Scholarlink

<p>One of the maximum deadly types of cancer is the lung cancer that has a high death rate because it is often detected too late. The knowledge of radiologists is crucial to current diagnostic procedures, which results in an extensive range of interpretations. In this work, a sophisticated, Deep Learning (DL) with optimization approach and Explainable AI (XAI) based diagnostic system that automates CT image interpretation for lung cancer prediction is proposed. To maintain contrast and edge details, the system deploys preprocessing based fast local Laplacian Filter (FLLF). To maintain spatial coherence, Fuzzy C-Means (FCM) clustering combined with Markov Random Fields (MRF) is used for segmentation, which segment preprocessed image, to improve the prediction level of lung cancer. Contour shape descriptors is utilized to extract morphological information from segmented regions to provide a robust characterization of lung cancer shapes. After that, these characteristics are classified by Pyramid Vision Transformer (PVT)–ResNet model, which enhances classification accuracy by fusing the local feature learning power of Attention Residual and Convolution Neural Network (CNN) with the global attention mechanism of PVT. For improving the performance of classifier, its parameters gets fine-tuned with the assistance of Horned Lizard Optimization Algorithm (HLOA). Local Interpretable Model-agnostic Explanations (LIME) provides visual insight into model’s decision-making process, is integrated for XAI to guarantee transparency in forecasts. A very precise, comprehensible framework for diagnosis of lung cancer is guaranteed by a results generated from python software, which accomplishes higher accuracy of (99.1%).</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

PVT-ResNet Fusion: A Hybrid DL Framework for Lung Cancer Prediction Using Image Analysis and Explainable AI

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • R. Robert,
  • V. Muneeswaran,
  • Jose Saji Kumar,
  • P. Nagaraj

摘要

One of the maximum deadly types of cancer is the lung cancer that has a high death rate because it is often detected too late. The knowledge of radiologists is crucial to current diagnostic procedures, which results in an extensive range of interpretations. In this work, a sophisticated, Deep Learning (DL) with optimization approach and Explainable AI (XAI) based diagnostic system that automates CT image interpretation for lung cancer prediction is proposed. To maintain contrast and edge details, the system deploys preprocessing based fast local Laplacian Filter (FLLF). To maintain spatial coherence, Fuzzy C-Means (FCM) clustering combined with Markov Random Fields (MRF) is used for segmentation, which segment preprocessed image, to improve the prediction level of lung cancer. Contour shape descriptors is utilized to extract morphological information from segmented regions to provide a robust characterization of lung cancer shapes. After that, these characteristics are classified by Pyramid Vision Transformer (PVT)–ResNet model, which enhances classification accuracy by fusing the local feature learning power of Attention Residual and Convolution Neural Network (CNN) with the global attention mechanism of PVT. For improving the performance of classifier, its parameters gets fine-tuned with the assistance of Horned Lizard Optimization Algorithm (HLOA). Local Interpretable Model-agnostic Explanations (LIME) provides visual insight into model’s decision-making process, is integrated for XAI to guarantee transparency in forecasts. A very precise, comprehensible framework for diagnosis of lung cancer is guaranteed by a results generated from python software, which accomplishes higher accuracy of (99.1%).