Can preoperative MR imaging predict optic nerve invasion of retinoblastoma?

PURPOSE: To evaluate the accuracy of pre-operative MRI for the detection of optic nerve invasion in retinoblastoma.
MATERIALS AND METHODS: Institutional review board approval and informed consent were waived for this retrospective study. A total of 41 patients were included. Inclusion criteria were histologically proven retinoblastoma, availability of diagnostic-quality preoperative MR images acquired during the 4 weeks before surgery, unilateral retinoblastoma, and normal-sized optic nerve. Two radiologists retrospectively reviewed the MR images independently. Five imaging findings (diffuse mild optic nerve enhancement, focal strong optic nerve enhancement, optic sheath enhancement, tumor location, and tumor size) were evaluated against optic nerve invasion of retinoblastoma. The predictive performance of all MR imaging findings for optic nerve invasion was also evaluated by the receiver operating characteristic curve analysis.
RESULTS: Optic nerve invasion was histopathologically confirmed in 24% of study population (10/41). The differences in diffuse mild enhancement, focal strong enhancement, optic sheath enhancement, and tumor location between patients with optic nerve invasion and patients without optic nerve invasion were not significant. Tumor sizes were 16.1mm (SD: 2.2mm) and 14.9 mm (SD: 3.6mm) in patients with and without optic nerve involvement, respectively (P=0.444). P-Values from binary logistic regression indicated that all five imaging findings were not significant predictors of tumor invasion of optic nerve. The AUC values of all MR imaging findings for the prediction of optic nerve invasion were 0.689 (95% confidence interval: 0.499-0.879) and 0.653 (95% confidence interval: 0.445-0.861) for observer 1 and observer 2, respectively.
CONCLUSION: Findings of MRI in patients with normal-sized optic nerves have limited usefulness in preoperatively predicting the presence of optic nerve invasion in retinoblastoma.