An Automated Method for Determining Vestibular Schwannoma Size and Growth

Vestibular schwannoma (VS) growth in the greatest linear dimension is one metric used to determine VS progression. Many studies characterizing VS growth also use manual volumetric segmentation of VS to increase the sensitivity of detecting growth and to better characterize lesions; however, high-resolution scans and radiologist-time required to make these measurements are rarely available. To create an automated means of measuring tumor size, we sought to compare manual versus automated measurements of tumor length, area, and volume in a cohort of VS patients.

Methods: We used selected cases from a database of VS cases at a tertiary care center. A total of 36 patients were selected who had at least two T1-weighted axial MRI scan with contrast at least six months after the initial scan, prior to any intervention. We manually measured the greatest linear dimension on axial imaging, in addition to manually segmenting each tumor slice. In comparison, we automatically segmented the tumor using the Chan-Vese image segmentation method. Afterwards, we calculated the area within the contour of the tumor, the major axis length of a “best-fit ellipse” around the tumor, and the tumor volume if the tumor can be contoured in more than one slice.

Results: The average age of the patients was 59.7 years (range, 35.4–76.9), with an average tumor size based on the greatest linear axial diameter was 9.0 mm, (range: 1.8–26.6). When comparing manual linear measurements with the major axis of the “best-fit ellipse,” the intraclass correlation coefficients (ICC) for the first and second measurements were 0.982 and 0.984, respectively. When comparing the absolute difference between the first and second measurements between the manual and the major axis of the “best-fit ellipse” linear measurements, the ICC was 0.918. When defining growth as an increase in tumor size ≥ 2 mm, 17/37 (45.9%) demonstrated linear growth when utilizing the manual measurement of the greatest linear dimension. However, only 4/37 (10.8%) demonstrated growth when utilizing the major axis length of the “best-fit ellipse” around the tumor, which is significantly different as compared with the manual measurement (p = 0.002). To date, by defining volumetric tumor growth as an increase in tumor size ≥ 20%, 10/19 (52.6%) of tumors demonstrate growth via semi-automated volumetric measurements. Manual segmentation of volumes will be completed and compared with semi-automated measurements.

Conclusion: With our analysis to date, manual versus automated linear measurements are in excellent agreement. However, when determining if the tumor grew, there is a significant difference in defining which tumors grew. Conclusions are pending as to volumetric measurements.