Objective:

To review the use of various CNN state-of-the-art architectures in colon cancer detection (bi-classification) and multi-classification tasks from January 2020 to the first quarter of 2025.

Approach:

• Search Strategy: Adopted the PRISMA method to retrieve metadata from IEEE Xplore, ScienceDirect, and Springer Nature databases using a specific boolean search expression.
• Data Collection: Utilized a stratified two-phase screening process for sampling and extracted accuracy metrics from included studies.
• Eligibility Standard: Applied the PICOS protocol to select studies based on population, intervention, comparator, outcome, and study type.
• Risk of Bias Assessment: Employed the PROBAST framework to evaluate methodological quality and potential biases in the selected studies.
• Reviewer Workflow: Implemented a three-reviewer workflow for screening and validating the selected studies.
• Data Synthesis: Conducted quantitative comparisons for data synthesis based on the selected studies.

Key Findings:

• Only 3,147 (0.7%) of CNN studies focused on colon cancer classification.
• 2,567 papers in the last 5 years studied state-of-the-art CNN variants for colon cancer diagnosis.
• A systematic review comparing CNN performance in colon cancer classification tasks is lacking.

Interpretation:

Limitations:

• Limited to studies published in English, which may exclude relevant research in other languages.
• Focused on peer-reviewed articles only, potentially missing important findings in non-peer-reviewed literature.

Conclusion:

The review provides an overview of CNN applications in colon cancer classification and identifies gaps in systematic comparisons.