Evaluation of a deep learning magnetic resonance imaging reconstruction method for synthetic computed tomography generation in prostate radiotherapy

Abstract

Background and purpose: In magnetic resonance imaging (MRI) only radiotherapy computed tomography (CT) is excluded. The method relies entirely on synthetic CT images generated from MRI. This study evaluates the compatibility of a commercial synthetic CT (sCT) with an accelerated commercial deep learning reconstruction (DLR) in MRI-only prostate radiotherapy.

Materials and methods: For a group of 24 patients (cohort 1) the effects of DLR were studied in isolation. MRI data were reconstructed conventionally and with DLR from identical k-space data, and sCTs were generated for both reconstructions. The sCT quality, Hounsfield Unit (HU) and dosimetric impact were investigated. In another group of 15 patients (cohort 2) effects on sCT generation using accelerated MRI acquisition (40 % time reduction) reconstructed with DLR were investigated.

Results: sCT images from both cohorts, generated from DLR MRI data, were of clinically expected image quality. The mean dose differences for targets and organs at risks in cohort 1 were <0.06 Gy, corresponding to a 0.1 % prescribed dose difference. Similar dose differences were observed in cohort 2. Gamma pass rates for cohort 1 were 100 % for criteria 3 %/3mm, 2 %/2mm and 1 %/1mm for all dose levels. Mean error and mean absolute error inside the body, between sCTs, averaged over all cohort 1 subjects, were -1.1 ± 0.6 [-2.4 0.2] and 2.9 ± 0.4 [2.3 3.9] HU, respectively.

Conclusions: DLR was suitable for sCT generation with clinically negligible differences in HU and calculated dose compared to the conventional MRI reconstruction method. For sCT generation DLR enables scan time reduction, without compromised sCT quality.

Keywords: Deep learning reconstruction; MRI; Prostate cancer; Radiotherapy; Synthetic CT.

Fig. 1

A. Data workflow for the patients in cohort 1 using two parallel image reconstruction pipelines on the same k-space data, one with conventional reconstruction (top row in A) and one with the GE HealthCare AIR Recon DL reconstruction (bottom row in A). B. Data workflow for the patients in cohort 2 using two image reconstruction pipelines, each one with unique k-space data, one with GE HealthCare AIR Recon DL reconstruction (clinical protocol, one average acquired, top row in B) and one with the conventional reconstruction (two averages acquired, bottom row in B). Images presented in B are symbolic representations and might not represent actual patient data.

Fig. 2

The figures represent differences between MRI using conventional reconstruction and DLR and their corresponding sCT in cohort 1. The left-hand side originates from the patient with the largest absolute value of body volume difference (27.2 cm³), and the right-hand side originates from the patient with the smallest absolute value of body volume difference (3.5 cm³). The images represent the following: a) Difference image between MR volumes with conventional and DLR reconstruction. Notice the homogeneous content of noise and Gibbs ringing artefacts. Also notice that the image differences do not geometrically correspond to the image differences seen for the sCT. b) Absolute value of image in a) with a wider display window for improved visualization. c) Difference image of the two corresponding sCT volumes for a window width of −15 to 15 Hounsfield units (HU). d) Absolute value of image in c) with a wider window of 0 to 150 HU for improved visualization of peripheral differences. The different window width and level enhance visibility, with higher signal values than the defined scale adopting the color of the highest defined window value. Notice the residual difference in left and right side of the patient for the patient with the largest difference and the small deviation on the patient left side for the patient with the smallest absolute body volume difference (HU scale saturated, true difference value was larger than 150).

Fig. 3

A. Boxplot of mean total dose differences for clinical target volume (CTV) and planning target volume (PTV) and organs at risk (sCT_orig-sCT_DL) in cohort 1 (n = 23). Prescribed dose was 42.7 Gy to planning target volume (PTV). sCT_orig defined the image geometry for the created treatment plan which was copied to sCT_DL and recalculated. B. Boxplot of mean total dose differences for clinical target volume (CTV) and planning target volume (PTV) and organs at risk (sCT_Acc_DL_1nex -sCT_Acc_orig_2nex) in cohort 2 (n = 15). Prescribed dose was 42.7 Gy to planning target volume (PTV). sCT_Acc_DL_1nex defined the image geometry for the created treatment plan which was copied to sCT_Acc_orig_2nex and recalculated.