INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 868
Integrating AI‑Based Orbital Ultrasound with Tear Proteomic
Biomarkers for Precision Diagnosis of Orbital Inflammatory
Disorder (OID), Including Graves’ orbitopathy (GO)
Hadi Khazaei* ,Behrooz Khajehee1, Danesh Khazaei 2, Kaneez Abbas3, Nashrah Junejo 4, Majd Oteibi5,
Faryar Etesami 2, Bala Balaguru 3
1 University of Milano-Bicocca, USA
2Department of Chemistry, Portland State University, USA
3Athreya Medtech, USA
4Washington State University, USA
5
Validus Institute Inc. , USA
*
Corresponding Author
DOI: https://doi.org/10.51584/IJRIAS.2025.100900086
Received: 16 Sep 2025; Accepted: 22 Sep 2025; Published: 23 October 2025
ABSTRACT
Graves’ orbitopathy (GO) is a potentially blinding manifestation of Orbital Inflammatory Disorder (OID). Although
characteristic clinical signs and imaging features (bilateral exophthalmos, extra‑ocular muscle swelling) often guide
diagnosis, there remain substantial diagnostic and prognostic challenges. Approximately half of patients with Orbital
Inflammatory Disorder (OID) develop orbitopathy, but only a minority experience sight‑threatening diseases. The
inability to predict who will progress and the lack of biomarkers to distinguish inflammatory versus fibrotic stages hinder
timely intervention. Recent research highlights two complementary advances: (1) tear proteomics can reveal
disease‑specific protein signatures and offers a non‑invasive source of biomarkers; and (2) an AI‑assisted orbital
ultrasound proof-of-concept model has been developed, using Google Vertex AI Platform, that can differentiate orbital
inflammatory disease (OID) from non‑inflammatory orbitopathy (NIO) with high precision. This proposal aims to
integrate tear proteomic biomarkers with AI‑based orbital ultrasound to create a hybrid diagnostic workflow that
enhances sensitivity, specificity, and prognostication in OID. Longitudinal tear sampling from OID, GD without
orbitopathy and other inflammatory controls will be coupled with mass–spectrometry–based proteomic profiling and
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 869
ELISA validation. Concurrently, an updated AI model will be retrained on new OID cases and used as a confirmatory
imaging layer. The hybrid algorithm will cross‑validate proteomic and imaging signals to provide a precision diagnostic
framework for early detection and staging of OID.
Understanding Orbital Inflammatory Disorders
Understanding Orbital Inflammatory Disorders Orbital inflammatory disorders encompass a diverse group of conditions
characterized by inflammation in the tissues surrounding the eye. These disorders can lead to a myriad of symptoms,
including proptosis, pain, diplopia, and visual impairment. While the etiology of OIDs can vary, encompassing
autoimmune, infectious, and idiopathic causes, the heterogeneity of presentations often poses a challenge for clinicians in
determining the most appropriate and effective treatment strategies.
The Role of Precision Medicine
Precision medicine in OIDs involves a comprehensive analysis of the patient’s genetic makeup,
molecular profiles, and other relevant factors to identify targeted therapeutic interventions. Genetic
testing, advanced imaging techniques, and biomarker analysis play pivotal roles in this approach,
allowing clinicians to pinpoint specific molecular pathways involved in the inflammatory process.
For instance, identifying specific genetic markers associated with autoimmune OIDs can guide the
selection of immunomodulatory agents tailored to the patient's genetic predisposition. Precision
medicine also enables the identification of potential side effects and helps predict treatment response,
thereby minimizing adverse reactions and optimizing therapeutic outcomes.
Evidence-Based Approaches in Ophthalmology
The foundation of evidence-based medicine involves integrating the best available scientific evidence with clinical
expertise and patient preferences. In the context of OIDs, this means relying on well-designed clinical trials, systematic
reviews, and meta-analyses to inform treatment decisions.
Randomized controlled trials (RCTs) evaluating different treatment modalities, including
corticosteroids, immunosuppressive agents, and biologics, have provided valuable insights into their
efficacy and safety profiles. Understanding the level of evidence supporting each intervention is crucial
for clinicians to make informed decisions about the most appropriate course of action for individual
patients.
