Deep Learning for Automated Contouring of Primary Tumor Volumes by MRI for Nasopharyngeal Carcinoma

Lin, Li; Dou, Qi; Jin, Yueming; Zhou, Guan Qun; Tang, Yi; Chen, Wei Lin; Su, Bao An; Liu, Feng; Chang, Tao; Jiang, Ning; Li, Jun Yun; TANG, Ling; Xie, Chuan; Huang, Shao Min; Ma, Jun; Heng, Pheng‐Ann; Wee, Ji‐Hyang; Chua, Melvin L.K.; Chen, Hao; Sun, Ying

doi:10.1148/radiol.2019182012

Cited by 271 publications

(223 citation statements)

References 29 publications

Supporting

Mentioning

219

Contrasting

Unclassified

Order By: Relevance

“…However, some variations could be minimized when unmotivated personal preferences are removed. Recently, automated contouring of NPC GTV using machine learning yielded promising results (23). Artificial intelligence (AI)-based innovative tools are now expected to help reduce inter-observer and inter-institution variance on CTV delineation in the near future.…”

Section: Discussionmentioning

confidence: 99%

Variations of Clinical Target Volume Delineation for Primary Site of Nasopharyngeal Cancer Among Five Centers in China

Zhang

Yang

et al. 2020

Front. Oncol.

Self Cite

View full text Add to dashboard Cite

The purpose of this study is to investigate the current status of clinical target volume (CTV) delineation for primary site of nasopharyngeal cancer (NPC) among five large tertiary cancer centers in China. Materials and Methods: The simulation CT and MR images of a patient with T3N2M0 NPC were sent to the centers participating. Fourteen experienced physicians contoured the targets independently, and the outlined structures were compared. The consistency and differences among these 14 CTVs are discussed. Results: Two different CTV designs were used in the centers. "One-CTV" design defines one CTV with a dose of 60 Gy, whereas "two-CTV" design has a high-risk CTV with dose of 60 Gy and a medium risk CTV with dose of 54 Gy. We found that the coverage of prophylactic area is very consistent between these two designs. The variances on the coverage of some sites were also significant among physicians, including covering cavernous sinus at un-involved side, posterior space of styloid process, and caudal border on posterior pharyngeal wall. Conclusions: Standardization is the main requirement for personalization of care; our study shows that among the 14 physicians in the five centers the coverage of prophylactic areas is in excellent agreement. Two distinct strategies on CTV design are currently being used, and multiple controversies were found, suggesting further optimization of CTV for primary site of NPC is needed.

show abstract

Section: Discussionmentioning

confidence: 99%

Variations of Clinical Target Volume Delineation for Primary Site of Nasopharyngeal Cancer Among Five Centers in China

Zhang

Yang

et al. 2020

Front. Oncol.

Self Cite

View full text Add to dashboard Cite

show abstract

“…It is also probable that the bottleneck for this question lies in the quality of the ground truth segmentation than in the deep learning analysis as many teams got very close results in terms of dice score (table 6). Higher quality ground truth segmentation may have been achieved by keeping only segmentation of class 3 and 4 or asking 3 expert radiologists to decide by consensus of the segmentation as it is done in [23].…”

Section: Discussionmentioning

confidence: 99%

Three artificial intelligence data challenges based on CT and MRI

Lassau

Bousaid²,

Chouzenoux

et al. 2020

Diagnostic and Interventional Imaging

View full text Add to dashboard Cite

Purpose: The 2019 edition of the data challenge was organized by the French Society of Radiology (SFR) during the Journées Francophones de Radiologie with the aim to: (i) work on relevant problematics of public health (ii) build large multicentric and prospective databases and (iii) boost the French AI community around radiologists. In comparison to the 2018 edition a first objective was to increase the question's complexity by including 3D information and prognostic analysis. The second objective was to improve the database quality and quantity with more balance among classes and data from at least 1000 examinations per question. Material and method: Relevant clinical questions were proposed by organ societies of the SFR. Their feasibility was assessed by experts in the field of AI. A dedicated platform was set up for inclusion centers to safely upload their anonymized examinations in compliance with European regulation. The quality of the database was checked by experts weekly with annotations performed by radiologists. Multidisciplinary teams competed between September 11 th and November 13 th 2019. Results: Three questions were selected using different imaging and evaluation modalities, including: pulmonary nodule detection and classification from 3D CT, prediction of expanded disability status scale in multiple sclerosis using 3D MRIs and segmentation of muscular surface for sarcopenia estimation from 2D CT. A total of 4347 examinations were gathered of which only 6% were excluded. Three independent databases from 24 individual centers were created. A total of 143 participants was split into 20 multidisciplinary teams. Conclusion: Three data challenges with over 1200 GDPR compliant, multicentric, 2D/3D CT and MRI databases were organized for 20 multidisciplinary teams.

show abstract

“…In the last decade, we have seen that digitalization, automation, and artificial intelligence (AI) have significant impact in transforming healthcare. Of note, we have seen emerging data on the use of AI in NPC contouring [74] and head and neck cancer radiotherapy planning [75]. The utility of AI in NPC is primarily using prior data to augment decision making.…”

Section: Advanced Computing Technologiesmentioning

confidence: 99%

“…Two different AI methodologies are commonly employed, CNN [81,82] and atlas-based auto-segmentation [83][84][85], in a bid to improve the accuracy of contouring. Lin et al [74] applied a 3D convolutional neural network to 818 MRI data sets to develop an AI contouring tool that can automate the primary gross tumor volume in NPC. The AI tool successfully improved the accuracy in contouring with radiation oncologists having a higher median Dice similarity coefficient (DSC) after assistance by the AI contouring tool.…”

Section: Advanced Computing Technologiesmentioning

confidence: 99%

Emerging radiotherapy technologies and trends in nasopharyngeal cancer

Tseng

Leong

et al. 2020

Cancer Communications

View full text Add to dashboard Cite

Technology has always driven advances in radiotherapy treatment. In this review, we describe the main technological advances in radiotherapy over the past decades for the treatment of nasopharyngeal cancer (NPC) and highlight some of the pressing issues and challenges that remain. We aim to identify emerging trends in radiation medicine. These include advances in personalized medicine and advanced imaging modalities, standardization of planning and delineation, assessment of treatment response and adaptive re‐planning, impact of particle therapy, and role of artificial intelligence or automation in clinical care. In conclusion, we expect significant improvement in the therapeutic ratio of radiotherapy treatment for NPC over the next decade.

show abstract

Deep Learning for Automated Contouring of Primary Tumor Volumes by MRI for Nasopharyngeal Carcinoma

Cited by 271 publications

References 29 publications

Variations of Clinical Target Volume Delineation for Primary Site of Nasopharyngeal Cancer Among Five Centers in China

Variations of Clinical Target Volume Delineation for Primary Site of Nasopharyngeal Cancer Among Five Centers in China

Three artificial intelligence data challenges based on CT and MRI

Emerging radiotherapy technologies and trends in nasopharyngeal cancer

Contact Info

Product

Resources

About