Automated Diagnosis for Colon Cancer Diseases Using Stacking Transformer Models and Explainable Artificial...

Area of Science:

• Oncology
• Computer Science
• Artificial Intelligence

Background:

• Colon cancer is a leading global cancer, with nearly two million cases in 2020.
• Accurate early detection is crucial for successful colon cancer treatment.
• Deep learning offers potential for enhanced diagnostic accuracy in medical imaging.

Purpose of the Study:

• To propose a novel heterogenic stacking deep learning model for colon cancer prediction.
• To improve the performance of colon cancer detection using integrated deep learning approaches.
• To evaluate the proposed model against established deep learning architectures.

Main Methods:

• Developed a heterogenic stacking deep learning model integrating pretrained convolutional neural network (CNN) models.
• Utilized a metalearner to enhance prediction performance within the stacking framework.
• Evaluated the model on the LC25000 and WCE colon cancer image datasets (binary and multiclassified).
• Compared performance against VGG16, InceptionV3, Resnet50, and DenseNet121 using accuracy, recall, precision, and F1 score.

Main Results:

• The proposed stacking deep learning model achieved superior performance on both datasets.
• For LC25000, the stacked model attained 100% accuracy, recall, precision, and F1 score.
• For WCE, the stacked model achieved 98% accuracy, recall, precision, and F1 score.
• Stacking-SVM demonstrated higher performance than individual models like VGG16, InceptionV3, Resnet50, and DenseNet121.

Conclusions:

• Heterogenic stacking deep learning models significantly enhance colon cancer prediction accuracy.
• The proposed stacking approach offers a robust and highly effective method for colon cancer detection.
• Explainable AI (XAI) methods can be applied to understand black-box deep learning models in this context.