Scholarlink

In recent years, the fast universal spread of Coronavirus Disease 2019 (COVID-19) has created a crucial need for efficient medical imaging tools to traditional methods like Real-time reverse Transcriptase-Polymerase Chain Reaction (RT-PCR). Moreover, previous methods frequently suffer from data limitations, suboptimal feature utilization, and imbalanced imaging information while expanding datasets. In this research, proposed Multi-head Attention (MHA) with Temporal Convolutional Networks (TCN) increases detection accuracy by capturing long-range dependencies in temporal data and analyzing serial X-ray patterns. The data is obtained from Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) dataset and preprocessed using an effective technique, namely image resizing and Z-score normalization to ensure consistent input size, and standardizes pixel values. Then, the Denoising Convolutional Neural Network (DnCNN) is used to extract essential features such as textures, edges, shapes from medical images. From the results, the proposed MHA-TCN approach provided outperforming results with an accuracy of 98.89%, F1 score of 98.88%, precision of 98.64% and recall of 97.67%, when compared with existing EfcientNetB0 method.

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

Anomaly Detection in Medical Imaging Based on Multi-head Attention with Temporal Convolutional Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ensteih Silvia,
  • B. N. Manjunatha,
  • Nelli Sreevidya,
  • B. G. Mamatha Bai,
  • P. V. Geetha

摘要

In recent years, the fast universal spread of Coronavirus Disease 2019 (COVID-19) has created a crucial need for efficient medical imaging tools to traditional methods like Real-time reverse Transcriptase-Polymerase Chain Reaction (RT-PCR). Moreover, previous methods frequently suffer from data limitations, suboptimal feature utilization, and imbalanced imaging information while expanding datasets. In this research, proposed Multi-head Attention (MHA) with Temporal Convolutional Networks (TCN) increases detection accuracy by capturing long-range dependencies in temporal data and analyzing serial X-ray patterns. The data is obtained from Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) dataset and preprocessed using an effective technique, namely image resizing and Z-score normalization to ensure consistent input size, and standardizes pixel values. Then, the Denoising Convolutional Neural Network (DnCNN) is used to extract essential features such as textures, edges, shapes from medical images. From the results, the proposed MHA-TCN approach provided outperforming results with an accuracy of 98.89%, F1 score of 98.88%, precision of 98.64% and recall of 97.67%, when compared with existing EfcientNetB0 method.