Interval Kernels for Combining Biometric Measurements from Multiple Prostate Samples per Patient in Prognostic Models...

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Prostate cancer is a leading cause of cancer death in the US.
• Accurate prognostic models are crucial for treatment decisions.
• Existing models often use single measurements from multiple tumor samples.

Purpose of the Study:

• To develop improved prognostic models for prostate cancer.
• To leverage information from multiple tumor samples per patient.
• To combine interval kernels with semi-supervised transduction for enhanced prediction.

Main Methods:

• Utilized support vector regression with a semi-supervised transduction framework.
• Implemented interval kernels to represent measurements across multiple samples using Hausdorff distance.
• Applied this combined approach to build prognostic models for prostate cancer.

Main Results:

• The proposed method using interval kernels yielded more accurate prognostic models.
• The semi-supervised transduction framework further enhanced the performance of these models.
• This novel combination demonstrates superior predictive capabilities.

Conclusions:

• Combining interval kernels and semi-supervised transduction offers a powerful new approach for prostate cancer prognostication.
• This method effectively utilizes multi-sample tumor data for improved accuracy.
• The findings suggest potential for better treatment strategies in prostate cancer.