MR image placenta segmentation method of multi-task generative adversarial model

Abstract

The invention discloses an MR image placenta segmentation method for a multi-task generative adversarial model. The method comprises the steps: constructing a training set and a test set, constructing a segmentation model through employing a generative adversarial network, training the segmentation model through employing the training set, the test set, and a total adversarial loss function of the segmentation model, and obtaining a trained segmentation model, finally, the placenta MR image is segmented through the segmentation model, segmented pictures of the placenta MR image are obtained, and a total adversarial loss function of the segmentation model is constructed based on maximization of discrimination loss and minimization of multi-task generation loss; the method has the advantages of being high in segmentation precision and capable of meeting clinical requirements.

Description

Technical field [0001] The present invention relates to a MR image placenta-division method, in particular, involving a MR image placenta segmentation method of a multi-task generating counterming model. Background technique [0002] In recent years, with the increase in the angels, the increasing abortion, the incidence of placenta implantation has increased the rise and has become one of the common clinical diseases in obstetric. The soft tissue resolution of magnetic resonance (MR) is high, and the image quality can be imagined in arbitrary direction, and the image quality is not affected by the fetal position, the maternal body type and the amount of water, and has gradually become an important to diagnose placental implantation. means. In clinical practice, precise segmentation of placental tissue is to identify placental implantation and assessment of implantation. By analysis of placenta and its surrounding tissue organs, it is expected to achieve placenta adhesion, implan...

AI Technical Summary

Problems solved by technology

However, the loss function of the traditional U-Net network regards each pixel in the image as independent of other pixels, and predicts the category of each pixel, which cannot reflect the connection between adjacent pixels and the relationship between pixels. Interrelationships are easily overlooked, resulting in incomplete objects in the segmentation results

When it is applied to placenta segmentation in pelvic MR images, it is difficult to express the outline of placenta tissue and its relationship with adjacent tissues and organs, and the segmentation accuracy still needs to be improved

Images