Scholarlink

Introduction <p>Detecting high risk characteristics in intraocular retinoblastoma is important for determining metastatic risk.</p> Methods <p>In this pilot study, preoperative Magnetic Resonance Imaging (MRI) radiomics data were compared to high-risk histopathology features after enucleation. A total of 2142 MRI features were extracted, and relevant features were selected to create predictive models for optic nerve and choroidal invasion.</p> Results <p>Among 31 eyes of 29 patients, 14 eyes had optic nerve invasion and 6 had choroidal invasion. The predictive model for optic nerve (ON) invasion achieved Area under the receiver operating characteristic curve (AUROC) of 0.85 (accuracy 0.84, sensitivity 0.93, specificity 0.76) using the radiomic features of maximum 2D diameter slice on the unaffected optic nerve (ON), correlation of grey level co-occurrence matrix (GLCM) on the affected ON, and kurtosis of affected ON. For choroidal invasion the model achieved AUROC of 0.86 (accuracy 0.94, sensitivity 0.83, specificity 0.96) using the radiomic features of sphericity of affected vitreous and surface volume ratio of affected ON.</p> Conclusions <p>Radiomics could be useful to guide the decision of globe salvage versus enucleation in advanced retinoblastoma. Larger studies are required to validate and enhance this predictive model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Radiomics to predict high risk features in intraocular retinoblastoma

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Aparna Ramasubramanian,
  • Kun-Han Lu,
  • Lauren A. Dalvin,
  • Monique Riemann,
  • Todd Abruzzo,
  • Chi Wah Wong,
  • Cristian Tomasetti,
  • Justin Moser,
  • William Adam Libling,
  • Ron Korn

摘要

Introduction

Detecting high risk characteristics in intraocular retinoblastoma is important for determining metastatic risk.

Methods

In this pilot study, preoperative Magnetic Resonance Imaging (MRI) radiomics data were compared to high-risk histopathology features after enucleation. A total of 2142 MRI features were extracted, and relevant features were selected to create predictive models for optic nerve and choroidal invasion.

Results

Among 31 eyes of 29 patients, 14 eyes had optic nerve invasion and 6 had choroidal invasion. The predictive model for optic nerve (ON) invasion achieved Area under the receiver operating characteristic curve (AUROC) of 0.85 (accuracy 0.84, sensitivity 0.93, specificity 0.76) using the radiomic features of maximum 2D diameter slice on the unaffected optic nerve (ON), correlation of grey level co-occurrence matrix (GLCM) on the affected ON, and kurtosis of affected ON. For choroidal invasion the model achieved AUROC of 0.86 (accuracy 0.94, sensitivity 0.83, specificity 0.96) using the radiomic features of sphericity of affected vitreous and surface volume ratio of affected ON.

Conclusions

Radiomics could be useful to guide the decision of globe salvage versus enucleation in advanced retinoblastoma. Larger studies are required to validate and enhance this predictive model.