# Artificial Intelligence-Enhanced Cone-Beam Computed Tomography for the Diagnostic Evaluation of Chronic Rhinosinusitis: A Systematic Review

## TL;DR

This review assesses how artificial intelligence improves cone-beam CT for diagnosing chronic rhinosinusitis, finding it accurate but needing more validation.

## Contribution

The study systematically evaluates AI-enhanced CBCT for CRS diagnosis, highlighting its potential and current limitations.

## Key findings

- AI-enhanced CBCT shows high sensitivity and specificity for maxillary sinusitis and mucosal thickening detection.
- Current AI methods for CRS have moderate methodological quality and limited external validation.
- Applications include anatomical variant detection and classification, but evidence certainty varies by outcome.

## Abstract

Chronic rhinosinusitis (CRS) is one of the most common chronic inflammatory conditions globally, affecting a significant proportion of adults. It is associated with reduced quality of life, recurrent healthcare utilization, and substantial economic burden. Imaging plays a pivotal role in its diagnosis and management. Cone-beam computed tomography (CBCT) has emerged as a valuable imaging modality that provides high-resolution, three-dimensional reconstructions with lower radiation exposure compared with conventional CT. However, CBCT interpretation requires expert knowledge, remains time-intensive, and is subject to inter-observer variability. Artificial intelligence (AI), particularly deep learning, has recently shown promise in enhancing diagnostic precision, standardizing interpretation, and improving efficiency. This systematic review evaluates the diagnostic performance, methodological quality, and clinical applicability of AI-enhanced CBCT for CRS.

A systematic search of PubMed, Embase, Cochrane Library, IEEE Xplore, and Web of Science was conducted from inception through March 2025. Studies were included if they applied AI to CBCT for CRS diagnosis or sinonasal evaluation and reported quantitative diagnostic outcomes. Two reviewers independently screened, extracted data, and appraised study quality using QUADAS-2. Certainty of evidence was graded using GRADE.

A total of 24 studies met the inclusion criteria. AI approaches encompassed convolutional neural networks (CNNs), U-Net architectures, Mask R-CNN, and hybrid deep learning frameworks. Applications included maxillary sinusitis detection (seven studies), mucosal thickening quantification (six studies), anatomical variant detection (five studies), CRS classification (four studies), and fungal sinusitis detection (two studies). Sensitivity ranged from 77.8% to 96.7%, specificity from 78.9% to 94.2%, and AUC values from 0.83 to 0.98. Inter-reviewer agreement for study selection and extraction was excellent (κ = 0.78-0.84). QUADAS-2 indicated moderate overall quality; GRADE certainty was moderate for sinusitis detection and anatomical variants, but low to very low for other outcomes due to retrospective designs, heterogeneity, and limited external validation.

AI-enhanced CBCT demonstrates high diagnostic accuracy for CRS, particularly for maxillary sinusitis and mucosal thickening. However, current evidence is constrained by methodological limitations and a lack of multicenter prospective validation. Standardized protocols, integration of patient-centered outcomes, and economic evaluations are needed before widespread clinical implementation.