Hyperspectral + AI Empowers Microscopic Blood Analysis: A Nature Portfolio Study Opens a New Path for Non-Invasive Oxidative Stress Screening

A joint study published in Communications Medicine relied on Hyperspectral Dark-Field Microscopy (HDFM) imaging technology combined with artificial intelligence algorithms to establish a label-free, micro-volume blood red blood cell (RBC) oxidative stress detection protocol. This provides a completely new optical detection path for the non-invasive screening of oxidative stress-related conditions such as Autism Spectrum Disorder (ASD), and once again proves the application value of hyperspectral technology in identifying the molecular features of biological cells.

I. Oxidative Stress Detection: Traditional Methods Face Multiple Limitations

Oxidative stress is a physiological disorder formed by an imbalance between intracellular oxidation and antioxidant defenses. The red blood cell membrane is rich in polyunsaturated fatty acids and is extremely susceptible to oxidative damage, making it an ideal observation carrier to reflect systemic oxidation levels. Such damage alters membrane lipid composition, membrane protein structure, and ion-transporting enzyme activity, presenting clear associations with neurodevelopmental abnormalities and metabolic diseases.

Previous methods for detecting oxidative stress had obvious drawbacks: biochemical lipidomics analysis requires complex pretreatment and has a long detection cycle; fluorescent labeling imaging alters the cell's native state and is unsuitable for scenarios involving micro-volume blood sampling in children; conventional spectroscopy equipment can only collect single-band data, making it difficult to capture subtle changes in membrane structure. Clinical practice lacks a standardized detection workflow that balances micro-volume sampling, label-free operation, low cost, and high-throughput batch analysis capabilities.

II. Hyperspectral Imaging: Capturing the Exclusive "Spectral Fingerprint" of Red Blood Cells

This study utilized Hyperspectral Dark-Field Microscopy (HDFM). Unlike ordinary optical cameras that only record RGB three-color information, this technology can capture hundreds of continuous spectral bands in the 400–1000 nm visible to near-infrared range. It generates a 3D spectral data cube for each RBC pixel, simultaneously preserving spatial cell morphology and molecular scattering characteristics.

The entire standardized experimental process requires only 2 μL of EDTA-anticoagulated whole blood, with no need for staining or fluorescent labeling:

1.After sample preparation, place it under a 60x oil-immersion objective lens to select RBC observation areas with complete outlines and no overlapping;

2.The system collects dark-field scattering spectra and uses the Spectral Angle Mapper (SAM) algorithm to extract 8 set feature spectral endmembers, corresponding to different molecular components of the cell membrane such as cholesterol, phospholipids, hemoglobin, and membrane skeleton proteins;

3.Calculate the distribution proportions of these 8 types of spectra within the cells to quantify subtle changes in cell membrane lipid and protein structures.

The research team first established an in vitro oxidation model: healthy human RBCs were treated with 1.5% hydrogen peroxide to simulate the state of oxidative damage in vivo. Detection results showed that oxidative stimulation significantly changed the proportions of multiple spectral endmembers. Meanwhile, gas chromatography confirmed a decrease in total polyunsaturated fatty acids and an increase in the proportion of saturated fatty acids in the cell membrane. The spectral changes formed a stable correspondence with lipidome remodeling, proving that hyperspectral signals can objectively reflect the degree of membrane oxidative damage.

III. Clinical Cohort Validation: Spectral Features Match Physiological Changes in Children with Autism

The study included 58 male children aged 3–8 years, divided into 27 children with Autism Spectrum Disorder (ASD) and 31 neurotypical (NT) controls. Age and BMI indicators were matched between the two groups, and confounding factors such as infection, epilepsy, and nutritional supplements were excluded.

Comparing the RBC spectral data of the two groups revealed that the spectral change trends in ASD children were highly consistent with those of RBCs with hydrogen peroxide-induced oxidative damage:

Proportions of Spectra 1 and 2 showed a significant drop, while proportions of Spectra 5, 7, and 8 increased synchronously;

The Na+/K+-ATPase activity of the RBC membrane in children with ASD was significantly reduced, and enzyme activity levels showed significant correlations with the distribution of multiple spectral endmembers. This confirms that oxidative stress damages membrane protein functions while synchronously altering the cell's optical scattering characteristics.

Simply put, the membrane structural changes appearing in the RBCs of children with autism highly overlap with the spectral changes brought by artificially induced *in vitro* oxidative damage. This indicates that hyperspectral imaging can capture the systemic oxidative stress features associated with such conditions.

IV. AI Empowers Spectral Data to Achieve Sub-Population Stratification and Auxiliary Discrimination

Massive hyperspectral data features complex dimensions and redundant variables, making it difficult for manual statistics to mine deep patterns. The study introduced an Artificial Neural Network (ANN) combined with the TWIST variable selection algorithm to parse the data:

1.The algorithm automatically screens out invalid spectral variables while retaining core features that distinguish between the two populations;

2.A two-way cross-training validation was adopted: one set of samples was used for training and the other for blind testing, taking the mean of the two-way results to reduce bias;

3.The model demonstrated stable sensitivity and specificity in distinguishing ASD from neurotypical children within the pediatric cohort, achieving an Area Under the Curve (AUC) of 0.92.

The unsupervised neural network Auto-CM combined with Minimum Spanning Tree visualizes data associations. The mapping intuitively shows that ASD samples are strongly bound to Spectrum 5 features, while normal samples cluster with Spectrum 4, providing a reference direction for simplifying future testing metrics.

V. Value of Technical Implementation: A New Optical Detection Direction for Precision Medicine

The entire detection framework established by this study brings multiple practical values for monitoring oxidative stress-related conditions:

1.Sampling-Friendly: Requires only a micro-volume of peripheral blood with no invasive chemical labels, making it suitable for multiple follow-up monitoring sessions in children;

2.Rapid and Reproducible Detection: Standardized imaging workflow allows testing completion within 5 hours of sample collection, with small repeat-measurement errors;

3.Multi-Scenario Extension Potential: In addition to neurodevelopmental disorders, cardiovascular and metabolic oxidative stress-related conditions can also be observed relying on RBC spectral models;

4.Compatible with Intelligent Detection: Spectral data can be batch-imported into AI models to reduce subjective bias caused by manual interpretation, fitting primary-level mass screening scenarios.

It should be objectively noted that this scheme currently serves only as an auxiliary tool for assessing oxidative stress status and cannot be used alone as a basis for definitive disease diagnosis. Larger-scale, multi-center clinical cohorts are still required to improve and validate the system.

Product Recommendation:FigSpec FS-23 Imaging Hyperspectral Camera

Image Resolution: 1920*1920

Spectral Range: 400-1000nm

Spectral Resolution (FWHM): 2.5nm

Spectral Channels: 1200

Disclaimer

This content is compiled based on public academic literature (Literature link:https://www.nature.com/articles/s43856-026-01581-y solely for industry technical discussions and popular science learning. It does not constitute any commercial commitment or investment reference basis. Various experimental data and conclusions listed in the text are subject to interference from multiple variables such as testing environments, individual sample differences, and model construction plans. If applied in actual scenarios, relevant effects must be verified through separate on-site testing according to specific conditions.