ATR_FTIR spectroscopy of saliva and machine learning as a screening test for the Sjögren disease

Background:

Sjögren’s disease is a chronic autoimmune disorder that primarily targets the body’s moisture-producing glands, leading to symptoms such as dry mouth and dry eyes. Diagnosing this condition is particularly challenging due to the overlap of its symptoms with other diseases and the absence of a single, definitive biomarker. The field of non-invasive diagnostics has therefore become increasingly important, as clinicians and researchers seek reliable, accessible, and patient-friendly methods for early detection and monitoring of autoimmune diseases. Saliva, as a readily available and non-invasively collectible biofluid, offers a promising window into the body’s biochemical state, making it an attractive medium for disease screening and diagnosis. Current diagnostic approaches for Sjögren’s disease, such as minor salivary gland biopsies, Schirmer’s tests, and serological assays, are often invasive, time-consuming, and lack sufficient specificity and sensitivity. These methods can be uncomfortable for patients, require specialized clinical settings, and may not always yield conclusive results, especially in early or atypical cases. Additionally, traditional spectroscopic and chemometric techniques used to analyze saliva or other biofluids struggle to distinguish diagnostically relevant signals from noise or unrelated biochemical variations, particularly given the heterogeneity of biological samples. This limitation hampers the effectiveness of machine learning models, as irrelevant or low-quality data can lead to poor generalization, reduced interpretability, and increased risk of misclassification—highlighting the pressing need for more robust, accurate, and user-friendly diagnostic solutions.

Technology Overview:

This technology offers a non-invasive diagnostic solution for Sjögren’s Disease by integrating attenuated total reflectance Fourier-transform infrared (ATR-FTIR) spectroscopy of saliva with a sophisticated machine learning framework. Saliva samples are analyzed using ATR-FTIR to generate comprehensive spectral fingerprints that reflect the biochemical composition associated with disease states. These high-dimensional spectra are then processed by an artificial neural network, which is enhanced by Monte Carlo Dropout (MCD) uncertainty estimation. The MCD mechanism serves as a critical preprocessing step, filtering out spectra with low classification confidence to ensure that only diagnostically relevant and high-quality data are used for model training. This approach not only increases the accuracy of disease detection but also supports rapid, repeatable, and painless sample collection, making it suitable for point-of-care diagnostics, home testing, and longitudinal monitoring. What differentiates this technology is its innovative use of MCD-based uncertainty estimation to address a fundamental challenge in biospectroscopic diagnostics: the inherent variability and noise within biofluid samples, where only a subset of spectra may be diagnostically informative. By systematically identifying and excluding ambiguous or uninformative spectra before model training, the solution enhances the interpretability, generalization, and robustness of the neural network classifier. This results in improved diagnostic performance, reduced risk of overfitting, and greater resilience to sample heterogeneity—critical for diseases like Sjögren’s, where biomarkers are sparse and unevenly distributed. The methodology is broadly applicable to other diseases and biofluids, enabling scalable, affordable, and accessible diagnostics across diverse clinical and research settings, and represents a significant advancement in the intersection of vibrational spectroscopy and artificial intelligence.

Advantages:

•    Non-invasive and painless saliva-based diagnostic method for Sjögren’s Disease
•    Rapid and accessible screening suitable for point-of-care and home testing
•    Improved diagnostic accuracy through machine learning classification of biochemical spectral fingerprints
•    Enhanced model robustness and interpretability via Monte Carlo Dropout uncertainty filtering of low-confidence spectra
•    Reduction of noise and irrelevant data, leading to better generalization and reliability
•    Applicable to longitudinal monitoring and disease progression tracking
•    Potentially extendable to other diseases and biofluids with heterogeneous diagnostic signals
•    Supports development of scalable, affordable diagnostic tools for diverse clinical and research settings

Applications:

•    Point-of-care Sjögren’s disease screening
•    Home-based saliva diagnostic kits
•    Longitudinal disease monitoring tools
•    Portable autoimmune disease diagnostics
•    Research tool for biofluid analysis

Intellectual Property Summary:

Patent application filed

Stage of Development:

Inquire for more information

Licensing Status:

This technology is available for licensing.