recognition of pathological voice conditions.