Ultrasonic thyroid nodule segmentation method based on asymmetric network

Abstract

The invention discloses an ultrasonic thyroid nodule segmentation method based on an asymmetric network, and belongs to the field of deep learning image segmentation. The coding layer is a U-Net network, more context semantic feature information can be obtained, thyroid nodules with variable scales, shapes, positions and the like can be accurately detected, and the decoding layer uses the combination of hole convolution, residual connection and convolution kernel decomposition, so that detail textures and edge information can be gradually restored from high-dimensional global context semanticfeature information. According to the model constructed by the method, the segmentation precision similar to that of a high-precision model is achieved on a thyroid data set, the IOU is about 2% higher than that of a U-Net model, the time for averagely segmenting one picture is about 1.9 s, and the efficiency is greatly improved compared with the manual segmentation efficiency of doctors. The method solves the problems of excessive interference factors and serious loss of image edge information in a traditional method, and has good accuracy and generalization ability.

Description

technical field [0001] The invention belongs to the field of deep learning image segmentation, and in particular relates to a thyroid nodule ultrasound image segmentation method based on an asymmetric network. Background technique [0002] Thyroid disease is currently the most common nodular disease, and ultrasonography is the primary method for diagnosing thyroid nodules. However, the ultrasonic medical images collected by medical instruments have low contrast, more speckle noise, and thyroid nodules change in shape and position, which brings great difficulty to the diagnosis of thyroid diseases. [0003] In recent years, with the continuous development of medical imaging technology, computer-aided diagnosis has gradually attracted the attention of researchers. The use of neural networks to establish medical image segmentation models has been widely used, and the research on thyroid ultrasound image-aided diagnosis has also made great progress. In 2015, Ronneberger et al. ...

AI Technical Summary

Problems solved by technology

However, for thyroid ultrasound images, the image contrast is low, the thyroid nodules are small, the ratio of foreground to background is extremely unbalanced, and there are many burrs on the border of nodules, which are easily interfered by tissues and organs around the thyroid gland, so good segmentation results cannot be obtained

Images