Scholarlink

<p>Early diagnosis of lung cancer is important for enhancing treatment outcomes and increasing survival rates, which derives from its aggressive progression and high mortality. Computational search techniques, particularly those leveraging deep learning methods like Convolutional Neural Networks (CNNs), are revolutionizing medical image analysis because they can automatically extract and evaluate critical features from medical images, making them essential for accurate diagnosis and informed clinical decision-making. In this study, we propose a multi-level fine-tuning method for classifying lung cancer images using three different pre-trained models: VGG16, MobileNetV2, and ResNet50, based on two commonly used datasets: the IQ-OTH/NCCD and Kaggle Chest CT lung cancer datasets. These models will be unfrozen at different levels, which involves the last quarter, third, and half of the model’s layers to investigate the impact of the multi-level fine-tuning strategy on the model’s efficiency. The findings demonstrate that the proposed multi-level fine-tuned VGG16 architecture performed best, achieving the highest classification accuracies of 99.54% and 98.01% on the IQ-OTH/NCCD and Kaggle Chest CT datasets, respectively, indicating a high degree of accuracy in classifying lung cancer images into multiple distinct categories.</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 innovative multi-level fine-tuning deep learning approach for enhanced lung cancer classification

  • 期刊论文

  • 发表日期：

    2025年8月19日

查看全文
  • Zahia Benamara,
  • Soraya Zehani,
  • Athmane Zitouni,
  • Tawfiq Beghriche

摘要

Early diagnosis of lung cancer is important for enhancing treatment outcomes and increasing survival rates, which derives from its aggressive progression and high mortality. Computational search techniques, particularly those leveraging deep learning methods like Convolutional Neural Networks (CNNs), are revolutionizing medical image analysis because they can automatically extract and evaluate critical features from medical images, making them essential for accurate diagnosis and informed clinical decision-making. In this study, we propose a multi-level fine-tuning method for classifying lung cancer images using three different pre-trained models: VGG16, MobileNetV2, and ResNet50, based on two commonly used datasets: the IQ-OTH/NCCD and Kaggle Chest CT lung cancer datasets. These models will be unfrozen at different levels, which involves the last quarter, third, and half of the model’s layers to investigate the impact of the multi-level fine-tuning strategy on the model’s efficiency. The findings demonstrate that the proposed multi-level fine-tuned VGG16 architecture performed best, achieving the highest classification accuracies of 99.54% and 98.01% on the IQ-OTH/NCCD and Kaggle Chest CT datasets, respectively, indicating a high degree of accuracy in classifying lung cancer images into multiple distinct categories.