Deep Learning and Contrast Enhancement Boost Optimize Low-Contrast CTPA Image Quality

Combining deep learning reconstruction with contrast enhancement boost synergistically optimizes low-contrast-dose computed tomography pulmonary angiography, enhancing image quality beyond that achieved with standard contrast doses. High-quality CT pulmonary angiography remains vital for the accurate diagnosis of pulmonary embolism, yet standard protocols require substantial contrast media administration. In a prospective randomized trial evaluating this combined post-processing approach, researchers found that integrating deep learning reconstruction with contrast enhancement boost permitted a 40 percent reduction in total iodine load and a 25 percent reduction in contrast injection rate while simultaneously improving vascular imaging quality over conventional standard-dose examinations.

The prospective randomized investigation was conducted between September and December 2023, enrolling a total of 106 patients allocated equally between two imaging regimens. Group A received a low-contrast-dose protocol, whereas Group B received a standard-contrast-dose protocol. Each cohort comprised 53 patients with closely matched demographic profiles: Group A had a mean age of 59 ± 15 years and included 27 women, while Group B had a mean age of 58 ± 18 years and included 26 women. The low-contrast-dose protocol administered to Group A utilized 30 mL of contrast agent at an injection flow rate of 3 mL/s. In contrast, the standard-contrast-dose protocol in Group B administered 50 mL of contrast medium at a higher injection flow rate of 4 mL/s.

To systematically isolate the effects of the reconstruction algorithms and post-processing tools, investigators established a 2 × 2 factorial design generating four distinct image datasets from each patient in the low-contrast-dose group. Dataset A1 applied hybrid iterative reconstruction alone to represent the baseline low-dose condition. Dataset A2 applied deep learning reconstruction using the Advanced Intelligent Clear-IQ Engine (AiCE). Dataset A3 combined hybrid iterative reconstruction with contrast enhancement boost (CE-boost). Dataset A4 integrated both deep learning reconstruction and CE-boost into a unified post-processing pipeline. In comparison, imaging datasets for all patients in standard-contrast-dose Group B were generated using standard hybrid iterative reconstruction.

The research team established both quantitative and qualitative diagnostic endpoints to compare the reconstructed image series across the study cohorts. The quantitative endpoints assessed CT attenuation, objective image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR). The qualitative endpoints assessed overall image quality along with subsegmental pulmonary artery visibility. To account for repeated measurements and factorial structures, the researchers evaluated these parameters using linear mixed models and aligned rank transform (ART) analysis of variance (ANOVA), incorporating Holm correction across the statistical tests to manage multiple comparisons.

The quantitative analysis revealed distinct individual performance profiles for deep learning reconstruction and contrast enhancement boost. Deep learning reconstruction alone significantly reduced image noise and produced statistically significant increases in both SNR and CNR across all evaluations, with all Holm-corrected p-values falling below 0.001. In qualitative grading, deep learning reconstruction significantly improved both overall image quality and subsegmental artery visibility. Conversely, contrast enhancement boost significantly increased CT attenuation, SNR, and CNR (all p < 0.001) without altering measured image noise. Qualitatively, contrast enhancement boost significantly improved subsegmental artery visibility alone, without producing a significant enhancement in overall image quality.

Crucially, the evaluation revealed a statistically significant interaction between deep learning reconstruction and contrast enhancement boost for CNR across all evaluated regions of interest (all p < 0.05), demonstrating a synergistic effect between the two post-processing methods. When the combined protocol in dataset A4 was compared against the alternative low-dose reconstructions, it achieved the highest CNR values and the best subsegmental artery visibility among all four Group A datasets. Furthermore, dataset A4 demonstrated image quality that significantly surpassed both the baseline low-contrast-dose dataset (A1) and the standard-contrast-dose protocol in Group B, with all comparisons meeting Bonferroni-corrected significance thresholds of p < 0.0083 and p < 0.025.

The authors concluded that pairing deep learning reconstruction with contrast enhancement boost synergistically optimizes low-contrast-dose CT pulmonary angiography. By simultaneously enabling a 40 percent reduction in iodine load and a 25 percent reduction in injection rate while delivering image quality superior to that of standard-contrast-dose CTPA, the combined protocol directly addresses technical constraints in contrast administration. Consequently, the researchers noted that this optimized low-contrast-dose protocol may offer a viable clinical imaging strategy for patients who require strict limitations on contrast media exposure during diagnostic evaluation for suspected pulmonary embolism.

Subject of Research: Synergistic optimization of low-contrast-dose CT pulmonary angiography using deep learning reconstruction and contrast enhancement boost

Article Title: Deep learning reconstruction combined with contrast enhancement boost improves image quality of low-contrast-dose ctpa: a prospective randomized study

Article References: Xie, D., Lin, J., Huang, M., Lai, Z., Yang, X., Xu, R., Cai, H., & Ma, H. (2026). Deep learning reconstruction combined with contrast enhancement boost improves image quality of low-contrast-dose ctpa: a prospective randomized study. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02998-6

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02998-6

Keywords: computed tomography pulmonary angiography, CTPA, deep learning reconstruction, contrast enhancement boost, CE-boost, low-contrast-dose CT, pulmonary embolism diagnosis, hybrid iterative reconstruction, AiCE, contrast-to-noise ratio, signal-to-noise ratio, CT attenuation, image noise reduction, subsegmental artery visibility, prospective randomized trial, 2×2 factorial design, iodine load reduction