AI‑assisted orbital ultrasound
Figure 1: Normal/Abnormal Orbital Ultrasound Images
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 870
Vertex AI, a cloud-based machine learning platform developed by Google, supports both customized model development
and automated (AutoML) training approaches. In the “OIDvsNIO-NoAn” case study, a point-of-care B-scan ultrasound
dataset was systematically curated to enable the differentiation of Orbital Inflammatory Disease (OID) from Non-
Inflammatory Orbitopathy (NIO). A balanced dataset of 140 non‑annotated images (112 Training, 14 Validation,
14 Test) was used; the AutoML classification trained model achieved a PR AUC of 0.986 with 92.9 % precision and
92.9 % recall. Notably, 100 % of OID images and 86 % of NIO images were correctly identified, emphasizing high
sensitivity for inflammatory disease. The study highlighted that balancing the dataset and removing annotation noise
markedly improved performance. These findings show that AI-assisted orbital ultrasound has the potential to be a
reliable tool for real-time confirmatory diagnosis of orbital pathology.
DISCUSSIONS
Despite advances in clinical assessment, imaging, and novel therapies, accurate diagnosis and staging of
Graves’ orbitopathy remain difficult. Current reliance on clinical features and serum thyroid-
stimulating immunoglobulin (TSI) involves subjective grading and imaging that do not reliably capture
molecular activity. No biochemical marker currently distinguishes the active inflammatory phase from
the chronic fibrotic phase or predicts progression to vision-threatening disease.
Imaging plays an essential role in assessment. Computed tomography (CT) and magnetic resonance imaging (MRI) can
localize orbital changes and assess disease activity, but they are expensive and may involve radiation exposure or
contrast administration. There is a lack of predictive biomarkers to identify which patients with OID may develop
orbitopathy or transition from inflammatory to fibrotic phases. Meanwhile, AI models trained to distinguish OID from
NIO have demonstrated high accuracy, but these models have yet to be applied to OID-specific cohorts or integrated
with molecular biomarkers. Orbital ultrasonography offers a more accessible and safer alternative, though its
interpretation can be highly operator-dependent and subjective. That’s why we propose a combined approach to integrate
tear proteomics to identify disease-specific proteins using AI-enhanced orbital ultrasound to assess inflammatory
changes. This approach may offer a robust, non-invasive approach for diagnosing and staging Graves’ orbitopathy.
Using tear proteomics to detect disease‑specific proteins and AI‑driven orbital ultrasound to confirm inflammatory
changes could provide a robust, non‑invasive framework for diagnosing and staging OID. Such integration may enable
precision medicine, matching therapy to disease activity, predicting progression, and reducing unnecessary interventions.
Figure 2: Precision-Recall Curve/ by Threshold OIDvsNIO-NoAn Project.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 871
RESEARCH & METHODOLOGY
Tear Proteomic Biomarkers Validation
Validate tear proteomic biomarkers for differentiating OID from other orbital inflammatory diseases. Conduct
longitudinal mass spectrometry profiling of tear samples from patients with GO, Graves’ disease without orbitopathy,
and control groups, including sarcoidosis, IgG4-related orbitopathy, granulomatosis with polyangiitis, Sjögren’s
syndrome, and nonspecific orbital inflammation. Identify individual proteins or protein panels that distinguish GO from
these conditions and evaluate their correlation with clinical activity scores.
Proteomic Analysis and Quantification of Human Tears in Graves Orbitopathy (GO)…
Figure 3: The data analysis flagged around 170 differential candidates between 6 pairs of OD-pinch (DPIN)
and OS-proparacaine (SPRO) samples.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 872
AI Imaging Integration as a confirmatory diagnostic layer for tear proteomic findings.
Integrate an AI-based imaging model as a confirmatory diagnostic layer for tear proteomic findings. Retrain and refine
the Vertex AI model training methods using newly acquired orbital B-scan ultrasound images from the study cohort.
Assess the model’s ability to differentiate GO from non-inflammatory orbitopathy and to evaluate disease activity or
fibrosis. Use imaging outcomes to validate tear-based molecular signatures and provide anatomical confirmation.
Figure 5: Butterfly Handheld Ultrasound - Used in the OIDvsNIO‑NoAn project.
Clinical Workflow
1. Study design and participants: Recruit three cohorts: (a) patients with clinically active GO; (b) patients with
GD without orbitopathy; and (c) controls with other orbital inflammatory disorders (sarcoidosis, IgG4‑related
orbitopathy, GPA, Sjögren’s syndrome, nonspecific orbital inflammation).
