Title:
METHOD FOR AUTOMATICALLY TURNING AND SEGMENTING LEFT VENTRICLE IN PET THREE-DIMENSIONAL IMAGE
Document Type and Number:
WIPO Patent Application WO/2023/040628
Kind Code:
A1
Abstract:
Disclosed is a method for automatically turning and segmenting the left ventricle in a PET three-dimensional image. The method comprises: constructing and training an encoder composed of a convolution module, a plurality of residual-convolution modules and a down-sampling module, a space transformation network, a decoder containing a plurality of upsampling modules and residual-convolution modules, and an automatic turning and segmentation model of the left ventricle in a PET 3D image having skip connection between the encoder and the decoder; using a regular view A as a model input; using said model to automatically rotate an image to a clinical standard view, and, on the basis of the standard view, obtaining a segmentation result of a left ventricular structure in said view. In the present invention, a position feature and a semantic feature of an image are extracted by using a multi-task learning deep learning network, and automatic turning from different angles to a standard view, heart positioning, and structural segmentation of the left ventricle are achieved. Moreover, operation using one-stop processing reduces manual turning, the complexity of segmentation, and human error, increasing the convenience and accuracy of image operation.
Inventors:
ZHU WENTAO (CN)
ZHANG DUO (CN)
ZHU HAI (CN)
ZHANG DUO (CN)
ZHU HAI (CN)
Application Number:
PCT/CN2022/114975
Publication Date:
March 23, 2023
Filing Date:
August 26, 2022
Export Citation:
Assignee:
ZHEJIANG LAB (CN)
International Classes:
G06T7/00; G06N3/04; G06T7/33
Domestic Patent References:
| WO2020199593A1 | 2020-10-08 |
Foreign References:
| CN113516658A | 2021-10-19 | |||
| CN112508949A | 2021-03-16 | |||
| US20210174513A1 | 2021-06-10 |
Attorney, Agent or Firm:
HANGZHOU QIUSHI PATENT OFFICE CO., LTD. (CN)
Download PDF: