Transfer learning based deep architecture for lung cancer classification using CT image with pattern and entropy...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading global cause of death, making early detection crucial for improved patient outcomes.
• CT scanning is a key diagnostic tool, but classifying lung cancer from CT images is challenging due to variable features.

Purpose of the Study:

• To present a novel hybrid deep learning architecture, ILN-TL-DM, for precise lung cancer classification from CT scan images.
• To enhance the accuracy and robustness of lung cancer detection through advanced image processing and classification techniques.

Main Methods:

• Pre-processing involved Adaptive Gaussian filtering for noise reduction and image enhancement.
• Segmentation utilized an Improved Attention-based ResU-Net (P-ResU-Net) to isolate lung and tumor regions.
• Feature extraction included Local Gabor Transitional Pattern (LGTrP), PHOG, deep features, and entropy-based features.
• Classification employed a hybrid deep learning model (ILN-TL-DM) combining improved LeNet, Transfer Learning, and DeepMaxout with soft voting.

Main Results:

• The ILN-TL-DM model achieved high accuracy (0.962), specificity (0.955), and Negative Predictive Value (NPV) (0.964).
• The proposed method demonstrated superior performance compared to conventional approaches in lung cancer detection.

Conclusions:

• The ILN-TL-DM model effectively classifies lung cancer from CT images, enhancing diagnostic accuracy.
• Combining sophisticated neural networks, feature engineering, and ensemble methods offers a promising strategy for medical image classification.