Leveraging explainable artificial intelligence for transparent and trustworthy cancer detection systems

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Timely cancer detection is crucial for improving patient outcomes.
• Artificial Intelligence (AI), particularly Deep Learning (DL), shows promise in cancer diagnostics but faces transparency challenges.
• Explainable AI (XAI) offers solutions by enhancing the interpretability and transparency of AI models.

Purpose of the Study:

• To systematically review recent applications of XAI in cancer detection across various cancer types.
• To categorize XAI techniques based on cancer type, interpretability methods, dataset usage, simulation environments, and security considerations.
• To identify current challenges and research gaps in trustworthy and interpretable AI for oncology.

Main Methods:

• Systematic literature review of recent XAI applications in cancer detection.
• Categorization of studies by cancer type (breast, skin, lung, colorectal, brain, etc.).
• Analysis of interpretability methods, dataset utilization, simulation environments, and security aspects.

Main Results:

• Convolutional Neural Networks (CNNs) are used in 31% of models.
• SHAP is the most common interpretability framework (44.4%).
• Python is the leading programming language (32.1%), with security issues addressed in only 7.4% of studies.

Conclusions:

• XAI is increasingly important for transparent and interpretable AI in cancer diagnostics.
• Further research is needed to address security concerns and advance trustworthy AI applications in oncology.
• This review provides a roadmap for future research in explainable AI for cancer detection.