A Novel U-Like Network For The Segmentation Of Thoracic Organs

Accurate segmentation of organs at risk (OARs) in computed tomography (CT) is an essential step in radiation therapy for thoracic cancer treatment. However, the manual segmentation of OARs is time-consuming and subject to inter-observer variation. In this paper, we propose a novel U-like deep convol...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2020 IEEE 17th International Symposium on Biomedical Imaging Workshops (ISBI Workshops) s. 1 - 4
Hlavní autoři: Shi, Jun, Wen, Ke, Hao, Xiaoyu, Xue, Xudong, An, Hong, Zhang, Hongyan
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.04.2020
Témata:
Biomedical imaging
Computed tomography
Convolution
convolutional neural network (CNN)
Decoding
hybrid dilated convolution
Image segmentation
Lung
multi-task learning
OARs segmentation
radiation therapy
Two dimensional displays
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Accurate segmentation of organs at risk (OARs) in computed tomography (CT) is an essential step in radiation therapy for thoracic cancer treatment. However, the manual segmentation of OARs is time-consuming and subject to inter-observer variation. In this paper, we propose a novel U-like deep convolutional neural network (CNN) architecture, which adopts the encoder-decoder design, to automatically segment the OARs in thoracic CT images. In our method, hybrid dilated convolution (HDC) is employed to enlarge the receptive field of the encoder part, and a pyramid backbone with lateral connections between encoder and decoder is utilized to capture contextual information at multiple scales. To reduce the false-positive segmentation results, we use the multi-task learning strategy to add an auxiliary classifier branch to the network. The experiments demonstrate that the proposed method outperforms other state-of-the-art models and the results have a good consistency with that of experienced radiologists.
DOI:10.1109/ISBIWorkshops50223.2020.9153358