Enhanced 3D Dose Prediction for Hypofractionated SRS (Gamma Knife Radiosurgery) in Brain Tumor Using Cascaded-Deep-Supervised Convolutional Neural Network-Reference-Cited by-同舟云学术

Enhanced 3D Dose Prediction for Hypofractionated SRS (Gamma Knife Radiosurgery) in Brain Tumor Using Cascaded-Deep-Supervised Convolutional Neural Network

Published:2024-01-30 Issue: Volume: Page:
ISSN:
Container-title:
language:
Short-container-title:

Author:

Li Nan¹^ORCID,Wang Jinyuan²,Fang Chunfeng³,Zhou Dongxue⁴,Liu Yaoying¹,Wang Yanping⁵,Zhang Chunsu⁴,Cao Lin⁴,Zhang Gaolong¹,Xu Shouping⁶^ORCID

Affiliation:

1. Beihang University

2. 301 Military Hospital: Chinese PLA General Hospital

3. Tsinghua University

4. HeBei YiZhou Proton Center

5. HeiBei YiZhou Proton Center

6. Chinese Academy of Medical Sciences and Peking Union Medical College

Abstract

Abstract Gamma Knife radiosurgery (GKRS) is a well-established radiation therapy (RT) technique for treating brain tumors. However, the planning process for GKRS is complex and time-consuming, heavily reliant on the expertise of medical physicists. Incorporating deep learning approaches for GKRS dose prediction can reduce this dependency, improving planning efficiency and homogeneity, streamlining clinical workflows, and reducing patient lagging times. Despite this, precise Gamma Knife plan dose distribution prediction using existing models remains a significant challenge. The complexity stems from the intricate nature of dose distributions, subtle contrasts in CT scans, and the interdependence of dosimetric metrics. In an effort to overcome these challenges, we have developed a "Cascaded-Deep-Supervised" Convolutional Neural Network (CDS-CNN) that employs a hybrid-weighted optimization scheme. Our innovative method incorporates multi-level deep supervision along with a strategic sequential multi-network training approach. It enables the extraction of intra-slice and inter-slice features, leading to more realistic dose predictions with additional contextual information. CDS-CNN was trained and evaluated using data from 105 brain cancer patients who underwent GKRS treatment, with 85 cases used for training and 20 for testing. Quantitative evaluations and statistical analyses demonstrated high consistency between the predicted dose distributions and the reference doses from the treatment planning system (TPS). The 3D overall gamma passing rates (GPRs) reached 97.15%±1.36% (3mm/3%, 10% threshold). When evaluated using the more stringent criteria of 2mm/3%, 10% threshold, the overall GPRs still achieved 96.33%±1.08%. Furthermore, the average target coverage (TC) was 98.33%±1.16%, dose selectivity (DS) was 0.57±0.10, gradient index (GI) was 2.69±0.30, and homogeneity index (HI) was 1.79±0.09. The experimental results showed that the proposed CDS-CNN outperformed other models in predicting GKRS dose distributions, with the prediction being the closest to the TPS dose.

Publisher

Research Square Platform LLC

Reference42 articles.

1. Ganz, Jeremy C. Development of the Gamma Knife. Springer Vienna, 10.1007/978-3-7091-3313-2_6, english, 1993, 2023-01-09, Ganz, Jeremy C., Gamma Knife Surgery, 55--65, http://link.springer.com/10.1007/978-3-7091-3313-2_6, 978-3-211-82476-4 978-3-7091-3313-2, Vienna

