Optimised Pre-Processing of Raman Spectra for Colorectal Cancer Detection Using High-Performance Computing.

Area of Science:

• Biomedical Diagnostics
• Spectroscopy
• Computational Biology

Background:

• Raman spectroscopy is vital for biomedical diagnostics, but spectral pre-processing is crucial yet time-consuming.
• Manual optimization of pre-processing methods for Raman spectra lacks efficiency and guarantees of optimality.
• Automated optimization is needed to improve diagnostic accuracy and facilitate clinical adoption.

Purpose of the Study:

• To optimize spectral pre-processing for human serum Raman spectra using high-performance computing (HPC).
• To enhance the diagnostic accuracy of colorectal cancer detection.
• To provide recommendations for pre-processing optimization with and without HPC access.

Main Methods:

• Utilized high-performance computing (HPC) to evaluate over 2.4 million pre-processing permutations.
• Investigated the impact of varying pre-processing order (Extended Multiplicative Scatter Correction, smoothing, baseline correction, binning, normalization).
• Assessed permutations using a Random Forest (RF) algorithm on Raman spectra from 102 patients (training) and 439 patients (testing).

Main Results:

• HPC-driven optimization significantly improved diagnostic performance.
• Sensitivity increased by 14.6%, specificity by 6.9%, positive predictive value by 3.4%, and negative predictive value by 2.4% compared to standard methods.
• Optimized parameters provide a basis for researchers without HPC access.

Conclusions:

• HPC-driven spectral pre-processing optimization substantially enhances diagnostic accuracy for colorectal cancer detection via Raman spectroscopy.
• The optimized methods offer significant improvements over standard approaches, crucial for clinical diagnostic adoption.
• Recommendations are provided for both HPC-based and basic pre-processing optimization strategies.