Automatic image analysis of episcleral hemangioma applied to the prognosis prediction of trabeculotomy in Sturge–Weber syndrome

Abstract

Sturge–Weber syndrome (SWS) is a rare neurocutaneous disorder, and it can cause eye diseases such as glaucoma. As the disease progresses, the SWS patients will not be able to see clearly what is in front of them, and the reduced quality of visual experience will cause a reduction in quality of life. In the current work, we intend to use computer vision to improve the success rate of surgery for SWS patients. Image analysis techniques were first applied to automatically classify episcleral hemangioma distribution patterns of young SWS patients, and the extracted image features were subsequently used to predict their trabeculectomy prognosis. The experimental data was the snapshot captured from high-resolution surgery video in Shanghai Ninth People’s Hospital from February 2016 to July 2017. Subsequently, we used the local entropy threshold method on the snapshot to segment episcleral vessels and realized binarization to extract the density of episcleral vessels. The two indexes viz. episcleral vessel entropy and episcleral vessel density were extracted by the feature extraction method. After obtaining the feature values, K-means unsupervised clustering was carried out, to group the two categories which were the successful operation case and the failed operation case. Based on two categories obtained by unsupervised clustering method, survival analysis was conducted to obtain the P-value. The classification result was that 21 eyes were grouped into successful operation cases and 19 eyes were grouped into failed operation cases. After survival analysis, the P-value of two groups clustered by K-means was 0.046, and this suggested there were significant differences between the successful operation case and the failed operation case. Two extracted features represented the severity of the disease as the ophthalmologists observed. The new classification of SWS patients achieved by automatic image analysis was significantly related to the trabeculotomy outcome which might provide a new approach to the prognostic prediction for SWS patients.