Focus on pre-processing step to ensure the clinical transferability of Raman data acquired on lymphocytes in different experimental and instrumental conditions

Abstract The efficiency of Raman spectroscopy for the analysis of biomedical samples has been largely demonstrated at the proof-of-concept stage during one-off measurement campaigns. However, bringing these results to the patient bedside requires to fill the gap for transferability of Raman data acquired in different experimental and instrumental conditions during multicentric measurement campaigns. In this study, we propose to evaluate a solution consisting in the application of Raman data preprocessing specifically developed to remove the spectral variability induced by such different conditions. For this purpose, we compared Raman data of lymphocytes acquired during two independent measurement campaigns from fresh unstained glass blood smears originating from healthy individuals and patients with a B-cell chronic lymphocytic leukemia (CLL) at an advanced stage. The differences between these campaigns were i) the instrumental configuration of the Raman devices, ii) the hospital partner, iii) the smear preparation method. A preprocessing developed previously and efficient for a specific measurement campaign is shown obsolete for these multicentric data. A second preprocessing based on Extended Multiplicative Signal Correction is able to homogenize the data by neutralizing the signal variability between the two measurement campaigns. These conclusions are drawn from the analysis of the data by Principal Component Analysis to study the source of variability between the two campaigns and by Partial Least Squares – Discriminant Analysis to assess the separability between healthy and CLL patients.