CT Effective Dose in Critical Patients: Comparison between Deep Learning Image Reconstruction (DLIR), Filtered Back Projection (FBP) and Iterative Algorithms

Background: Since the introduction of computed tomography (CT) imaging in 1971, different algorithms have been developed to reconstruct images starting from the countless attenuation values acquired by the rotating detector. Such algorithms significantly impact radiation exposure, since a reconstruction that improves image quality at a given dose translates into the possibility of achieving the same baseline quality required for diagnosis at a lower dose. At the time of writing, three main categories of reconstruction methods have been developed: filtered back projection (FBP) in 1972, iterative reconstruction (IR) in 2008 and deep learning image reconstruction (DLIR) in 2018. DLIR has been proven effective in reducing radiation dose compared with the previous techniques in both phantom studies and studies where patients were randomly assigned to be examined using either one method or the other. However, studies regarding how far this reduction may have an impact on the same individual in real clinical settings are still limited. Objective: To evaluate if there is radiation dose reduction and image quality im- provement in the same acute patient when using DLIR in comparison to FBP and IR in daily clinical practice. Methods: This retrospective study included 83 critical care patients who under- went CT imaging of the same anatomical region multiple times within a period of 30 days using both DLIR (TrueFidelity) and FBP or IR (AIDR3D and ADMIRE). Regions included were chest, abdomen and trunk (chest + abdomen). All examinations were performed using automatic exposure control (AEC) which modulates the tube current and hence radiation exposure according to the algorithm applied. Radiation dose was assessed using CT dose index volume (CTDI volume), dose-length product (DLP) and Effective Dose. For the quantification of image quality, Noise and Signal to Noise Ratio (SNR) were used. All parameters were compared across the different reconstruction methods for each patient using both parametric and non-parametric testing. In cases of contrast-enhanced CT (CECT), all parameters were retrieved for every acquisition phase (direct, arterial, venous or delayed) as well as for their total value as stated in the patient protocol. Results: Our analysis suggested that DLIR majorly and consistently outperformed the traditional techniques with regards to image quality, and to a lesser extent, it correlated with dose reduction. Specifically on average, the total values for FBP were 24.67 ± 61.01 mGy for CTDI, 1350.62 ± 1191.68 mGy * cm for DLP, 20.19 ± 17.91 mSv for Effective Dose, image noise was 28.85 ± 32.77 HU, and SNR was 3.99 ± 1.23 HU. Those values were improved in DLIR: 9.56 ± 5.86 mGy for CTDI, 1085.33 ± 626.30 mGy * cm for DLP, 16.13 ± 9.55 mSv for Effective Dose, image noise was 8.45 ± 3.24 HU, and SNR was 11.53 ± 9.28 HU. Regarding IR, the total dose was not found to be affected by the use of DLIR, but image quality was improved. The mean values for examinations with IR were 14.00 ± 12.46 mGy for CTDI, 1235.53 ± 873.67 mGy * cm for DLP, 18.45 ± 13.16 mSv for Effective Dose, image noise was 14.85 ± 2.73 HU, and SNR was 4.84 ± 2.74 HU. Conclusion: According to our study, DLIR provides benefits in terms of dose and image quality over the traditional FBP. It also outperforms IR methods for image quality, but not for dose. Further research is needed to see if those improvements translate into safer imaging practices, higher diagnostic confidence, and ultimately better patient care. Keypoints: • Compared to FBP, DLIR both reduces radiation dose and improves image quality; • Compared to IR, DLIR doesn’t necessarily reduce radiation dose, but it improves image quality