Approach: Collect clinical activity scores, quality‑of‑life metrics, and imaging data at baseline and follow‑up
visits (e.g., 0, 6, and 12 months). Identify single proteins or multi-protein panels to differentiate GO from
comparators.
2. Tear sample collection and proteomic analysis: Tear fluid will be collected using a standardized method
(either Schirmer strips or cellulose sponges, per protocol) to ensure reproducibility and minimize sampling-
induced variability in protein content. To control for technique-dependent biases, collection and extraction
procedures will be carefully harmonized across centers. Extracted proteins will be labeled with tandem mass tags
and analyzed by liquid chromatography-tandem mass spectrometry. Rigorous phased validation—including
ELISA in independent patient sets and planned multi-center expansion—will bolster clinical adoption and
generalizability. Only consistently differentially expressed proteins across replicates will be carried forward as
diagnostic biomarkers.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 873
Figure 6: Tear protein digestion and tandem mass tag (TMT) labeling, mass spectrometric analysis,
Figure 7: Tear protein digestion and tandem mass tag (TMT) labeling, mass spectrometric analysis.
3. Orbital ultrasound imaging and AI model development:
Retrain and refine the Vertex AI model with new orbital B-scan images from the study cohort. using a portable
handheld device analogous to the Butterfly iQ+ used in the OID vs NIO study. Images will include
cross‑sectional scans through the optic nerve and extra‑ocular muscles. Data will be anonymized and uploaded to
Vertex AI, a cloud-based platform designed for building machine learning models, for training model training
methods to perform image prediction with minimal coding. Test the model’s ability to distinguish OID from
non-inflammatory orbitopathy.. Then differentiate active inflammation from fibrosis. Training, validation, and
testing sets will be balanced, and hyperparameters optimized. Performance metrics (precision, recall, and
PR AUC) will be compared with those reported for “OIDvsNIO‑NoAn”. The model will be evaluated for its
ability to differentiate active OID from inactive OID and from other disorders.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 874
4. Hybrid diagnostic algorithm: Develop a decision‑support pipeline that integrates tear proteomic profiles
(continuous variables or risk scores) with AI‑derived imaging probabilities. Machine‑learning techniques (e.g.,
logistic regression, random forest, or neural networks) will be employed to combine features. The primary
endpoint will be diagnostic accuracy for OID; secondary endpoints include sensitivity to disease activity and
prediction of progression. Cross‑validation and external validation cohorts will ensure generalizability.
5. Statistical analysis: Proteomic data will be normalized and analyzed using multivariate statistics (principal
component analysis, hierarchical clustering). Differential protein expression between groups will be assessed
with false discovery rate control. AI-trained model performance will be assessed by confusion matrices,
precision–recall curves, and ROC analysis. The hybrid model will be compared against individual modalities
using the area under the curve (AUC) and decision‑curve analysis.
6. Ethical considerations: Securing approval from an institutional review board and obtaining informed consent.
Patient privacy must be protected through de-identification and secure data storage. Participants will be informed
about the experimental nature of proteomic analysis and AI imaging. The process also includes registration on
ClinicalTrials.gov, and compliance with GDPR/HIPAA standards for storage.
Anticipated Outcomes
1. Identification of tear biomarkers: We anticipate discovering a panel of tear proteins whose expression
distinguishes GO from GD without orbitopathy and other inflammatory orbitopathies. These may include
inflammatory mediators, extracellular matrix proteins, or immune signaling molecules.
2. Validated AI imaging Model: By training on new OID cases, the Vertex AI model is expected to achieve
precision and recall comparable to the “OIDvsNIO” study (PR AUC ~ 0.986). We expect the model to
accurately detect active versus inactive OID, identifying inflammatory changes such as muscle swelling while
distinguishing non‑inflammatory conditions.
3. Hybrid diagnostic workflow: Combining tear proteomic signatures with AI imaging probabilities should
improve diagnostic sensitivity and specificity over either modality alone. The algorithm will generate a
composite risk score and provide staging information (inflammatory vs fibrotic). We anticipate that this
framework will detect subclinical OID before overt clinical signs and better predict disease progression.
