Navigated intraoperative ultrasound in neuro-oncology: volumetric accuracy and correlation with high-field MRI

Abstract

OBJECTIVE The use of intraoperative techniques to detect residual tumors has recently become increasingly important. Intraoperative MRI has long been considered the gold standard; however, it is not widely used because of high equipment costs and long acquisition times. Consequently, real-time intraoperative ultrasound (ioUS), which is much less expensive than MRI, has gained popularity. The aim of the present study was to evaluate the capacity of ioUS to accurately determine the primary tumor volume and detect residual tumors. METHODS A prospective study of adult patients who underwent surgery for intra-axial brain tumors between November 2017 and October 2020 was performed. Navigated intraoperative ultrasound (nioUS) of the brain was used to guide tumor resection and to detect the presence of residual disease. Both convex (5–8 MHz) and linear array (6–13 MHz) probes were used. Tumor volume and residual disease were measured with nioUS and compared with MR images. A linear regression model based on a machine learning pipeline and a Bland-Altman analysis were used to assess the accuracy of nioUS versus MRI. RESULTS Eighty patients (35 females and 45 males) were included. The mean age was 58 years (range 25–80 years). A total of 88 lesions were evaluated; there were 64 (73%) gliomas, 19 (21.6%) metastases, and 5 (5.7%) other tumors, mostly located in the frontal (41%) and temporal (27%) lobes. Most of the tumors (75%) were perfectly visible on ioUS (grade 3, Mair grading system), except for those located in the insular lobe (grade 2). The regression model showed a nearly perfect correlation (R2 = 0.97, p < 0.001) between preoperative tumor volumes from both MRI and nioUS. Ultrasonographic visibility significantly influenced this correlation, which was stronger for highly visible (grade 3) tumors (p = 0.01). For residual tumors, the correlation between postoperative MRI and nioUS was weaker (R2 = 0.78, p < 0.001) but statistically significant. The Bland-Altman analysis showed minimal bias between the two techniques for pre- and postoperative scenarios, with statistically significant results for the preoperative concordance. CONCLUSIONS The authors’ findings show that most brain tumors are well delineated by nioUS and almost perfectly correlated with MRI-based measurements both pre- and postoperatively. These data support the hypothesis that nioUS is a reliable intraoperative technique that can be used for real-time monitoring of brain tumor resections and to perform volumetric analysis of residual disease.