An optimized convolutional neural network architecture for lung cancer detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer diagnosis relies heavily on invasive biopsy, which is costly and traumatic.
• Current diagnostic methods face challenges with limited datasets and diagnostic inaccuracies.
• There is a need for automated, non-invasive tools for early lung cancer detection.

Purpose of the Study:

• To develop an automated diagnostic system for lung cancer screening using Computerized Tomography (CT) scans.
• To enhance the generalization of Convolutional Neural Network (CNN) models for diverse lung pathology CT slices.
• To optimize CNN hyperparameters for improved diagnostic accuracy and reliability.

Main Methods:

• A novel preprocessing methodology was developed for lung CT scans to prevent information loss during image smoothing.
• A Sine Cosine Algorithm (SCA) was integrated into the CNN model to optimize tuning parameters.
• The SCA algorithm minimized the error rate, serving as the objective function for hyperparameter selection.

Main Results:

• The proposed automated system achieved an average classification accuracy of 99% for lung scans.
• The model successfully classified scans into normal, benign, and malignant categories.
• The system demonstrated strong generalization ability on unseen datasets, validating its efficacy.

Conclusions:

• The developed automated diagnostic tool shows high accuracy and efficacy for lung cancer screening.
• The optimized CNN model with SCA integration offers a promising, less invasive alternative to traditional biopsy.
• The system's performance suggests its potential utility for radiologists in clinical settings.