Significance and Innovation
This study integrates two advanced diagnostics—omics-based tear biomarkers and AI-assisted orbital ultrasound—to
meet an unmet clinical need in OID. Tear sampling is noninvasive and reflects ocular pathology, enabling early detection
and personalized monitoring. Validating tear proteomic markers and correlating them with clinical activity addresses the
lack of predictive indicators. AI-based orbital ultrasound is portable and radiation-free, improving accuracy in
distinguishing inflammatory from noninflammatory disease. Together, these methods enhance diagnostic confidence,
reduce misclassification, enable precise staging, and support earlier treatment, better selection for emerging therapies
(such as IGF-1 receptor inhibitors), and robust response monitoring. The approach is adaptable to other orbital
inflammatory diseases and advances precision ophthalmology.
Figure 8: Integrated framework for diagnosing and staging Graves’ Orbitopathy.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 875
Strengths
1. Non-invasive diagnostics: Tear sampling avoids radiation or invasive procedures, while ultrasound is safe and
repeatable.
2. Complementary modalities: Biomarker detection provides molecular specificity, and AI-ultrasound offers
anatomical and functional insight.
3. Clinical impact: Together, these tools may enable earlier therapeutic intervention and more tailored treatment
strategies.
4. Broader applicability: The framework could be extended to other orbital and autoimmune inflammatory
diseases.
Limitations
1. Validation required: Tear proteomic biomarkers are still under investigation and require large-scale, multi-
center validation before clinical adoption.
2. Operator dependency: While AI improves ultrasound interpretation, variability in acquisition and image
quality may still affect diagnostic accuracy.
3. Resource constraints: Implementation may be limited in low-resource settings without access to advanced
omics platforms or AI integration.
4. Dynamic disease course: Biomarkers and imaging findings may change over time, requiring repeated
assessments to remain clinically useful.
Future research :
1. Biomarker validation will be phased, progressing from carefully controlled pilot cohorts to large, multi-center
studies, with external validation at each stage.
2. Operator dependency in ultrasound will be minimized via standardized protocols, centralized training, and
certification.
3. For low-resource settings, platform scaling and biomarker panel streamlining will be actively investigated.
4. Dynamic and longitudinal data analysis will be integrated to reflect the real-world course of OID, with advanced
statistical modeling for time-dependent biomarkers and image changes.
5. Diverse, external validation cohorts will be prioritized to establish broad clinical applicability and robustness of
the diagnostic pipeline.
CONCLUSIONS
Orbital Inflammatory Disorder (OID) remains a clinically challenging disorder to diagnose and stage due to its
heterogeneous presentation, unpredictable disease trajectory, and the absence of reliable biomarkers that distinguish
active inflammatory from chronic fibrotic phases. While orbital ultrasound is accessible, there is still operator
dependency, variability, and interpretive subjectivity. These diagnostic shortcomings contribute to delays in appropriate
treatment and complicate patient selection for emerging therapies. The proposed integration of tear proteomics and AI-
assisted orbital ultrasound represents a significant step toward overcoming these limitations. Tear proteomics, as a non-
invasive and easily repeatable approach, offers a unique opportunity to capture disease-specific molecular signatures
reflective of orbital pathology. By validating proteomic biomarkers and correlating them with disease activity, it is
possible to establish predictive and prognostic indicators that have long been missing in OID management.
Complementing this, AI-assisted orbital ultrasound provides a radiation-free portable imaging modality that enhances
diagnostic precision by reducing operator variability and enabling reliable differentiation between inflammatory and non-
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 876
inflammatory disease stages. This approach also aligns with the broader movement toward precision ophthalmology,
where individualized risk stratification and treatment monitoring can improve both short- and long-term patient
outcomes and provide early diagnostic interventions.
ACKNOWLEDGEMENT
We acknowledge Oregon Health Science University, Department of Orbital Inflammatory Disorders, led by Professor
Jim Rosenbaum, Dr John Ng, and Dr. Davin, for their professional advice and contribution in the tear prototype project.
We also acknowledge Portland State University, Department of Electrical and Computer Engineering, led by Dr. Faryar
Etesami and his team, for their guidance and training.
All the above materials are intellectual properties of the authors with the original Datasets and publications.
