Evaluation method of breast ultrasound image quality based on auxiliary diagnosis and subjective aesthetics

Abstract

The invention provides a breast ultrasound image quality evaluation method based on auxiliary diagnosis and subjective aesthetics. Firstly, construct an ultrasound image database with a small amount of strong-supervised annotation information and a large amount of weak-supervised annotation information; then, adopt a variety of small-sample learning strategies to train the deep network model DNN-1 for automatic segmentation of lesion ROI; then, combine the ROI image with the original image Input together, and train to obtain the ultrasound-assisted diagnosis network DNN-2; finally, transfer the convolutional layer of the auxiliary diagnosis network, combined with the doctor's aesthetic score, train the aesthetic evaluation convolutional neural network DNN-3, and combine DNN-2 based on the attention mechanism , to obtain the breast ultrasound image quality evaluation results.

Description

technical field [0001] The invention belongs to the technical field of medical image processing, and in particular relates to a breast ultrasound image quality evaluation method based on auxiliary diagnosis and subjective aesthetics. Background technique [0002] During medical sonography, image quality is critical to diagnostic accuracy. At present, although there is no unified standard worldwide, for gray-scale ultrasound (B-ultrasound) and Doppler ultrasound, each country or region has relevant measurement standards, such as probe configuration, display performance, image uniformity, sensitivity ( Maximum viewing depth, signal-to-noise ratio), geometric accuracy, spatial resolution and contrast resolution, etc. For example, "Sassaroli Elisabetta, Crake Calum, Scorza Andrea, Kim Don-Soo, Park Mi-Ae. Image quality evaluation of ultrasound imaging systems: advanced B-modes. [J]. Journal of applied clinical medical physics, 2019, 20(3) .” proposed a method to measure the im...

AI Technical Summary

Problems solved by technology

The model trained in the above literature can only achieve the approximate positioning of the target area, but cannot achieve accurate segmentation. Accurate segmentation requires a large number of manually marked images, and images of different parts need to be marked by physicians with corresponding rich experience

Images