Lung cancer classification using neural networks for CT images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early cancer detection significantly improves patient survival rates.
• Computed tomography (CT) imaging is crucial for visualizing lung structures.
• Artificial neural networks (ANNs) offer potential for automated image analysis in cancer diagnosis.

Purpose of the Study:

• To develop and evaluate a computer-aided classification method for lung cancer detection in CT images.
• To compare the performance of different artificial neural network architectures and training functions.
• To introduce and validate novel training functions for improved classification accuracy.

Main Methods:

• Lung segmentation from CT images.
• Calculation of statistical parameters (mean, standard deviation, skewness, kurtosis, moments) from segmented lung images.
• Classification using feed-forward and back-propagation neural networks, including standard and proposed training functions.

Main Results:

• Back-propagation neural networks outperformed feed-forward networks.
• The statistical parameter 'skewness' yielded the highest classification accuracy.
• The proposed training function 1 achieved 93.3% accuracy, 100% specificity, and 91.4% sensitivity.
• The proposed training function 2 achieved 93.3% accuracy with a minimal mean square error of 0.0942.

Conclusions:

• The developed computer-aided classification method using ANNs shows high potential for accurate lung cancer detection.
• The proposed novel training functions significantly enhance classification performance compared to existing methods.
• This approach can aid in earlier diagnosis, potentially leading to better patient outcomes.