AI-based autism identification from hyperspectral imaging detection of oxidative stress in pediatric red blood cells

Oxidative stress (OS) plays a key role in many pathologies, yet the non-invasive, label-free, and cost-effective detection remains a challenge. This study evaluates Hyperspectral Imaging (HSI) combined with AI to detect OS by identifying changes in red blood cell (RBC) membranes.

An OS model for the HSI procedure is established by treating EDTA-anticoagulated whole blood with 1.5% hydrogen peroxide (H₂O₂) to induce stress without cell lysis. Membrane fatty acid composition (lipidome) is analysed via gas chromatography, while HSI in dark-field microscopy captures spectral signatures and their distributions in healthy and insulted RBC. The HSI methodology is then applied to RBC samples from 31 neurotypical (NT) children and 27 children with Autism Spectrum Disorder (ASD), a condition linked to OS. A deep learning algorithm is used to classify the clinical samples based on the identified OS signatures.

Here, we show that significant spectral distribution differences are present in OS-exposed RBCs, which correlate with membrane lipidome remodelling. Notably, the OS-induced spectral differences in the H₂O₂ model mirror those observed between the ASD and NT groups. The AI-assisted analysis successfully classifies the pediatric cohort, achieving 93.2% accuracy in identifying ASD subjects.

HSI, guided by OS-specific modeling and integrated with AI, provides a robust, scalable method for membrane diagnostics. This approach offers a promising pathway for personalized medicine and the non-invasive monitoring of oxidative stress-related conditions.