Enhancing pancreatic cancer classification through dynamic weighted ensemble: a game theory approach

Area of Science:

• Medical Imaging Analysis
• Computational Oncology
• Artificial Intelligence in Healthcare

Background:

• Pancreatic cancer is a leading cause of cancer death, often diagnosed late due to its abdominal location.
• Accurate and early classification of pancreatic cancer is crucial for effective treatment.
• Computer-aided diagnosis (CAD) systems are increasingly used in radiological imaging for cancer detection.

Purpose of the Study:

• To develop and evaluate a dynamic weighted ensemble framework for pancreatic cancer classification.
• To leverage game theory for improved classification accuracy in medical imaging.
• To integrate transfer learning and traditional machine learning methods for robust diagnosis.

Main Methods:

• Feature extraction using Grey Level Co-occurrence Matrix (GLCM).
• Feature reduction via Gaussian kernel-based fuzzy rough sets theory (GKFRST).
• Classification using Random Forest (RF), ResNet50, and VGG16 with transfer learning (TL).
• Ensemble classification integrating TL and RF outcomes using a game theory approach.

Main Results:

• The proposed game theory-based ensemble classifier achieved significantly higher accuracy in pancreatic cancer classification.
• The framework demonstrated exceptional performance compared to current state-of-the-art models.
• Integration of multiple computational techniques led to improved diagnostic capabilities.

Conclusions:

• The novel ensemble framework effectively enhances pancreatic cancer classification accuracy.
• Game theory provides a unique and powerful mathematical paradigm for strategic interactions in cancer categorization.
• This research offers a distinctive methodological contribution to the field of computational oncology.