This study presents a novel methodology for optimizing the number of Raman spectra required per sample for human bone compositional analysis. The methodology integrates Artificial Neural Network (ANN) and Monte Carlo Simulation (MCS). We demonstrate the robustness of ANN in enabling prediction of Raman spectroscopy-based bone quality properties even with limited spectral inputs. The ANN algorithms tailored to individual sex and age groups, which enhance the specificity and accuracy of predictions in bone quality properties. In addition, ANN guided MCS systematically explores the variability and uncertainty inherent in different sample sizes and spectral datasets, leading to the identification of an optimal number of spectra per sample for characterizing human bone tissues. The findings suggest that as low as 2 spectra are sufficient for biochemical analysis of bone, with R2 values between real and predicted values of v1/PO4/Amide I and ∼I1670/I1640 ratios, ranging from 0.60 to 0.89. Our results also suggest that up to 8 spectra could be optimal when balancing other factors. This optimized approach streamlines experimental workflows, reduces data and acquisition costs. Additionally, our study highlights the potential for advancing Raman spectroscopy in bone research through the innovative integration of ANN-guided probabilistic modeling techniques. This research could significantly contribute to the broader landscape of bone quality analyses by establishing a precedent for optimizing the number of Raman spectra with sophisticated computational tools. It also sets a novel platform for future optimization studies in Raman spectroscopy applications in biomedical field.
Keywords: Artificial intelligence; Artificial neural network; Bone; Bone quality; Monte Carlo simulation; Raman spectroscopy.
Copyright © 2024 Elsevier B.V. All rights reserved.