Quantitative Assessment of Diagnostic Image Quality Adequacy In Neck CT: Evidence for Physics-Informed Protocol Redesign
Abstract
Purpose
Neck CT protocols often present challenges for objective lesion detectability tasks. We systematically assessed lesion detectability in clinical scans to identify quality gaps and dose utilization inefficiencies.
Methods
We analyzed 42 clinical neck CT examinations using a custom task-based automated image quality (IQ) assessment tool (AnatomIQ). We drew circular ROIs around lesions with the tool automatically extracting Hounsfield Unit (HU) values from the ROI center and sampled multiple background candidate regions with similar HU values in the surrounding area. Evaluation metrics included CNR, detectability index (d'=CNR×√Area), Weber contrast, and the dose-normalized figure of merit (CNR2/dose). Lesions were classified using the Rose criterion-based threshold for IQ-adequacy (i.e., adequate: CNR ≥1.0). CTDIvol was used to estimate Size-specific dose estimates (SSDE).
Results
Systematic assessment demonstrated 45% (19/42) inadequacy (CNR0.80), while d’ strongly correlated with both contrast (r=0.60, p0.60), with IQ-inadequate cases receiving a higher mean dose (10.3±4.6 mGy) than IQ-adequate cases (9.7±3.1 mGy, p=0.90). Notably, 78% of IQ-inadequate cases occurred at doses below the AD. However, IQ-inadequacy was also observed in high dose cases up to 19.76 mGy (CNR=0.13) demonstrating that elevated doses alone might not compensate for insufficient lesion contrast.
Conclusion
Automated quantitative assessment identified lesion detectability deficiencies in clinical neck CT independent of radiation dose. These findings emphasize the need for contrast and reconstruction driven protocol redesign guided by task-based IQ metrics rather than traditional dose benchmarks alone.