E2718. Improving Radiologist Agreement in Assessing Neurovascular Compression of the Trigeminal Nerve
  1. Can Ozutemiz; University of Minnesota
  2. Kellen Mulford; University of Minnesota
  3. Yu-Hui Huang; University of Minnesota
  4. Kamran Masood; University of Minnesota
  5. Sean Moen; University of Minnesota
  6. Sam Ndoro; University of Minnesota
Trigeminal neuralgia (TN) is a painful neuropathic condition. Although TN is primarily idiopathic, MRI is used to identify secondary causes including multiple sclerosis and neurovascular conflict (NVC) of the cisternal portion of the trigeminal nerve. The degree of NVC is used to determine whether the patient is an appropriate candidate for surgical intervention and can help guide the surgical approach. Existing studies examining the relationship of TN and NVC either do not report agreement statistics or only include patients with TN, which biases readers towards an aggressive read of NVC. Our aim is to evaluate the level of agreement in assessing NVC and the factors that affect that agreement using a large database of skull base MRI images.

Materials and Methods:
Submillimetric T2-weighted skull base MRI images that fully captured the course of the trigeminal nerve were acquired from the institutional PACS along with the acquisition parameters. Two radiology residents and one neuroradiology fellow were trained under the guidance of a senior board-certified neuroradiologist to grade NVC categorically on 150 training cases. NVC compression was graded as A: no compression, B: nerve abutment (simple contact), C: nerve compression (thinning or displacement of the nerve). The four readers then independently graded 2596 nerves (representing 1298 cases). Nerve morphometry measurements were recorded by an independent reader for each nerve. Fleiss’ Kappa and absolute agreement were calculated and compared between factors of acquisition and nerve morphometry.

Eighty-one cases were removed for image quality issues. A total of 997 1.5-Tesla and 220 3.0-Tesla MRI cases consisting of the following sequences: 44 SSFP, 68 3D-TSE, 321 CISS, and 805 T2-SPACE were included in the analysis. No reader drift was observed over the 7 month reading window. Fleiss’ Kappa calculated on the reported degree of NVC was 0.51 (pairwise: 0.47 - 0.6), demonstrating moderate agreement. Of the 1932 (78%) nerves in the reading set, with 3 or 4 readers in agreement, 40% of nerves had some degree of NVC, with 659 (34%) nerves found to be abutted and 110 (6%) nerves compressed. TN affected nerves were much more likely to exhibit compression level NVC, while unaffected nerves were more likely to exhibit abutment level NVC (p < 0.001). A slight decreasing trend in agreement was observed as slice thickness increased (p = 0.1). Agreement was higher in coronal acquisitions (K: 0.55 vs 0.50) and lower with the PSIF sequence (K: 0.40 with SSFP vs. 0.54 others). No difference in agreement was observed between differing field strengths, patient age or gender, or between nerves with differing shape.

This study demonstrates that radiologists agree only moderately on the features of NVC present on MRI, and that compression level NVC is a particular feature of TN. To maximize the information that can be obtained from MRI scans targeting the trigeminal nerve: CISS or SPACE sequences acquired in the coronal plane with minimized slice thicknesses should be used.