2. Dong, Peng and P érez-And újar, Ang élica and Pinnaduwage, Dilini and Braunstein, Steve and Theodosopoulos, Philip and {McDermott}, Michael and Sneed, Penny and Ma, Lijun Dosimetric characterization of hypofractionated Gamma Knife radiosurgery of large or complex brain tumors versus linear accelerator –based treatments. 125: 97--103 https://doi.org/10.3171/2016.7.GKS16881, Dong 等 - 2016 - Dosimetric characterization of hypofractionated Ga.pdf:C\:\\Users\\JIMMY.G3.3579\\Zotero\\storage\\EYJ97VFP\\Dong 等 - 2016 - Dosimetric characterization of hypofractionated Ga.pdf:application/pdf, english, 2016-12, {JNS}, Journal of Neurosurgery, Supplement\_1, Objective Noninvasive Gamma Knife ({GK}) platforms, such as the relocatable frame and on-board imaging, have enabled hypofractionated {GK} radiosurgery of large or complex brain lesions. This study aimed to characterize the dosimetric quality of such treatments against linear accelerator –based delivery systems that include the {CyberKnife} ({CK}) and volumetric modulated arc therapy ({VMAT}). Methods Ten patients treated with {VMAT} at the authors ’ institution for large brain tumors ({\textgreater} 3 cm in maximum diameter) were selected for the study. The median prescription dose was 25 Gy (range 20 –30 Gy) in 5 fractions. The median planning target volume ({PTV}) was 9.57 cm3 (range 1.94 –24.81 cm3). Treatment planning was performed using Eclipse External Beam Planning V11 for {VMAT} on the Varian {TrueBeam} system, Multiplan V4.5 for the {CyberKnife} {VSI} System, and Leksell {GammaPlan} V10.2 for the Gamma Knife Perfexion system. The percentage of the {PTV} receiving at least the prescription dose was normalized to be identical across all platforms for individual cases. The prescription isodose value for the {PTV}, conformity index, Paddick gradient index, mean and maximum doses for organs at risk, and normal brain dose at variable isodose volumes ranging from the 5-Gy isodose volume (V5) to the 15-Gy isodose volume (V15) were compared for all of the cases. Results The mean Paddick gradient index was 2.6 ± 0.2, 3.2 ± 0.5, and 4.3 ± 1.0 for {GK}, {CK}, and {VMAT}, respectively (p {\textless} 0.002). The mean V15 was 7.5 ± 3.7 cm3 (range 1.53 –13.29 cm3), 9.8 ± 5.5 cm3 (range 2.07 –18.45 cm3), and 16.1 ± 10.6 cm3 (range 3.58 –36.53 cm3) for {GK}, {CK}, and {VMAT}, respectively (p ≤ 0.03, paired 2-tailed t-tests). However, the average conformity index was 1.18, 1.12, and 1.21 for {GK}, {CK}, and {VMAT}, respectively (p {\textgreater} 0.06). The average prescription isodose values were 52% (range 47% –69%), 60% (range 46% –68%), and 88% (range 70% –94%) for {GK}, {CK}, and {VMAT}, respectively, thus producing significant variations in dose hot spots among the 3 platforms. Furthermore, the mean V5 values for {GK} and {CK} were similar (p {\textgreater} 0.79) at 71.9 ± 36.2 cm3 and 73.3 ± 31.8 cm3, respectively, both of which were statistically lower (p {\textless} 0.01) than the mean V5 value of 124.6 ± 67.1 cm3 for {VMAT}. Conclusions Significantly better near-target normal brain sparing was noted for hypofractionated {GK} radiosurgery versus linear accelerator –based treatments. Such a result supports the use of a large number of isocenters or confocal beams for the benefit of normal tissue sparing in hypofractionated brain radiosurgery. http://thejns.org/doi/abs/10.3171/2016.7.{GKS}16881, https://thejns.org/view/journals/j-neurosurg/125/Supplement_1/article-p97.xml, 0022-3085, 1933-0693

3. Sheikh-Bagheri, Daryoush and Rogers, D. W. O. and Ross, Carl K. and Seuntjens, Jan P. Comparison of measured and Monte Carlo calculated dose distributions from the {NRC} linac. 27(10): 2256--2266 https://doi.org/10.1118/1.1290714, Sheikh-Bagheri 等 - 2000 - Comparison of measured and Monte Carlo calculated .pdf:C\:\\Users\\JIMMY.G3.3579\\Zotero\\storage\\M7SUERPA\\Sheikh-Bagheri 等 - 2000 - Comparison of measured and Monte Carlo calculated .pdf:application/pdf, english, 2000-10, Med. Phys., Medical Physics, http://doi.wiley.com/10.1118/1.1290714, 00942405

