A deep learning-based framework for lung cancer survival analysis with biomarker interpretation.

Area of Science:

• Computational pathology
• Biomedical data science
• Artificial intelligence in oncology

Background:

• Lung cancer is a leading cause of cancer death with low survival rates.
• Accurate survival analysis is critical for lung cancer diagnosis and treatment.
• Current methods need improvement for better patient outcomes.

Purpose of the Study:

• To develop an advanced survival analysis system for lung cancer.
• To leverage deep learning for improved survival prediction accuracy.
• To identify and visualize image-based biomarkers for lung cancer prognosis.

Main Methods:

• A deep neural network for end-to-end cellular feature learning.
• Locality-constrained linear coding (LLC) bag-of-words (BoW) for patient-level features.
• Cox proportional hazards model with elastic net for feature selection and survival analysis.
• Biomarker interpretation module for visualizing predictive image regions.

Main Results:

• The system achieved excellent predictive power on a public lung cancer dataset.
• Demonstrated high accuracy using log-rank test (p-value) and concordance index (c-index).
• Successfully identified and visualized image regions contributing to survival predictions.

Conclusions:

• A novel segmentation-free survival analysis system was developed using deep learning and Cox models.
• The system effectively predicts lung cancer patient survival.
• Visualizable biomarkers provide evidence for the model's prognostic capabilities.