Thyroid nodule identification and segmentation method and system, storage medium and equipment

Abstract

The invention discloses a thyroid nodule identification and segmentation method, and the method comprises the steps: obtaining a preprocessing sample, and inputting the preprocessing sample into a skeleton network for sample feature extraction; respectively inputting the extracted feature samples into a first branch and a second branch for training; respectively calculating the classification loss of the trained first branch and the segmentation loss of the trained second branch; obtaining a total loss training model through linear weighting of classification loss and segmentation loss, wherein the total loss training model can judge benign and malignant categories and lesion areas of thyroid nodules of the input image. According to the invention, dependence on human resources such as doctors and the like can be reduced, the possibility of human errors is reduced, and intelligent detection is realized. Meanwhile, the iconography characteristics of the thyroid nodules are fully considered, so that the interference problem caused by inconsistency of image categories and pathological categories of the deep learning model is relieved, the classifier can better learn proper iconography characteristics, and accurate segmentation of lesion areas on the classifier is achieved.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a method, system, storage medium and equipment for identifying and segmenting thyroid nodules. Background technique [0002] Thyroid nodule is a very common clinical disease with an incidence of about 19% to 68% of the population, of which about 5% to 15% are malignant. Ultrasound imaging is the first choice for the examination of thyroid nodules because of its low cost and no damage to the human body. At present, the identification and segmentation of thyroid nodules are mainly divided into two methods: manual identification and segmentation and computer-aided identification and segmentation. [0003] However, due to the characteristics of low contrast and noise in ultrasound images, the identification of nodules requires high skill and experience of radiologists, which may easily lead to a certain degree of missed and false detections. Moreover, althou...

AI Technical Summary

Problems solved by technology

[0003]However, due to the characteristics of low contrast and noise in ultrasound images, the identification of nodules has high requirements on the skills and experience of imaging doctors, and it is easy to Lead to a certain degree of missed detection and false detection

Moreover, although the existing automatic identification and segmentation technology has been continuously developed and the classification effect is getting better and better, it ignores the inconsistency between the imaging standard and the pathological grading standard: that is, some images should be defined as malignant by the imaging standard. , but the gold standard for clinical diagnosis of the thyroid gland—the result of pathology is benign, so it is easy to occur

Images