Compared to state-of-the-art iterative reconstruction, DLIR allows radiation dose for CTPA to be reduced by an additional 41% with non-inferior image quality.
Key Findings
Results
In phantom testing, DLIR-H allowed radiation dose to be reduced by up to 71% with equivalent or higher SNR compared to standard-dose examinations reconstructed with ASiR-V 90%.
Medium and large phantoms were used to simulate different body sizes.
The comparator reconstruction was ASiR-V 90%, representing state-of-the-art adaptive statistical iterative reconstruction.
DLIR-H (high-strength deep learning-based image reconstruction) was the experimental reconstruction algorithm.
SNR was equivalent or higher at up to 71% dose reduction in phantom conditions.
Results
In the patient cohort, radiation dose was reduced by 41% after transitioning to the DLIR-based protocol.
Median dose length product (DLP) decreased from 116 to 68 mGy*cm (p < 0.001).
Effective dose decreased from 1.69 to 0.99 mSv (p < 0.001).
The study included 307 consecutive patients: 152 examined before and 155 after the protocol change.
The change was a clinically driven protocol transition, analyzed retrospectively.
Results
Objective image quality as measured by SNR was superior for the central pulmonary artery and non-inferior for the segmental pulmonary arteries with the reduced-dose DLIR protocol.
Median SNR for the central pulmonary artery was 13.6 (standard protocol) vs. 22.3 (modified protocol), indicating superiority for the modified protocol.
Median SNR for the segmental pulmonary arteries was 16.4 (standard protocol) vs. 16.8 (modified protocol), meeting the non-inferiority criterion.
The non-inferiority margin was pre-specified as a less than 5% difference in image quality parameters.
The standard protocol used ASiR-V 90% reconstruction; the modified protocol used DLIR.
Results
Subjective image quality averaged over both readers was superior with the modified reduced-dose DLIR protocol compared to the standard-dose iterative reconstruction protocol.
Subjective image quality was rated by two radiologists.
The rating was averaged across both readers.
The modified protocol (reduced dose + DLIR) was rated superior, not merely non-inferior, in subjective assessment.
This finding was consistent despite the 41% reduction in radiation dose.
Methods
The study design used a retrospective analysis of consecutive patients undergoing a clinically driven protocol change, with a pre-specified non-inferiority margin.
307 consecutive patients were included: 152 before and 155 after the protocol change.
The non-inferiority margin was pre-specified as a less than 5% difference in image quality parameters.
Both objective (SNR quantification) and subjective (radiologist rating) image quality measures were assessed.
A phantom study was conducted prior to the clinical study to estimate achievable dose reduction.
What This Means
This research suggests that a newer type of computer-based image processing called deep learning image reconstruction (DLIR) can significantly reduce the radiation dose patients receive during CT pulmonary angiography (CTPA) — a common scan used to detect blood clots in the lungs — without sacrificing image quality. In laboratory phantom testing simulating different body sizes, the researchers found that DLIR allowed radiation doses to be cut by up to 71% while maintaining or improving image sharpness compared to the previous standard method (iterative reconstruction). When they applied this approach clinically, 307 real patients were studied before and after the protocol change, and the radiation dose fell by 41% — from a median effective dose of 1.69 millisieverts to 0.99 millisieverts.
Despite delivering substantially less radiation, the images produced with the new protocol were at least as good — and in some measures better — than those from the old higher-dose approach. Objective measurements of image clarity (signal-to-noise ratio) in the main pulmonary arteries were actually higher with the new protocol, and two radiologists independently rated the overall image quality as superior with the reduced-dose DLIR images. The pre-specified threshold for declaring non-inferiority (less than 5% difference in quality measures) was met or exceeded across all key metrics.
This research suggests that hospitals using modern CT scanners equipped with deep learning reconstruction software could meaningfully lower the radiation exposure associated with routine lung clot screening while maintaining — or even improving — diagnostic image quality. Given that CTPA is one of the more commonly performed CT examinations and radiation exposure carries cumulative health risks, a nearly 41% dose reduction in clinical practice represents a potentially important benefit for patients, particularly those who require repeated imaging.