Scholarlink

<p>The second-most frequent neurological condition is Parkinson’s Disease (PD). There is no cure; however, therapy and drugs can lessen symptoms. Therefore, early PD recognition is crucial for control. The computer-Aided Diagnosis (CAD) systems can identify PD via voice cues. Low-quality signals and improper feature selection make it difficult to design an accurate PD detection system utilizing patient voice signals. This research provides an effective PD detection method using patient voice signals. The model uses improved automatic voice filtering and feature engineering techniques. First, a U-Net-based Twin Lightweight Deep Learning Model (TLDLM) is proposed to enhance the quality of speech signals. The proposed TLDLM model is applied to any input speech signal to enhance its quality for accurate classification problems. After automatically filtering speech signals, we design the consolidated model for optimized feature engineering and classification. In this work, we extend the TLDLM model for automatic feature learning using the optimization algorithm. The DL feature extraction process, due to its high speed, leads to significant redundancy and dimensionality, which in turn causes poor classification results. The extracted features are optimized using the Whale Optimization Algorithm (WOA). The main purpose of the WOA algorithm is to improve PD classification performance by making appropriate feature selections. Support Vector Machine (SVM), Naive Bayes (NB), and Random Forest (RF) are machine learning classifiers used in the development of the proposed model for classification. According to the results of the simulation, the recommended strategy outperformed current approaches in terms of improving speech quality and classifying PD.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advances in parkinson’s disease detection using intelligent deep learning-based speech signal filtering and feature engineering

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Chaitali Shamrao Raje,
  • Pramodkumar H. Kulkarni,
  • Rupali Ravindra Deshmukh

摘要

The second-most frequent neurological condition is Parkinson’s Disease (PD). There is no cure; however, therapy and drugs can lessen symptoms. Therefore, early PD recognition is crucial for control. The computer-Aided Diagnosis (CAD) systems can identify PD via voice cues. Low-quality signals and improper feature selection make it difficult to design an accurate PD detection system utilizing patient voice signals. This research provides an effective PD detection method using patient voice signals. The model uses improved automatic voice filtering and feature engineering techniques. First, a U-Net-based Twin Lightweight Deep Learning Model (TLDLM) is proposed to enhance the quality of speech signals. The proposed TLDLM model is applied to any input speech signal to enhance its quality for accurate classification problems. After automatically filtering speech signals, we design the consolidated model for optimized feature engineering and classification. In this work, we extend the TLDLM model for automatic feature learning using the optimization algorithm. The DL feature extraction process, due to its high speed, leads to significant redundancy and dimensionality, which in turn causes poor classification results. The extracted features are optimized using the Whale Optimization Algorithm (WOA). The main purpose of the WOA algorithm is to improve PD classification performance by making appropriate feature selections. Support Vector Machine (SVM), Naive Bayes (NB), and Random Forest (RF) are machine learning classifiers used in the development of the proposed model for classification. According to the results of the simulation, the recommended strategy outperformed current approaches in terms of improving speech quality and classifying PD.