Automated diagnosis and quantitative analysis of plus disease in retinopathy of prematurity based on deep convolutional neural networks

Abstract

Background: The purpose of this study was to develop an automated diagnosis and quantitative analysis system for plus disease. The system provides a diagnostic decision but also performs quantitative analysis of the typical pathological features of the disease, which helps the physicians to make the best judgement and communicate the decisions. Methods: The deep learning network provided segmentation of the retinal vessels and the optic disc (OD). Based on the vessel segmentation, plus disease was classified and tortuosity, width, fractal dimension and vessel density were evaluated automatically. Results: Thetrainednetworkachievedasensitivityof95.1%with97.8%specificityfor the diagnosis of plus disease. For detection of preplus or worse, the sensitivity and specificity were 92.4% and 97.4%. The quadratic weighted k was 0.9244. The tortuosities for the normal, preplus and plus groups were 3.61 AE 0.08, 5.95 AE 1.57 and 10.67 AE 0.50 (10 4 cm À3 ). The widths of the blood vessels were 63.46 AE 0.39, 67.21 AE 0.70 and 68.89 AE 0.75 lm. The fractal dimensions were 1.18 AE 0.01, 1.22 AE 0.01 and 1.26 AE 0.02. The vessel densities were 1.39 AE 0.03, 1.60 AE 0.01 and 1.64 AE 0.09 (%). All values were statistically different among the groups. After treatment for plus disease with ranibizumab injection, quantitative analysis showed significant changes in the pathological features. Conclusions: Our system achieved high accuracy of diagnosis of plus disease in retinopathy of prematurity. It provided a quantitative analysis of the dynamic features of the disease progression. This automated system can assist physicians by providing a classification decision with auxiliary quantitative evaluation of the typical pathological features of the disease.

Comparative Study on Optic Disc Features of Premature Infants and Term Infants

“…Current study of neonatal optic discs mainly focus on optic discs in preterm infants in different premature stages of preterm infant development, including premature infants with xed eyeballs in formalin due to premature death [1,19], birth babies [20,21] and premature birth children [15,18,8,22]. Different methods have also been introduced into analytical research.…”

Comparative Study on Optic Disc Features of Premature Infants and full-term newborns

“…Current study of neonatal optic discs mainly focused on optic discs in preterm infants in different premature stages of preterm infants development, including premature infants with xed eyeballs in formalin due to premature death [1,21], birth babies [20,21] and premature birth children [17,20,10,24]. Different methods have also been introduced into analytical research.…”

Artificial Intelligence in Retinopathy of Prematurity Diagnosis

“…One approach to have more explainable AI is to combine DL methods with traditional feature extraction, and several groups have attempted this for plus disease. 42,43 Mao et al 42 trained a DL network to segment retinal vessels and the optic disc and to diagnosis plus disease based on automated quantitative characterization of pathologic features, such as vessel tortuosity, width, fractal dimension, and density. Graziani et al 43 compared the black box CNN features with known hand-crafted features using regression concept vectors.…”

“… 26 Deep CNNs differ from traditional feature extraction and machine learning systems by allowing the CNN to learn features that best correlate the input image (step 1) with the diagnosis (4) with or without preprocessing but without explicit human defined features (step 2). 27 , 28 , 42 …”