Corresponding Author Prof. Hadi Khazaei is the author of the Springer Nature textbook titled:
Fundamentals of Orbital Inflammatory Disorders. Springer Nature Switzerland; 2025:15-29. doi:10.1007/978-3-031-
85768-3_3
REFERENCES
1. Khazaei H, Khazaei D, Verma R, et al. The Potential of Tear Proteomics for Diagnosis and Management of
Orbital Inflammatory Disorders. In: Khazaei H, ed. Fundamentals of Orbital Inflammatory Disorders. Springer
Nature Switzerland; 2025:15-29. doi:10.1007/978-3-031-85768-3_3
2. H. Khazaei et al., Leveraging Vertex AI for Automated Ultrasound Image Analysis: A Comprehensive Review.
3. Khazaei H. Assessment of Disease Activity of Graves’ Using Orbital Ultrasonography. In: Khazaei H, ed.
Fundamentals of Orbital Inflammatory Disorders. Springer Nature Switzerland; 2025:67-74. doi:10.1007/978-
3-031-85768-3_7
4. Khazaei H. Proteomic Analysis and Quantification of Human Tears in Graves Orbitopathy (GO): A Novel
Approach. In: Khazaei H, ed. Fundamentals of Orbital Inflammatory Disorders. Springer Nature Switzerland;
2025:31-50. doi:10.1007/978-3-031-85768-3_4
5. Aass, C., Norheim, I., Eriksen, E. F., Thorsby, P. M., & Pepaj, M. (2015). Single unit filter-aided method for
fast proteomic analysis of tear fluid. Analytical Biochemistry, 480, 1–5.
https://doi.org/10.1016/j.ab.2015.04.017
6. Bartalena, L., Kahaly, G. J., Baldeschi, L., Dayan, C. M., Eckstein, A., Marcocci, C., ... & EUGOGO. (2021).
The 2021 European Group on Graves’ orbitopathy (EUGOGO) clinical practice guidelines for the medical
management of Graves’ orbitopathy. European journal of endocrinology, 185(4), G43-G67.
7. Khazaei H, Khazaei D, Ashraf D, Mikkilineni S, Ng JD. Orbital Ultrasonography for Meaningful Orbital
Inflammatory Responses in Thyroid-Associated Orbitopathy (TAO). In: Khazaei H, ed. Fundamentals of
Orbital Inflammatory Disorders. Springer Nature Switzerland; 2025:59-65. doi:10.1007/978-3-031-85768-3_6
8. Lipor J, Khazaei H, Khazaei D, Franck A, Ng JD, Etesami F. Deep Machine Learning in Eye Ultrasound. In:
Khazaei H, ed. Fundamentals of Orbital Inflammatory Disorders. Springer Nature Switzerland; 2025:145-158.
doi:10.1007/978-3-031-85768-3_17
9. Bahn, R. S. (2010). Graves’ ophthalmopathy. New England Journal of Medicine, 362(8), 726–738.
https://doi.org/10.1056/NEJMra0905750
10. Khazaei H, Khazaei D, Abbas K, Ashraf D, Ng JD. Integrating Imaging Bioinformatics in Ophthalmology. In:
Khazaei H, ed. Fundamentals of Orbital Inflammatory Disorders. Springer Nature Switzerland; 2025:127-132.
doi:10.1007/978-3-031-85768-3_15
11. Khazaei H, Khazaei D, Ashraf D, Mikkilineni S, Ng JD. Overview of Orbital Ultrasonography. In: Khazaei H,
ed. Fundamentals of Orbital Inflammatory Disorders. Springer Nature Switzerland; 2025:51-57.
doi:10.1007/978-3-031-85768-3_5
12. Oteibi M, Khazaei H, Abbas K, Balaguru B, Williams AR, Etesami F. Breast Imaging and omics for Non-
Invasive Integrated Classification (BIONIC). Int J Res Sci Innov. 2024;10(8):826–835.Manuscript ID:
10IJ08ASN8264.
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue IX September 2025
www.rsisinternational.org Page 877
13. Smith, T. J., & Hegedüs, L. (2016). Graves’ disease. New England Journal of Medicine, 375(16), 1552–1565.
https://doi.org/10.1056/NEJMra1510030
14. Dolman, P. J. (2017). Evaluating Graves’ orbitopathy. Best Practice & Research Clinical Endocrinology &
Metabolism, 26(3), 229–248. https://doi.org/10.1016/j.beem.2012.09.002
15. Huang, D., Chen, L., & Yang, H. (2021). Tear proteomics in ophthalmology: Progress, potential, and
perspectives. Proteomics: Clinical Applications, 15(3), e2000020. https://doi.org/ 10.1002/ prca .202000020