Thyroid nodule real-time segmentation method based on full convolution dense hole network

Abstract

The invention discloses a thyroid nodule real-time segmentation method based on a full convolution dense hole network. The method comprises the following steps: 1, obtaining thyroid data and carryingout preprocessing; 2, labeling the obtained data to serve as a data set for training a full convolution dense hole network model; 3, constructing a full convolution dense hole network model based on dense connection, and performing parameter training; 4, replacing a convolution kernel in the convolution layer with hole convolution, and decomposing the convolution kernel by convolution kernel decomposition; 5, performing data standardization and nonlinear activation processing on the input of the convolution layer; and 6, analyzing and comparing the segmentation effect and efficiency of the full convolution dense hole network model.

Description

technical field [0001] The invention belongs to the field of deep learning and image processing, and relates to a convolutional neural network and image semantic segmentation technology, in particular to a real-time segmentation method for thyroid nodules based on a fully convolutional dense hole network. Background technique [0002] Thyroid nodules are the most common abnormalities in the endocrine system, and their malignant potential makes them clinically important. Ultrasonography is the imaging modality of choice for the diagnosis of thyroid nodules. In clinical practice, radiologists based on the aspect ratio of nodules in ultrasound images, the presence of calcification, structure (diffusion, single or multiple), boundaries, echo characteristics (hyperechoic, isoechoic and hypoechoic), etc. The evaluation criteria are used to diagnose benign and malignant thyroid glands. However, due to the influence of cognitive ability, subjective experience, and fatigue, doctors ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the problem that the existing semantic segmentation model has too many parameters and cannot efficiently run on computer-aided diagnosis equipment with limited computing resources. The present invention designs a new network model, adopts dense connection, empty convolution and The convolution kernel decomposition method maintains high precision and greatly reduces the parameters of the model

Images