A Transfer Learning Approach with a Convolutional Neural Network for the Classification of Lung Carcinoma.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer diagnosis is critical but traditionally time-consuming and expensive.
• Early and accurate detection of lung nodules is vital for patient outcomes.
• Existing diagnostic methods can be invasive and resource-intensive.

Purpose of the Study:

• To develop a robust deep-learning-based model for lung cancer detection and recognition.
• To create an effective, non-invasive diagnostic tool for clinical use.
• To improve the efficiency and accuracy of lung cancer diagnosis.

Main Methods:

• A deep neural network was employed for feature extraction within a computer-aided diagnosis (CAD) system.
• The model involved data augmentation, classification using pre-trained Convolutional Neural Networks (CNNs), and localization.
• Transfer learning (TL) was utilized to address limited medical image data for training.

Main Results:

• The proposed model demonstrated high accuracy in lung cancer detection and recognition.
• Transfer learning techniques, including VGG 16, VGG 19, and Xception, were compared.
• At the 20th epoch, VGG 16 achieved 98.83% accuracy, VGG 19 achieved 98.05%, and Xception achieved 97.4%.

Conclusions:

• The deep learning model offers an effective, non-invasive approach to lung cancer diagnosis.
• The methodology provides a computationally efficient alternative to traditional diagnostic methods.
• The study highlights the robustness and potential of deep learning in medical image analysis for oncology.