Bridging TG‐116 and TG‐232: A retrospective multi‐vendor evaluation of deviation index re‐centering and process control in digital radiography
Ezzat AbuAzzah, Faisal Alrehily, Moawia Gameraddin, Sultan Alshoabi, Walaa Alsharif, Kamal Dahan Alsultan, Abdullah Alshamrani, Khaled Al‐Radadi, Ramy Badawy, Samer Al‐Jabri, Turki Al‐Harbi, Nawaf Al‐HazmeAbstract
Background
The deviation index (DI) is a vendor‐neutral indicator of detector exposure relative to the configured target exposure index (TEI). Because TEI is a protocol value rather than a calibrated physical quantity, poorly aligned TEI settings can create systematic DI offsets, obscure protocol problems, and prevent meaningful statistical process control.
Objective
This retrospective multi‐vendor study evaluated whether routine DICOM metadata can identify biased DI distributions, separate fixed TEI‐setting offsets from protocol‐adherence variation, estimate candidate TEI updates, and support TG‐232‐aligned standard deviation‐based monitoring after DI re‐centering.
Methods
We retrospectively analyzed 38 916 digital radiography examinations performed from January 1, 2024 through August 27, 2024 using three vendor platforms at a tertiary hospital. DICOM metadata were benchmarked against published technical standards using a composite adherence index (CAI) for SID, kVp, mAs, grid use, and AEC use. A secondary gradient‐boosting/SHAP analysis was retained as a non‐causal variance discovery to rank the metadata patterns used to predict DI. Candidate TEI updates were estimated within the vendor‐region‐projection groups by algebraically re‐centering the observed mean DI ( to 0.0. Robustness was assessed using sensitivity analyses restricted by automatic exposure control, beam energy, and patient habitus proxy measures.
Results
The baseline = −3.1 (SD 3.7), and only 11.3% of examinations were within |DI| ≤ 1. Forty of 45 vendor–region–projection groups (88.9%) had an absolute group greater than 1.0. Separately, the mean CAI was 0.176. The secondary SHAP analysis ranked the configured TEI, manufacturer, SID, exposure, and kVp as high‐contribution predictors of DI; these findings were interpreted as predictive attribution, not causal evidence. DI re‐centering reduced the overall variance by 42.9% (13.5 to 7.7), increased the proportion of examinations within |DI| ≤ 1 to 34.3%, and reduced severe deviations (|DI| > 3) from 60.3% to 23.1%.
Conclusion
In this retrospective, multi‐vendor study, the TEI setting review was operationalized as a structured, data‐driven responsibility of a clinical site. By verifying protocol adherence, equipment performance, image quality, and dose indicators before centering the , departments can support the conversion of the DI from a biased compliance signal into a practical process‐control metric for exposure governance.