4. Kendall, Ethan and Algan, Ozer and Chen, Yong and Ahmad, Salahuddin Dosimetric comparison of {TMR}10 and convolution dose calculation algorithms in {GammaPlan} treatment planning system. 19(1): 93--97 https://doi.org/10.1017/S1460396919000347, Kendall 等 - 2020 - Dosimetric comparison of TMR10 and convolution dos.pdf:C\:\\Users\\JIMMY.G3.3579\\Zotero\\storage\\BK5YIRIJ\\Kendall 等 - 2020 - Dosimetric comparison of TMR10 and convolution dos.pdf:application/pdf, english, 2020-03, J Radiother Pract, Journal of Radiotherapy in Practice, Aims: In this article, our goal is to compare the {TMR}10 and convolution dose calculation algorithm in {GammaPlan} used in stereotactic radiosurgery ({SRS}) treatments with Gamma Knife and to assess if the algorithms produce clinically significant differences., https://www.cambridge.org/core/product/identifier/S1460396919000347/type/journal_article, 1460-3969, 1467-1131

5. Yu, Cheng and Chen, Joseph C.T and Apuzzo, Michael L.J and O ’Day, Steven and Giannotta, Steven L and Weber, Jeffrey S and Petrovich, Zbigniew Metastatic melanoma to the brain: prognostic factors after gamma knife radiosurgery. 52(5): 1277--1287 https://doi.org/10.1016/S0360-3016(01)02772-9, Yu 等 - 2002 - Metastatic melanoma to the brain prognostic facto.pdf:C\:\\Users\\JIMMY.G3.3579\\Zotero\\storage\\B77X4A2E\\Yu 等 - 2002 - Metastatic melanoma to the brain prognostic facto.pdf:application/pdf, english, 2002-04, International Journal of Radiation Oncology*Biology*Physics, International Journal of Radiation Oncology*Biology*Physics, Purpose: To identify important prognostic factors predictive of survival and tumor control in patients with metastatic melanoma to the brain who underwent gamma knife radiosurgery. Methods and Materials: A total of 122 consecutive patients with 332 intracranial melanoma metastases underwent gamma knife radiosurgery over a 5-year period. Of these, 39 (32%) also received whole-brain irradiation ({WBI}). The median tumor volume was 0.8 cm3 (range: 0.02 –30.20 cm3), and the median prescribed dose was 20 Gy (range: 14 –24 Gy). Median follow-up was 6.8 months. Univariate and multivariate analyses of survival and freedom from progression were performed using the following parameters: status of systemic disease, intracranial tumor volume, number of lesions, tumor location, Karnofsky performance status, gender, age, and {WBI}. Results: Overall median survival was 7.0 months from time of radiosurgery and 9.1 months from the onset of brain metastasis. In multivariate analysis, improved survival was noted in patients with total intracranial tumor volume {\textless}3 cm3 (p ؍ 0.003) and inactive systemic disease (p ؍ 0.0065), whereas other parameters studied were of lesser importance (tumor location, p ؍ 0.056, and Karnofsky performance status, p ؍ 0.086), or of no signi ﬁcance (number of lesions, {WBI}, age, and gender). Freedom from subsequent brain metastasis depended on intracranial tumor volume (p ؍ 0.0018) and status of systemic disease (p ؍ 0.034). Conclusions: Stereotactic radiosurgery is an effective treatment modality for patients with intracranial metastatic melanoma. Tumor volume and status of systemic disease are good independent predictors of survival and freedom from tumor progression. © 2002 Elsevier Science Inc., Metastatic melanoma to the brain, https://linkinghub.elsevier.com/retrieve/pii/S0360301601027729, 03603016