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

Abstract

Spectral pre-processing is an essential step in data analysis for biomedical diagnostic applications of Ramanspectroscopy, allowing the removal of undesirable spectral contributions that could mask biological information usedfor diagnosis. However, due to the specificity of pre-processing for a given sample type and the vast number of potentialpre-processing combinations, optimisation of pre-processing via a manual ‘trial and error’ format is often time intensivewith no guarantee that the chosen method is optimal for the sample type. Here we present the use of high-performancecomputing (HPC) to trial over 2.4 million pre-processing permutations to demonstrate the optimisation on the pre-processing of human serum Raman spectra for colorectal cancer detection (CRC). The effect of varying pre-processingorder, using extended multiplicative scatter correction (EMSC), spectral smoothing, baseline correction, binning andnormalisation was considered. Permutations were assessed on their ability to detect patients with disease using arandom forest (RF) algorithm with training and testing data sets with 102 patients (510 spectra) and an independent testset of 439 patients (1317 spectra) in a primary care patient cohort. Optimising via HPC enables improved performancein diagnostic abilities, with sensitivity increasing by 14.6%, specificity increasing by 6.9%, positive predictive value (PPV)increasing by 3.4%, and negative predictive value increasing by 2.4% when compared to a standard pre-processingoptimisation. Ultimate values of these metrics are very important for diagnostic adoption, and once diagnosticsdemonstrate good accuracy these types of optimisations can make a significant difference to roll-out of a test anddemonstrating advantages over existing tests. We also provide tips/recommendations for pre-processing optimisationwithout the used of HPC. From the HPC permutations, recommendations for appropriate parameter constraints forconducting a more basic pre-processing optimisation are also detailed, thus helping model development for researchersnot having access to HPC.

A novel blood based triage test for colorectal cancer in primary care: a pilot study

“…All Raman spectra underwent data pre-processing including wavenumber calibration, data binning, smoothing, background subtraction, and normalisation. 19 , 20 A random forest-based machine learning model showed optimal performance and a diagnostic model was developed using a retrospective cohort of 300 patients with known clinical outcomes of CRC (histologically confirmed) or non-cancer control (normal colonoscopy) in a 50:50 split. 21 The Raman-CRC model was internally cross-validated using a repeated 5-fold cross validation of training data to produce a preliminary AUC and sensitivity and specificity values within R.…”

An observational cohort study to evaluate the use of serum Raman spectroscopy in a rapid diagnosis center setting

“…Spectra were processed using Savitsky-Golay [16] filtering with a filter length of 9 and a polynomial order of 4 for spectral smoothing, fluorescent background has been corrected using a modified polynomial fitting algorithm [17] modeling with a 5th order polynomial, and spectra have been normalized using standard-normal-variate [18] followed by normalization to the phenylalanine peak at 1002 cm − 1 . This procedure was developed in a high-performance computing workflow for optimization using a pre-processing package developed in-house [19]. A mean of 3 replicate spectra was taken as a patient's measurement.…”