Scholarlink

<p>The respiratory system is undoubtedly hampered by lung disorders. Also, one of the important reasons for death among people all around the world has been lung cancer. Early discovery can advance human survival probabilities. As a result, a unique ensemble-deep-learning paradigm for lung cancer detection and classification is established in the present research effort. The projected model includes five major phases: (a) image augmentation, (b) pre-processing, (c) segmentation, (d) feature extraction, (e) feature selection, and (f) lung cancer detection and classification, respectively. The collected raw CT images are augmentation with SMOTE. The augmented images are pre-processed via Median Filtering (for noise removal) and Contrast-limited adaptive histogram equalization (CLAHE) (for image contrast enhancement). Subsequently, from the pre-processed data, the ROI is identified via optimized U-NETS. The activation function (hyper-parameter) of U-NETS is optimized via a new hybrid optimization model-Digging Tunaswarm Optimizer (DTO). This DTO is the conceptual amalgamation of two standard meta-heuristic optimization models, namely “Honey Badger Algorithm (HBA)” and “Tuna Swarm Optimization (TSO)” models, respectively. Then, from the selected ROI area, the features like texture features (Manhattan Distance-based-GLCM, GLRM), Color features (Color Histogram), and Shape features (Moments, Area, Perimeter) are extracted. Among the extracted features, the optimal features are selected using DTO. This optimal feature selection reduces the computational complexity of the projected model. Finally, using these extracted optimal features, the two-fold-deep-learning-classifier framework is trained. This two-fold-deep-learning-classifiers framework encapsulates the <b>“</b>Bidirectional long-short term memory (Bidirectional LSTM)” and the “Recurrent Neural Network (RNN)” and the “Modified Convolutional Neural Network (M-CNN)”. In the first phase, the Bi-LSTM and RNN are clamped, and they are trained with the selected optimal factors. The outcome from Bi-LSTM and also RNN was fed as input to M-CNN. Final detected findings based on the existence or absence of lung cancer are acquired from the M-CNN, whose loss function has been modified with RMSE. Finally, a comparative evaluation is undergone to validate the efficiency of the projected model. The proposed model has a higher overall accuracy (92.4%) detecting modelling accuracy (96.3%) and classification accuracy (92.4%) compared to other models such as HBA, TSO, CNN, 3D CenterNet, and TSCNN. The use of a two-fold deep learning framework is responsible for these improvements, and the model also has lower failure rates (FPR and FNR) in detecting lung cancer. It is suggested that the proposed approach is effective in early-stage lung cancer detection.</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

Lung cancer detection and classification with optimal feature selection and two-fold-deep-learning-classifiers

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Babu Kumar S.,
  • M. Vinoth Kumar

摘要

The respiratory system is undoubtedly hampered by lung disorders. Also, one of the important reasons for death among people all around the world has been lung cancer. Early discovery can advance human survival probabilities. As a result, a unique ensemble-deep-learning paradigm for lung cancer detection and classification is established in the present research effort. The projected model includes five major phases: (a) image augmentation, (b) pre-processing, (c) segmentation, (d) feature extraction, (e) feature selection, and (f) lung cancer detection and classification, respectively. The collected raw CT images are augmentation with SMOTE. The augmented images are pre-processed via Median Filtering (for noise removal) and Contrast-limited adaptive histogram equalization (CLAHE) (for image contrast enhancement). Subsequently, from the pre-processed data, the ROI is identified via optimized U-NETS. The activation function (hyper-parameter) of U-NETS is optimized via a new hybrid optimization model-Digging Tunaswarm Optimizer (DTO). This DTO is the conceptual amalgamation of two standard meta-heuristic optimization models, namely “Honey Badger Algorithm (HBA)” and “Tuna Swarm Optimization (TSO)” models, respectively. Then, from the selected ROI area, the features like texture features (Manhattan Distance-based-GLCM, GLRM), Color features (Color Histogram), and Shape features (Moments, Area, Perimeter) are extracted. Among the extracted features, the optimal features are selected using DTO. This optimal feature selection reduces the computational complexity of the projected model. Finally, using these extracted optimal features, the two-fold-deep-learning-classifier framework is trained. This two-fold-deep-learning-classifiers framework encapsulates the “Bidirectional long-short term memory (Bidirectional LSTM)” and the “Recurrent Neural Network (RNN)” and the “Modified Convolutional Neural Network (M-CNN)”. In the first phase, the Bi-LSTM and RNN are clamped, and they are trained with the selected optimal factors. The outcome from Bi-LSTM and also RNN was fed as input to M-CNN. Final detected findings based on the existence or absence of lung cancer are acquired from the M-CNN, whose loss function has been modified with RMSE. Finally, a comparative evaluation is undergone to validate the efficiency of the projected model. The proposed model has a higher overall accuracy (92.4%) detecting modelling accuracy (96.3%) and classification accuracy (92.4%) compared to other models such as HBA, TSO, CNN, 3D CenterNet, and TSCNN. The use of a two-fold deep learning framework is responsible for these improvements, and the model also has lower failure rates (FPR and FNR) in detecting lung cancer. It is suggested that the proposed approach is effective in early-stage lung cancer detection.