Dosimetric assessment of patient dose calculation on a deep learning‐based synthesized computed tomography image for adaptive radiotherapy

Abstract

Purpose Dose computation using cone beam computed tomography (CBCT) images is inaccurate for the purpose of adaptive treatment planning. The main goal of this study is to assess the dosimetric accuracy of synthetic computed tomography (CT)‐based calculation for adaptive planning in the upper abdominal region. We hypothesized that deep learning‐based synthetically generated CT images will produce comparable results to a deformed CT (CTdef) in terms of dose calculation, while displaying a more accurate representation of the daily anatomy and therefore superior dosimetric accuracy. Methods We have implemented a cycle‐consistent generative adversarial networks (CycleGANs) architecture to synthesize CT images from the daily acquired CBCT image with minimal error. CBCT and CT images from 17 liver stereotactic body radiation therapy (SBRT) patients were used to train, test, and validate the algorithm. Results The synthetically generated images showed increased signal‐to‐noise ratio, contrast resolution, and reduced root mean square error, mean absolute error, noise, and artifact severity. Superior edge matching, sharpness, and preservation of anatomical structures from the CBCT images were observed for the synthetic images when compared to the CTdef registration method. Three verification plans (CBCT, CTdef, and synthetic) were created from the original treatment plan and dose volume histogram (DVH) statistics were calculated. The synthetic‐based calculation shows comparatively similar results to the CTdef‐based calculation with a maximum mean deviation of 1.5%. Conclusions Our findings show that CycleGANs can produce reliable synthetic images for the adaptive delivery framework. Dose calculations can be performed on synthetic images with minimal error. Additionally, enhanced image quality should translate into better daily alignment, increasing treatment delivery accuracy.

Deep learning methods for enhancing cone‐beam CT image quality toward adaptive radiation therapy: A systematic review

“…Segmentation accuracy can be thought of as a proxy measure for image quality and a useful metric for assessing the extent of anatomical preservation during image synthesis. To compare the preservation of anatomical structures, Lemus et al 52 . manually contoured CBCT, sCT, and DIR CT images of abdominal patients, with the CBCT used as reference.…”

“…GAN‐based studies all used DIR given their susceptibility of generating false anatomies. Cycle‐GAN, originally designed for unpaired data, was most commonly coupled with RR, for both paired 19,35,38,40,45,52 and unpaired approaches 36,37,42,46,49 . Alternatively, Liu et al 41 .…”

“…GANbased studies all used DIR given their susceptibility of generating false anatomies. Cycle-GAN, originally designed for unpaired data, was most commonly coupled with RR, for both paired 19,35,38,40,45,52 and unpaired approaches 36,37,42,46,49 .Alternatively,Liu et al 41 and Qiu et al 48 applied DIR to their training set for better anatomical correspondence, whereas Liang et al 39 and Dong et al 50 did not apply any registration to their training data, only resampling to the same grid. Liu et al 41 was the only cycle-GAN study to investigate the impact of DIR and RR preprocessing on the same dataset and found that DIR produced slightly better HU agreement (56.89 ± 13.84 vs. 58.45 ± 13.88 HU), with substantially less noise and sharper organ boundaries upon visual inspection.…”

“…Extensions to the cycle-GAN global loss included identity loss, 19,39,49,50 gradient loss, 35,36,44,[46][47][48]52 synthetic loss, 35,38,40,41,44,48 L1.5 loss, 35 histogram matching loss, 48 idempotent loss, 36,52 air loss, 36 total variation loss, 36,52 feature matching loss, 18 and perceptual loss. 48 The identity loss 19,39,49,50 aids network stability and generator accuracy by ensuring no additional effect occurs to real images when they are input into generators tasked to output images in the same domain.…”

“…The preferred method for implementing CBCT-based dose monitoring protocols has relied on DIR of the planning CT to the CBCT. [121][122][123] With the emergence of sCT generation, the authors have demonstrated better agreement in terms of MAE 39,57,61 and anatomical accuracy 52 over using deformed planning CTs, whilst maintaining comparable dose accuracy for proton and photon plans. 42,52,61 Hence, the use of sCT for dose monitoring will enhance the accuracy of research conclusions and result in improved decision-making for both online and offline ART.…”