Scholarlink

Lung Cancer is considered one of the deadliest forms of cancer world-wide. There have been limited successes in developing diagnostic systems to identify Lung Cancer. The objective of this research project is to create and evaluate an innovative hybrid model for identifying Lung Cancer using chest imaging (CT scans and X-ray). This research project created a new AI Hybrid Model utilizing GANs (Generative Adversarial Networks) to generate synthetic images for the purpose of data augmentation. Utilizing a variety of architectures of neural networks including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), combinations of CNN and RNN, GAN-augmented CNN’s, and lightweight architectures such as MobileNet V2 and SegNet, the researchers evaluated all models using a large number of experiments using a large set of diverse preprocessed images, including CT scan images and X-ray images and synthesized images representing the complexity and realism of lung disease. Each architecture was evaluated using multiple metrics including accuracy, precision, recall, F1-score, and Area Under Curve Receiver Operator Characteristic (AUC ROC). The best performing architecture was the combination of CNN and RNN achieving 96.3% accuracy, 0.94 precision, 0.96 recall, 0.95 F1-score, and 0.97 AUC ROC. All architectures trained with synthetic image augmentations outperformed those trained only on original images, particularly in terms of recall and F1-score. MobileNetV2 and SegNet performed slightly lower than other architectures but are still viable options due to their lightweight architecture which makes them ideal for rapid diagnostic, real-time or mobile diagnostics. For visualization of the architectures’ performance the researchers used both bar charts and radar plots. The researchers conclude that the use of hybrid workflows combined with data enhancement techniques will increase diagnostic accuracy. This research demonstrated the effectiveness of integrating data enhancement techniques into hybrid workflows and also provide an opportunity for future studies to investigate multimodal systems and large-scale clinical validations.

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

Hybrid Deep Learning and Generative Augmentation Framework for Robust Multi-Metric Lung Cancer Detection: A Comparative Evaluation of CNN, RNN, and Lightweight Architectures

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anupam Agrawal,
  • Asadi Srinivasulu,
  • Bodicherla Siva Sankar,
  • K. Vijaya Bhaskar,
  • A. Rama Mohan Reddy,
  • G. Pradeep,
  • Kalavagunta Varshita,
  • Goddindla Sreenivasulu

摘要

Lung Cancer is considered one of the deadliest forms of cancer world-wide. There have been limited successes in developing diagnostic systems to identify Lung Cancer. The objective of this research project is to create and evaluate an innovative hybrid model for identifying Lung Cancer using chest imaging (CT scans and X-ray). This research project created a new AI Hybrid Model utilizing GANs (Generative Adversarial Networks) to generate synthetic images for the purpose of data augmentation. Utilizing a variety of architectures of neural networks including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), combinations of CNN and RNN, GAN-augmented CNN’s, and lightweight architectures such as MobileNet V2 and SegNet, the researchers evaluated all models using a large number of experiments using a large set of diverse preprocessed images, including CT scan images and X-ray images and synthesized images representing the complexity and realism of lung disease. Each architecture was evaluated using multiple metrics including accuracy, precision, recall, F1-score, and Area Under Curve Receiver Operator Characteristic (AUC ROC). The best performing architecture was the combination of CNN and RNN achieving 96.3% accuracy, 0.94 precision, 0.96 recall, 0.95 F1-score, and 0.97 AUC ROC. All architectures trained with synthetic image augmentations outperformed those trained only on original images, particularly in terms of recall and F1-score. MobileNetV2 and SegNet performed slightly lower than other architectures but are still viable options due to their lightweight architecture which makes them ideal for rapid diagnostic, real-time or mobile diagnostics. For visualization of the architectures’ performance the researchers used both bar charts and radar plots. The researchers conclude that the use of hybrid workflows combined with data enhancement techniques will increase diagnostic accuracy. This research demonstrated the effectiveness of integrating data enhancement techniques into hybrid workflows and also provide an opportunity for future studies to investigate multimodal systems and large-scale clinical validations.