Comparison of spectral imaging modalities for identification and classification of air-liquid interface biofilms

Biofilms formed at air-liquid interfaces represent a significant challenge for the food industry due to their resistance to cleaning and the difficulty of detection. This study systematically investigates the ability of five bacterial species (E. coli, B. subtilis, C. sakazakii, P. fluorescens and L. plantarum) to form interfacial biofilms on stainless steel AISI 304, stainless steel AISI 316, and aluminium. Biofilm formation was confirmed by crystal violet staining and quantified via viable cell count. C. sakazakii produced the highest biofilm biomass (>6 log CFU cm^-2 on stainless steel, in milk), whereas P. fluorescens (minimal media) and L. plantarum (milk) yielded no viable cells, suggesting limitations in recovery or altered physiological states. Spectral imaging techniques (FT-IR, Raman, SWIR, VNIR) were used to analyse biofilm formation. Results demonstrated a strong influence of surface type and growth medium on spectral signatures and classification performance, with stainless steel substrates supporting higher bacterial growth than aluminium under nutrient-rich conditions. Principal Component Analysis (PCA) showed clear separation of contaminated samples in milk, while broth samples were more stable but less distinct. All techniques detected contamination, with SWIR and VNIR providing more consistent classification performance, particularly when combined with sub-regional analysis and data fusion. Broth-incubated biofilms were more reliably classified than those in milk, likely due to reduced spectral complexity. The variability across test sets underscores the need for careful sample selection and robust model validation. Overall, spectral imaging is a promising non-destructive approach for biofilm detection, with SWIR and VNIR showing the best performance.