Scholarlink

Pneumonia is a severe health issue in the world, especially in the rural and resource poor areas whereby access to radiological services and qualified health care providers is minimal. Early detection is important to minimize the severity and mortality of the disease but the conventional methods of diagnosis like the interpretation of the chest X-ray is expensive and time consuming. In order to reduce these problems, this paper introduces a multimodal deep learning-based pre-screening system, which uses cough audio, coupled with chest X-ray images, to diagnose pneumonia in its early stages. Audio samples of cough were sampled in the Asthma Detection Dataset Version 2 and the chest X-ray samples were sampled in a Kaggle dataset that had radiologist-verified cases of Pneumonia and Normal cases. Mel-spectrogram characteristics were recognized with help of a lightweight SimpleCNN, whereas the X-ray images were evaluated with the help of DenseNet121. Both modalities prediction was fused to obtain a confidence-weighted fusion strategy. Experimental findings were 97, 94 and 98 percent accuracy on cough-based classification, X-ray-based classification and fusion model, respectively. The suggested system offers a cost-effective and easy-to-use pneumonia pre-screening system that can fit the telemedicine and low-resource healthcare setting.

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

Multimodal Cough Audio and Chest X-Ray Based Deep Learning Framework for Early Pneumonia Prediction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Pravin Patil,
  • Mangesh Balpande,
  • Mehul Waghare,
  • Ganesh Gwali,
  • Kaushal Patil,
  • Lokesh Khairnar

摘要

Pneumonia is a severe health issue in the world, especially in the rural and resource poor areas whereby access to radiological services and qualified health care providers is minimal. Early detection is important to minimize the severity and mortality of the disease but the conventional methods of diagnosis like the interpretation of the chest X-ray is expensive and time consuming. In order to reduce these problems, this paper introduces a multimodal deep learning-based pre-screening system, which uses cough audio, coupled with chest X-ray images, to diagnose pneumonia in its early stages. Audio samples of cough were sampled in the Asthma Detection Dataset Version 2 and the chest X-ray samples were sampled in a Kaggle dataset that had radiologist-verified cases of Pneumonia and Normal cases. Mel-spectrogram characteristics were recognized with help of a lightweight SimpleCNN, whereas the X-ray images were evaluated with the help of DenseNet121. Both modalities prediction was fused to obtain a confidence-weighted fusion strategy. Experimental findings were 97, 94 and 98 percent accuracy on cough-based classification, X-ray-based classification and fusion model, respectively. The suggested system offers a cost-effective and easy-to-use pneumonia pre-screening system that can fit the telemedicine and low-resource healthcare setting.