Scholarlink

<p>Voice disorders among people are constantly increasing nowadays all across the globe. In recent years, numerous researchers have developed diverse tools for voice pathology detection to determine the type of voice disorder accurately. However, these existing multifarious tools have various limitations, including but not limited to high computational complexity, low effectiveness, and many more. In this article, an improvised CNN-based multi-level model (IML-CNN) is proposed for laryngeal cancer detection in the initial stage based on the voice dataset. This IML-CNN model has been built based on a two-stage pipeline. Initially, a CNN-based detector is used to determine normal and pathological voice samples. Furthermore, a fine-tuned balanced classifier is integrated with the IML-CNN model for classification of the laryngeal cancer and its associated symptoms in the growing phase. Our proposed IML-CNN multi-level model adopts a structured training technique that involves scheduling of adaptive learning rate as well as dropout regularization to improve the generalization capability of the model. These strategies allow the IML-CNN framework to proficiently learn from different voice data samples and enhance its utility to a widespread medical issue associated with pathological voice disease. The performance evaluation of the proposed IML-CNN model offers an accuracy of 97.36%. The outcome of the proposed models demonstrates that they offer higher accuracy scores in the identification of voice pathological disorders and laryngeal cancer in real-time diagnosis procedures.</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

An Optimized Multi-Level Convolutional Neural Network Model for Real Time Detection of Laryngeal Cancer

  • 期刊论文

  • 发表日期：

    2025年8月21日

查看全文
  • Pravat Kumar Sahoo,
  • Sushruta Mishra,
  • Biswajit Brahma,
  • Akash Kumar Bhoi

摘要

Voice disorders among people are constantly increasing nowadays all across the globe. In recent years, numerous researchers have developed diverse tools for voice pathology detection to determine the type of voice disorder accurately. However, these existing multifarious tools have various limitations, including but not limited to high computational complexity, low effectiveness, and many more. In this article, an improvised CNN-based multi-level model (IML-CNN) is proposed for laryngeal cancer detection in the initial stage based on the voice dataset. This IML-CNN model has been built based on a two-stage pipeline. Initially, a CNN-based detector is used to determine normal and pathological voice samples. Furthermore, a fine-tuned balanced classifier is integrated with the IML-CNN model for classification of the laryngeal cancer and its associated symptoms in the growing phase. Our proposed IML-CNN multi-level model adopts a structured training technique that involves scheduling of adaptive learning rate as well as dropout regularization to improve the generalization capability of the model. These strategies allow the IML-CNN framework to proficiently learn from different voice data samples and enhance its utility to a widespread medical issue associated with pathological voice disease. The performance evaluation of the proposed IML-CNN model offers an accuracy of 97.36%. The outcome of the proposed models demonstrates that they offer higher accuracy scores in the identification of voice pathological disorders and laryngeal cancer in real-time diagnosis procedures.