Elsevier

Displays

Volume 71, January 2022, 102118
Displays

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

https://doi.org/10.1016/j.displa.2021.102118Get rights and content

Highlights

  • A novel image analysis method was proposed to predict trabeculectomy prognosis of SWS patients.

  • Extracted image features were related to the severity of the disease as the ophthalmologists observed.

  • Results demonstrated that classification model was significantly related to the trabeculotomy outcome.

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.

Introduction

Sturge–Weber syndrome (SWS) is a rare neurocutaneous disorder with an incidence ranging from 1 in 50,000 to 1 in 20,000 [1]. It is a sporadic, congenital, neurocutaneous disorder with angiomas that is characterized by leptomeningeal angiomatosis, glaucoma, and cutaneous capillary malformations in the distributions of the trigeminal nerve [2]. Nowadays, trabeculotomy has some chance of curing young SWS with glaucoma [3], [4]. To improve the recovery rate, there is an urgent need to find indexes that characterize or predict the success of the surgery. Compared to other organs, the blood vessels of the eyes are visible and can indicate the health condition of the body organs [5]. SWS is a disease characterized by vascular malformations. In previous studies, episcleral hemangiomas were found in all patients with secondary glaucoma [6]. Therefore, analyzing the blood vessels in the eyes seems to provide a window for us to evaluate glaucoma. However, the researchers did not find an association between scleral hemangioma and glaucoma outcome.

Inspired by this, we observed the vascular distribution patterns in dozens of SWS patients, showing that the prognosis of glaucoma after trabeculotomy in SWS patients was closely related to the distribution pattern of scleral hemangioma in the eyes [7]. We had established a scoring system to distinguish the degree of vascular malformation. The more severe the vascular malformation of the scleral surface, the higher the score. The artificial scores were used to divide episcleral blood vessels into two categories: namely MEVAN (multiple episcleral vascular abnormal network) and SEVAN (simple episcleral vascular abnormal network) [7]. In our study, 24 of the 50 eyes (46 patients) were assigned to MEVAN, and the remaining 26 were assigned to SEVAN. In the subsequent two years of follow-up, the trabeculotomy success rate of the SEVAN group was 89.7%, and the MEVAN group was 36%. The cumulative success rates were significantly different between the 2 groups (p=0.001). Therefore, we concluded that the vascular distribution pattern could be an indicator of prognosis.

Our previous study was mainly based on the observation results of ophthalmologists from Shanghai Ninth People’s Hospital. Since SWS is a rare disease, the quantitative lacking-manual scoring system might be affected by doctors’ experience and other subjective factors. Hence, in the current work, we attempted to apply an image analysis algorithm to extract image features representing the distribution characteristics of episcleral vessels, and afterward used an unsupervised clustering method to classify the SWS patients into two categories namely the successful operation case and the failed operation case based on the extracted image features. The survival analysis was conducted to explore the relationship between clustering results and surgical prognosis. Automatic diagnosis by the designed algorithm is of great significance to the process of auxiliary medical diagnosis, and is not easily affected by the subjective experience of ophthalmologists.

Section snippets

Experimental data

The experimental data in this study were obtained from SWS patients under 4 years old who underwent trabeculotomy in Shanghai Ninth People’s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine from February 2016 to July 2017. The study was approved by the Review Committee of the Ninth People’s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine. Diagnosis, surgery, follow-up, and necessary treatment were provided by the ophthalmologist (Dr. Guo Wenyi). We

Segmentation results of episcleral vessels

Fig. 2 demonstrates the original image of the episcleral vessels (first line) and the binary image of the episcleral vessels (second line). As shown in Fig. 2, most blood vessels in images can be segmented by the selected threshold-based segmentation algorithm. In general, the overall structure of the segmented vessels is well maintained. At the same time, there is less discrete noise in segmented images, and this will facilitate the subsequent statistical image analysis such as image feature

Discussion

In previous image analysis of ophthalmic vascular disease, diabetic retinopathy, hypertension, and glaucoma were mainstream research objects [23]. The goal of these previous researches was to automatically and objectively distinguish different stages of the disease development using image analysis technique. For the standard datasets of these ophthalmic vascular diseases, by extracting image features specific to disease development characteristics, the classification accuracy of the

Conclusions

In conclusion, two image features viz. entropy and density extracted from episcleral blood vessel image can be used as indicators to automatically and objectively evaluate the severity of episcleral hemangioma, the severity of glaucoma and to predict trabeculotomy outcomes of young SWS patients. Guided by these two image features, ophthalmologists may develop the effective surgical strategy to raise the success rate of SWS treatment.

CRediT authorship contribution statement

Lejing Zhang: Methodology, Validation, Writing – original draft, Writing – review & editing. Yue Wu: Funding acquisition, Methodology, Validation, Writing – original draft, Writing – review & editing. Menghan Hu: Funding acquisition, Methodology, Validation, Writing – original draft, Writing – review & editing. Wenyi Guo: Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work is sponsored by the Interdisciplinary Program of Shanghai Jiao Tong University, China (YG2019QNA18), the “Chenguang Program” of Shanghai Education Development Foundation, China and Shanghai Municipal Education Commission, China (No. 19CG27), the Science and Technology Commission of Shanghai Municipality, China (No. 19511120100), the GHfund B (No. 20210702), and the Fundamental Research Funds for the Central Universities, China .

References (31)

  • AkbarShahzad et al.

    Automated techniques for blood vessels segmentation through fundus retinal images: A review

    Micros. Res. Tech.

    (2019)
  • RicciE. et al.

    Retinal blood vessel segmentation using line operators and support vector classification

    IEEE Trans. Med. Imaging

    (2007)
  • S. Garg, J. Sivaswamy, S. Chandra, Unsupervised curvature-based retinal vessel segmentation. In: 4th IEEE international...
  • BhuiyanA. et al.

    Blood vessel segmentation from color retinal images using unsupervised texture classification

    IEEE Int. Conf. Image Process.

    (2007)
  • T.A. Soomro, M.A.U. Khan, J. Gao, T.M. Khan, M. Paul, N. Mir, Automatic Retinal Vessel Extraction Algorithm. In:...
  • Cited by (0)

    This paper was recommended for publication by Zhi-Cheng Li.

    View full text