AI-Assisted Saliva Diagnostics Using an Electrochemical Sensor Platform for Periodontitis Detection (SALIENCE)

Background and Rationale Periodontitis is a chronic inflammatory disease that affects the supporting tissues of the teeth and is one of the most prevalent oral diseases worldwide. Despite its high prevalence, early diagnosis remains a major challenge, as current diagnostic tools rely on retrospective measures such as pocket depth, bleeding on probing, and radiographic bone loss. These indicators reflect past tissue destruction rather than current disease activity, leading to delayed diagnosis and treatment.

Saliva is an attractive diagnostic fluid for non-invasive, real-time disease monitoring because it contains a complex mixture of biomarkers that reflect both oral and systemic health. However, its biological variability, matrix effects, and susceptibility to contamination have limited its reliability as a diagnostic medium. To overcome these challenges, this project integrates electrochemical sensing with artificial intelligence (AI)-driven data interpretation and synthetic data generation to enable robust saliva-based diagnostics.

Study Objectives

The overall objective is to develop and validate an AI-assisted electrochemical sensor platform capable of detecting biochemical patterns in saliva associated with periodontitis. Specific aims are to:

Design and optimize a modular, multi-electrode sensor platform for saliva analysis.

Develop and validate AI algorithms for real-time signal interpretation and disease classification.

Generate feasibility and proof-of-concept data for future clinical implementation in dental care.

Study Design and Methods This is an observational study involving saliva sampling from adult participants diagnosed with periodontitis and healthy control subjects. Approximately 25 periodontitis patients and an expanded control group will be recruited from a collaborating dental clinic.

The electrochemical sensor platform will include multiple electrode types (carbon, gold, platinum, palladium), each selected for specific electrochemical and biochemical properties. The electrodes will detect a broad spectrum of salivary biomarkers, including metabolites related to inflammation, oxidative stress, and microbial activity. The system supports several analytical modes, such as differential pulse voltammetry (DPV), which provides high sensitivity and resolution across multiple analytes.

Collected saliva samples will first be tested under controlled laboratory conditions to establish baseline sensor responses. Selected samples will also undergo complementary analyses (e.g., H-NMR, LC-MS) for calibration and validation. Sensor signals will be processed and classified using supervised and unsupervised machine learning models, including support vector machines and random forests. Synthetic data augmentation, based on generative models such as variational autoencoders, will be used to increase dataset diversity and improve model generalization, particularly in the early phase when sample numbers are limited.

Implementation Plan The project is organized over 24 months, beginning in September 2025.

Months 1-4: Selection and benchmarking of electrode materials in artificial and healthy saliva samples.

Months 5-8: Assembly of the first prototype with integrated AI signal acquisition and secure, GDPR-compliant data transmission.

Months 9-12: Development of synthetic datasets and initial training of machine learning models.

Months 9-19: Clinical saliva sampling and iterative model refinement using new data.

Months 21-24: Real-world testing of the proof-of-concept system in a clinical setting, data analysis, and dissemination of results.

Data Handling and Analysis All collected data will be de-identified and processed according to GDPR and ethical standards. Sensor outputs will be linked to anonymized clinical metadata. Machine learning models will be evaluated based on classification accuracy, sensitivity, specificity, and robustness to biological variability. Statistical analyses will include cross-validation and performance benchmarking across electrode configurations and AI models.

Expected Outcomes and Significance The study will result in a functional proof-of-concept prototype of an AI-assisted saliva diagnostic system. The platform aims to provide rapid, non-invasive detection of periodontitis at the point of care. If successful, it will represent a paradigm shift in periodontal diagnostics, enabling earlier detection, better-targeted interventions, and improved patient outcomes.

Beyond periodontitis, the modular design of the system allows adaptation for other diseases, including oral cancer, systemic inflammation, and metabolic disorders. This flexibility enhances its long-term potential as a scalable diagnostic technology for both dental and medical applications.

Ethical Considerations and Risk Management The study has been approved by the Swedish Ethical Review Authority (reference number: 2025-04853-01). All participants will provide written informed consent prior to sample collection. The study involves minimal risk, limited to the discomfort associated with saliva sampling. No experimental treatment or invasive procedures will be performed.

Potential technical risks, such as variability in saliva samples or device performance, will be mitigated through standardized sampling protocols and iterative prototype refinement. The modular hardware design ensures that individual components can be adjusted or replaced without compromising the overall system.

Collaboration and Expertise This interdisciplinary project brings together experts in clinical dentistry, biomedical engineering, and computer science. The study is led by Assoc. Prof. Shariel Sayardoust (Linköping University and Östergötland County), with collaborators Assoc. Prof. Magnus Falk, Assoc. Prof. Julia Davies, and Dr. Erdal Akin from Malmö University. Industrial collaboration with Redoxme AB supports hardware prototyping and integration.

Together, the consortium aims to establish the foundation for next-generation saliva diagnostics that combine AI, electrochemical sensing, and patient-centered care in everyday clinical practice.