Explainable AI-based feature importance analysis for ovarian cancer classification with ensemble methods

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Ovarian cancer (OC) is a leading cause of cancer deaths in women.
• Early-stage OC diagnosis is challenging due to non-specific symptoms.
• Current treatments offer marginal diagnostic improvements.

Purpose of the Study:

• To design a computer-aided diagnostic system for accurate ovarian cancer classification and detection.
• To utilize ensemble machine learning and explainable AI for deeper insights into diagnostic patterns.

Main Methods:

• Developed a three-stage ensemble model incorporating a game-theoretic approach with SHAP values.
• Evaluated and visualized results to identify key predictive features.
• Validated SHAP values using statistical methods (p-test, Cohen's d-test).

Main Results:

• Achieved a high diagnostic accuracy of 98.66%.
• Demonstrated the model's consistency and advantages over single classifiers.
• Validated feature importance using p-values and Cohen's d-values.

Conclusions:

• The AI-based method accurately detects, diagnoses, and prognoses OC using multi-modal data.
• The approach mimics clinical decision-making, offering reliable and consistent AI solutions.
• Potential for improved patient outcomes, reduced costs, morbidity, and mortality, especially in resource-constrained settings.