Deep Neural Network Models for Colon Cancer Screening.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early colorectal cancer detection improves clinical decision-making and reduces workload.
• Automatic systems using endoscopic and histological images are key for early detection.
• Deep learning advancements have spurred development in image- and video-based polyp analysis.

Purpose of the Study:

• To review recent advances in AI models for colorectal cancer prediction.
• To summarize models with and without transparency for polyp identification and segmentation.
• To address the knowledge gap in explainable AI for clinical diagnostics.

Main Methods:

• Review of deep learning techniques, including convolutional neural networks (CNNs).
• Analysis of image patch processing and preprocessing methods.
• Examination of transfer learning and end-to-end learning for detection and localization.

Main Results:

• AI methods, particularly CNNs, show strong performance in predicting invasive cancer.
• Transfer and end-to-end learning enhance accuracy and reduce data dependency.
• Growing demand for explainable AI (XAI) to ensure transparency, interpretability, and fairness.

Conclusions:

• AI significantly aids in colorectal cancer diagnosis and workload reduction.
• Explainable AI models are crucial for reliable and trustworthy clinical applications.
• Future research should focus on developing and validating transparent AI systems.