Metalearning-based thyroid ultrasound nodule fuzzy boundary-oriented segmentation method

Abstract

The invention discloses a segmentation method for a thyroid ultrasound nodule fuzzy boundary based on meta-learning. The segmentation method comprises the following steps: 1) synthesizing a noise mark; 2) providing an objective function calculation method based on meta learning; 3) designing a meta-mask network based on a full convolution structure; and 4) training the model by adopting an iterative optimization algorithm. According to the method provided by the invention, a weight map is automatically estimated to evaluate the importance of each pixel in segmentation network learning. The loss value graph of the prediction segmentation result is used as the input of the meta-mask network, so that a damaged layer can be identified and a smaller weight can be distributed to the damaged layer. In the training process, the segmentation network and the meta-mask network are updated in an alternating mode, the method can train the powerful segmentation network from a large number of low-quality marked images, and the generalization performance of training the deep network on damaged training data is remarkably improved.

Description

technical field [0001] The invention relates to the field of deep learning and auxiliary medical technology, to a damaged image mining model based on meta-learning, and in particular to a segmentation method based on meta-learning for fuzzy boundaries of thyroid ultrasound nodules. Background technique [0002] The thyroid gland is an important organ located in the neck of the human body. It produces and secretes two important hormones, namely triiodothyronine and thyroxine, which are responsible for regulating the metabolism of the human body. Due to its important role in the human body, it is important to diagnose and treat thyroid disease. [0003] An important problem that is common in the thyroid region is the appearance of nodular lesions of the thyroid. Thyroid nodules are abnormal lumps that appear in the area of ​​a person's thyroid gland. They can be caused by many factors, including iodine deficiency, overgrowth of normal thyroid tissue, or thyroid cancer. Thyr...

AI Technical Summary

Problems solved by technology

However, most existing methods use fixed prior knowledge to guide the pseudo-label training network

Therefore, they are usually not stable enough for training samples with complex noise distributions

Caused the results of the segmentation network to be severely affected when using corrupted labels as supervised samples

Images