Mechanistic Characterization of Biologically Inspired Oral Neutrophil Isolation for AI-Assisted Oral Inflammatory...

Non-invasive quantitative assessment of oral inflammatory burden could complement conventional periodontal evaluation; however, current clinical methods primarily characterize disease after structural and tissue changes have occurred. In our previously reported DePerio framework, we demonstrated the feasibility of enriching oral polymorphonuclear neutrophils (oPMNs) from saliva through their differential adhesion to a hydrophilic cornstarch (CS)-coated surface, followed by brightfield imaging and AI-assisted quantification. Building on this framework, here we present BioSA, with an emphasis on the mechanistic and quantitative characterization of this biologically inspired adhesion-based oPMN isolation process, analogous to leukocyte adhesion behavior in blood capillaries. We systematically investigate the contributions of surface physicochemical properties, oPMN surface characteristics, and physical forces governing preferential oPMN retention and epithelial-cell removal. The optimized adhesion-based isolation reduced epithelial-cell contamination by 91% while preserving >98% of oPMNs, outperforming the evaluated filtration- and poly-L-lysine (PLL)-based approaches. Following isolation, brightfield images were analyzed using an AI-assisted deep-learning model for automated oPMN quantification. Across 30 independent test days, BioSA measurements were strongly associated with those obtained using the reference HEMO method (R² ≈ 0.99; MAE ≈ 2.06 × 10⁵ cells/10 mL), while method agreement and systematic differences were further evaluated using Bland-Altman analysis. The deep-learning detector achieved 98% sensitivity, 97% precision, and an F1 score of 0.97 on a dataset containing 4617 manually annotated oPMNs. Using data-driven oPMN concentration ranges informed by prior literature, we further explored five oral inflammatory load (OIL) strata. These strata are hypothesis-generating and should not be interpreted as validated diagnostic stages or as replacements for the 2017 periodontitis staging and grading framework. Operating with standard brightfield microscopy and a cloud-based AI interface, BioSA reduces analysis time from 15 to 20 min to less than 2 min per sample. Collectively, this study extends the DePerio framework by providing mechanistic insight into biologically inspired adhesion-based oPMN isolation and demonstrates its potential for rapid, quantitative, and exploratory assessment of oral inflammatory burden.