A network approach to compute hypervolume under receiver operating characteristic manifold for multi-class biomarkers

Area of Science:

• Biostatistics
• Medical Informatics
• Machine Learning

Background:

• Evaluating biomarkers for multi-class diagnostic discrimination requires computing the hypervolume under the ROC manifold (HUM).
• The traditional HUM calculation involves multiple integrations, leading to high computational costs in multi-class ROC analysis, especially with large sample sizes or numerous categories.
• Existing R packages for HUM computation can be inefficient for complex medical investigations.

Purpose of the Study:

• To introduce a novel, efficient graph-based computational method for hypervolume under the ROC manifold (HUM).
• To address the computational challenges associated with traditional HUM calculation in multi-class ROC analysis.
• To provide a faster and more scalable alternative for biomarker evaluation in medical research.

Main Methods:

• Developed a novel graph-based approach to compute HUM.
• The method avoids computationally intensive multiple summations.
• Implemented and tested the approach against existing R packages.

Main Results:

• The proposed graph-based method significantly improves computational efficiency for HUM calculation.
• Extensive simulation studies demonstrated superior performance compared to existing R packages.
• The method was successfully applied to two real biomedical datasets, showcasing its practical utility.

Conclusions:

• The novel graph-based approach offers an efficient and scalable solution for computing HUM.
• This method effectively reduces computational burden in multi-class ROC analysis, facilitating biomarker evaluation.
• The approach has practical implications for medical research involving complex diagnostic scenarios.