Explainable artificial intelligence in breast cancer detection and risk prediction: A systematic scoping review

Overview

• The paper explores the application of explainable artificial intelligence (XAI) in breast cancer detection and risk prediction.
• Researchers conducted a systematic review of studies that used XAI methods in breast cancer datasets.
• The review found that the SHAP model is the most widely used XAI technique in breast cancer research.
• SHAP is a model-agnostic approach that can explain the predictions of various machine learning models, including tree-based ensemble models.
• Explainable AI is important for improving the transparency, interpretability, fairness, and trustworthiness of AI-enabled health systems and medical devices.

Plain English Explanation

Artificial intelligence (AI) is becoming more common in the medical field, as data-driven algorithms can help with tasks like detecting and predicting breast cancer. However, these algorithms can be complex and difficult for doctors to understand. This makes it hard for them to trust the decisions made by the AI.

To address this, researchers have developed a type of AI called "explainable AI" or XAI. XAI aims to make the inner workings of AI models more transparent and understandable. This review looked at studies that used XAI techniques for breast cancer diagnosis and risk prediction.

The researchers found that the SHAP model was the most popular XAI method used in these studies. SHAP is a tool that can explain the predictions of various machine learning models, including the powerful tree-based models often used for medical applications.

The benefit of using SHAP and other XAI techniques is that they make AI-based health systems more transparent and trustworthy for doctors and patients. This can help improve the quality of care and outcomes for people with breast cancer.

Technical Explanation

The researchers conducted a systematic scoping review to investigate the application of XAI in breast cancer detection and risk prediction. They searched four major databases (Scopus, IEEE Explore, PubMed, and Google Scholar) for peer-reviewed studies published between January 2017 and July 2023 that used XAI methods with breast cancer datasets.

After screening the search results, 30 studies were included in the final analysis. The review found that the SHAP (SHapley Additive exPlanations) model was the most widely used XAI technique in breast cancer research. SHAP is a model-agnostic approach, meaning it can be used to explain the predictions of various machine learning models, including tree-based ensemble models, which are commonly used in this domain.

The researchers determined that the primary reasons for SHAP's popularity are its ability to explain model predictions, its suitability for high-performing models like tree-based ensembles, and its relative ease of implementation. SHAP has been used in breast cancer studies to explain model predictions, diagnose and classify biomarkers, and analyze prognosis and survival.

Critical Analysis

The review provides a comprehensive overview of the current state of XAI application in breast cancer research. However, the authors acknowledge that the study is limited by the scope of the included publications, which may not capture all relevant work in this rapidly evolving field.

Additionally, the review does not delve into the specific limitations or potential drawbacks of the SHAP model or other XAI techniques. While SHAP is widely used, there may be other XAI methods that are better suited for certain breast cancer applications or datasets. Further research is needed to explore the relative strengths and weaknesses of different XAI approaches in the medical domain.

It is also important to consider the broader challenges of implementing XAI in real-world healthcare settings, such as integrating these techniques into clinical workflows and ensuring that the explanations provided are meaningful and actionable for clinicians.

Conclusion

This systematic review highlights the growing importance of explainable AI in the field of breast cancer detection and risk prediction. By using XAI methods like SHAP, researchers and clinicians can gain more transparency and trust in the decision-making processes of AI-based medical systems.

As the use of AI continues to expand in healthcare, the adoption of XAI techniques will be crucial for ensuring the fairness, reliability, and interpretability of these technologies. The insights from this review can help guide future research and development efforts in this direction, ultimately leading to improved patient outcomes and enhanced trust in AI-powered medical applications.