Method for automatically checking additional retinopathy of premature infant

Abstract

The embodiment of the invention discloses a method for automatically checking additional retinopathy of a premature infant, which comprises the following steps of: inputting a retina image of the premature infant, firstly judging whether the image is a posterior pole image by using a deep neural network, and if the image is the posterior pole image, segmenting a blood vessel in an original image by using a semantic segmentation model to obtain a binary image of the blood vessel, positioning a posterior polar region according to the original image, cutting the blood vessel image by using the posterior polar region to obtain a posterior polar blood vessel image, and finally classifying the posterior polar blood vessel image by using a deep neural network so as to judge whether the original image belongs to the additional lesion or not. By the adoption of the method, business domain knowledge is combined, the influence of other lesion characteristics can be shielded, and the accuracy (including indexes such as sensitivity, specificity and F1) of an automatic examination system for the additional retinopathy of the premature infant is greatly improved.

Description

technical field [0001] The invention relates to the technical field of ophthalmic disease recognition and image recognition, in particular to an automatic inspection method for additional lesions of retinopathy of premature infants. Background technique [0002] Retinopathy of premature (Retinopathy of premature, ROP) is the main cause of visual impairment and irreversible blindness in children. About 1.2% (184,700) premature infants worldwide develop ROP each year, and about 30,000 premature infants develop permanent Sexual vision loss or even blindness. With the popularization of neonatal intensive care unit and the development of life support technology, the survival rate of premature infants, especially very low birth weight infants, has been greatly improved, and the incidence of ROP is also increasing. ROP screening can be used for early diagnosis and timely referral for treatment, and can effectively avoid or reduce ROP-related visual impairment. The diagnosis of RO...

AI Technical Summary

Problems solved by technology

However, most of these methods regard the classification of ROP additional lesions as ordinary image classification, and a few methods first segment blood vessels and then classify them. Since the incidence of ROP additional lesions is relatively low and the number of samples is relatively small, this simple classification method The algorithm design and business domain knowledge are not combined, resulting in low accuracy, especially the diagnosis of additional lesions is easily interfered by other features (bleeding, peripheral lesions)

Images