Evaluation of a cycle-generative adversarial network-based cone-beam CT to synthetic CT conversion algorithm for adaptive radiation therapy

Abstract

Image-guided radiation therapy could benefit from implementing adaptive radiation therapy (ART) techniques. A cycle-generative adversarial network (cycle-GAN)-based cone-beam computed tomography (CBCT)-to-synthetic CT (sCT) conversion algorithm was evaluated regarding image quality, image segmentation and dosimetric accuracy for head and neck (H&N), thoracic and pelvic body regions. Methods: Using a cycle-GAN, three body site-specific models were priorly trained with independent paired CT and CBCT datasets of a kV imaging system (XVI, Elekta). sCT were generated based on first-fraction CBCT for 15 patients of each body region. Mean errors (ME) and mean absolute errors (MAE) were analyzed for the sCT. On the sCT, manually delineated structures were compared to deformed structures from the planning CT (pCT) and evaluated with standard segmentation metrics. Treatment plans were recalculated on sCT. A comparison of clinically relevant dose-volume parameters (D 98 , D 50 and D 2 of the target volume) and 3D-gamma (3%/3mm) analysis were performed. Results: The mean ME and MAE were 1.4, 29.6, 5.4 Hounsfield units (HU) and 77.2, 94.2, 41.8 HU for H&N, thoracic and pelvic region, respectively. Dice similarity coefficients varied between 66.7 ± 8.3% (seminal vesicles) and 94.9 ± 2.0% (lungs). Maximum mean surface distances were 6.3 mm (heart), followed by 3.5 mm (brainstem). The mean dosimetric differences of the target volumes did not exceed 1.7%. Mean 3D gamma pass rates greater than 97.8% were achieved in all cases. Conclusions: The presented method generates sCT images with a quality close to pCT and yielded clinically acceptable dosimetric deviations. Thus, an important prerequisite towards clinical implementation of CBCTbased ART is fulfilled.

Geometric and Dosimetric Evaluation of Deep Learning-Based Automatic Delineation on CBCT-Synthesized CT and Planning CT for Breast Cancer Adaptive Radiotherapy: A Multi-Institutional Study

“…In our study, the synthetic CT image quality analysis of the thorax yielded an ME/MAE of 8.46/71.58 HU, and our results were better than those reported by Eckl et al. ( 36 ), with 29.6/94.2 HU. The SNU in synthetic CT was close to the SNU in planning CT.…”

Deep learning based synthetic‐CT generation in radiotherapy and PET: A review

“…124 Some of the works (5 out of 15) focused only on improving CBCT image quality for better IGRT. 83,[124][125][126][127] The remaining 10 proved the validity of the transformation with dosimetric studies for photons, 66,71,101,[128][129][130][131] protons, 120 and for both photons and protons. 85,132,133 Only three studies investigated unpaired training 84,128,133 ; in 11 cases, paired training was implemented by matching the CBCT and ground truth CT by rigid or deformable registration.…”

Dosimetric benefits of daily treatment plan adaptation for prostate cancer stereotactic body radiotherapy

“…This study revealed the deficiencies of an IGRT approach and, more importantly, offered several adaptation techniques focusing on available planning time (ART1) and increasing target coverage (ART2) or reducing OAR dosage (ART3). Although there has been a long-time debate about the actual benefit of adaptive strategies with respect to their costs [ 57 – 59 ], promising results for daily treatment modifications have been obtained by fast and automated tools for image correction [ 10 , 60 ], segmentation [ 61 , 62 ] and treatment planning [ 63 ]. As previously reported, converting CBCT into sCT and subsequent image segmentation including manual correction of the generated structures takes up to 30 s, 30 s and 5.2 ± 1.6 min [ 10 ].…”

“…Although there has been a long-time debate about the actual benefit of adaptive strategies with respect to their costs [ 57 – 59 ], promising results for daily treatment modifications have been obtained by fast and automated tools for image correction [ 10 , 60 ], segmentation [ 61 , 62 ] and treatment planning [ 63 ]. As previously reported, converting CBCT into sCT and subsequent image segmentation including manual correction of the generated structures takes up to 30 s, 30 s and 5.2 ± 1.6 min [ 10 ]. Together with the presented re-optimization approaches an end-to-end adaptive workflow can eventually become feasible within a reasonable timeframe of minimum 0.5 min + 0.5 min + 5.2 min + 2.6 min (ART1 approach) = 8.8 min, being in line with reported adaptation times of 10 min including online plan QA [ 64 ].…”

“…With the help of an individually trained cycle-generative adversarial network (cycle-GAN) based pelvis sCT model in the research software ADMIRE (Elekta AB) all CBCT were converted into sCT, taking on average 30 s per dataset. Specifications of the GAN framework and the pelvis sCT model can be taken from previously published literature [ 10 ].…”