Efficient deep learning model for classifying lung cancer images using normalized stain agnostic feature method and...

Area of Science:

• Computational pathology
• Medical imaging analysis
• Artificial intelligence in oncology

Background:

• Lung cancer is a leading cause of global mortality.
• Accurate histological classification is crucial for effective treatment and patient outcomes.
• Current diagnosis relies on manual analysis of tissue samples.

Purpose of the Study:

• To develop a computer-assisted diagnosis (CADx) system for non-small cell lung cancer (NSCLC) histology classification.
• To leverage deep learning and AI for improved diagnostic accuracy and efficiency.
• To evaluate the performance of a modified ResNet-34 architecture.

Main Methods:

• Utilized the FastAI-2 framework with a modified ResNet-34 deep learning architecture.
• Implemented stain normalization in LAB color space for consistent color representation.
• Trained the model on the LC25000 dataset and compared performance against VGG11 and SqueezeNet1_1.

Main Results:

• The proposed modified ResNet-34 model achieved 99.78% accuracy in classifying lung cancer histology.
• Demonstrated superior performance and an optimal balance between network depth and computational efficiency.
• FastAI-2 framework facilitated rapid model convergence and reduced training time.

Conclusions:

• Automated histopathology classification using AI shows high effectiveness for lung cancer diagnosis.
• AI-driven tools can significantly assist pathologists by enhancing accuracy and reducing workload.
• The developed system holds potential for improving clinical decision-making in oncology.