AI-powered quantitative analysis of sinus CT scans for enhanced clinical assessment and research

The National Institute of Biomedical Imaging and Bioengineering, part of the Department of Health and Human Services National Institutes of Health, awarded Towson University $211,406 on September 1, 2026, under the Discovery and Applied Research for Technological Innovations to Improve Human Health program (CFDA 93.286) to develop and validate AIRS (Artificial Intelligence-powered Rhinosinusitis Scoring), a clinically deployable system for automated, quantitative analysis of sinus CT scans in chronic rhinosinusitis patients.

AIRS integrates deep learning, computational fluid dynamics, and secure cloud deployment to overcome reproducibility and clinical alignment limitations in current CT-based disease assessment. The three-aim project establishes robust AI-driven segmentation tools on de-identified sinus CT datasets; integrates segmentations into a CFD pipeline to generate patient-specific physiologic biomarkers including airflow resistance, pressure gradients, and flow patterns, with surrogate models for rapid inference; and delivers a HIPAA-compliant cloud platform enabling clinical investigators to upload scans, receive automated segmentations and physiologic assessments, and export results in standardized formats (DICOM, CSV, JSON). Validation uses the UNC dataset across multiple CT scanners and acquisition protocols with cross-platform testing to confirm robustness in different computing environments and PACS-integrated workflows.

The award runs from September 1, 2026, through June 30, 2030, with place of performance in Towson, Maryland. Towson University is a subsidiary of the University System of Maryland.

Abstract

Chronic rhinosinusitis (CRS) affects millions of individuals annually and poses a substantial public health burden, yet current methods to evaluate disease severity remain subjective and poorly aligned with patient outcomes. Existing CT-based scoring systems lack reproducibility and fail to capture physiologic consequences of airway obstruction, limiting both clinical management and research innovation. This proposal aims to develop and validate a clinically deployable system-airs (artificial intelligence-powered rhinosinusitis scoring)-to provide automated, quantitative, and physiologically relevant analysis of sinus CT scans. Airs will integrate deep learning, computational fluid dynamics (CFD), and secure cloud deployment to overcome longstanding barriers in CRS imaging. Aim 1 will establish a robust segmentation and quantification tool by AI models on large DE-identified sinus CT datasets, generating reproducible measurements of anatomic structures and disease burden. Aim 2 will integrate these segmentations into a CFD pipeline, producing patient-specific physiologic biomarkers such as airflow resistance, pressure gradients, and streamline flow patterns. Surrogate models will be developed to enable rapid inference while maintaining accuracy. Aim 3 will deliver a secure, hipaa-compliant cloud platform for broad research use. This toolkit will allow clinical investigators to upload CT scans, receive automated segmentations and physiologic assessments, visualize overlays through a web-based viewer, and export standardized results (dicom, CSV, json). Validation will be conducted using the UNC dataset across multiple CT scanners and acquisition protocols, with additional cross-platform testing to confirm robustness in different computing environments and pacs-integrated workflows. The platform's utility will be assessed through reproducibility of automated measurements, correlation with patient symptoms, and efficiency of multicenter workflows. By uniting advanced AI methods with physiologic modeling and a user-friendly research toolkit, airs has the potential to transform CRS research. The system will enable standardized, automated analysis across institutions, foster discovery of objective biomarkers, and provide a foundation for future translational studies. If successful, airs will accelerate the development of precision diagnostics for CRS and establish a scalable framework applicable to other airway diseases. A central feature of the project is student engagement. Undergraduate and master's students at towson university will be trained in medical image annotation, algorithm development, and CFD modeling, gaining hands-on experience with modern computational and translational methods. These authentic research opportunities will prepare students for stem careers and advanced study, while contributing to towson's institutional mission of expanding biomedical research capacity.