Tensor-RT-Based Transfer Learning Model for Lung Cancer Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading global cause of death, necessitating accurate and early diagnosis.
• Computed tomography (CT) scans are crucial for identifying lung cancer stages, but automated classification remains challenging.
• Current computer-aided diagnosis systems often require manual annotation and struggle with real-time performance.

Purpose of the Study:

• To develop a high-speed, real-time transfer learning framework for classifying CT lung cancer slices as benign or malignant.
• To improve the accuracy and efficiency of automated lung cancer diagnosis using deep learning.
• To create a system deployable in real-time clinical applications without manual nodule annotation.

Main Methods:

• A novel framework utilizing K-means clustering and morphological operations for lung image pre-processing and segmentation.
• Development and regularization of a weighted VGG deep network (WVDN) model for classification.
• Real-time model inference using Nvidia Tensor-RT for efficient deployment.

Main Results:

• The proposed WVDN model achieved a classification accuracy of 0.932, with precision, recall, and F1 score of 0.93.
• Cohen's kappa score reached 0.85, indicating strong agreement.
• Statistical evaluation showed a p-value <0.0001, confirming the model's significant performance improvement over existing methods.

Conclusions:

• The proposed transfer learning framework demonstrates superior performance for real-time lung cancer classification from CT slices.
• The system's ability to process complete CT slices without manual annotation offers a significant advantage for clinical diagnosis.
• This high-speed, accurate model has the potential to enhance early lung cancer detection and patient outcomes.