Optimized CNN framework with VGG19, EfficientNet, and Bayesian optimization for early colon cancer detection.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Colon cancer remains a leading cause of cancer mortality globally.
• Early and accurate detection methods are crucial for improving patient outcomes.
• Histopathological image analysis is vital for cancer diagnosis.

Purpose of the Study:

• To develop an advanced deep learning framework for automated colon cancer identification.
• To enhance classification accuracy and minimize overfitting in histopathological image analysis.
• To create a robust computer-aided diagnosis (CAD) tool for pathologists.

Main Methods:

• Integration of Convolutional Neural Networks (CNNs) with Bayesian optimization for hyperparameter tuning.
• Training and testing on a merged dataset from Kaggle and KCDP, encompassing nine tissue types.
• Implementation of data augmentation and stain normalization techniques for improved generalization.

Main Results:

• The optimized CNN achieved 96.84% accuracy, 97.02% precision, 96.50% recall, and 96.71% F1-score.
• An Area Under Curve (AUC) of 0.97 indicated high discriminative capability.
• The proposed method outperformed baseline CNN and ResNet architectures in robustness and generalization.

Conclusions:

• The deep learning framework shows significant promise as a CAD tool for colon cancer diagnosis.
• The model's effectiveness can potentially be extended to other cancer types via transfer learning.
• External validation on multi-institutional cohorts is necessary before clinical deployment.