Abstract
Two custom fibre optic probes designed for bulk surface (Probe B) and sub-surface (Probe A) assessment of gastrointestinal tissue were evaluated for signal performance. Probe A was designed with the capability of using spatially offset Raman spectroscopy (fixed for 0 mm, 2.5 mm, and 5 mm lateral offset light collection sites) and near-infrared Raman spectroscopy. Probe B was equipped with a notch filter at the probe tip along with Raman (785 nm), near infrared, and optical coherence tomography fibre lines.
The quantitative performance of different configurations of Probe A and Probe B were investigated using model pharmaceuticals (hydroxypropyl methylcellulose, titanium dioxide, lactose, and indomethacin) in tablet formulations. The resulting data was analysed using principal component analysis and partial least squares regression individually or using a low-level fusion approach. Hydroxymethyl cellulose (RSMEP = 2.3%) was most effectively quantified by near-infrared spectroscopy via Probe B with a normal background reference and laser interference during measurements. Titanium dioxide (RSMEP = 0.42%) was best quantified by PLSR model developed using low-level fused data of Raman and near-infrared data collected by Probe B. Lactose (RSMEP = 3.7%) was best quantified by PLSR model developed using low-level fused 785-Raman data of all configurations of Probe A. Indomethacin (RSMEP = 0.79%) was best quantified by PLSR model developed using low-level fused 785-Raman and near-infrared data of all configurations of Probe B.
The qualitative performance of Probe A, Probe B, and a hand-held near-infrared sensor (M-NIR) were investigated by using low- and mid-level fusion methods, principal component analysis, support vector machines, linear discriminatory analysis, and partial least squares discriminant analysis to classify different types of store-bought kiwifruit. Gold and green kiwifruit types were best classified (classification accuracy = 0.98) using SVM model using M-NIR near-infrared data and PCA-LDA model using low-level fused Raman spatial offsets (0, 2.5, and 5 mm) and near-infrared data collected using Probe A. The relative ripening of kiwifruit samples during the 14-day study were attempted to be tracked using the spectroscopic techniques and classification models. Among the individual techniques, SVM model using A-NIR data best identified (classification accuracy = 1.00) kiwifruit samples measured on Day 1, Day 8, and Day 14.
A preliminary study was conducted using spatially offset Raman spectroscopy via Probe A, Fourier-transform Raman spectroscopy, and near-infrared spectroscopy measuring pig intestine biopsies to determine optimal instrument parameters and assess their suitability of measuring tissue samples. Biochemical signals associated with proteins and lipids were observed in the Fourier-transform Raman spectra of pig intestine biopsy sample. This demonstrated the potential of spatially offset Raman spectroscopy, Fourier-transform Raman spectroscopy, and near-infrared spectroscopy to be applied to ex-vivo biopsy tissue samples collected from individuals.