Assessment of immunotherapy response in intracranial malignancy using semi-automatic segmentation on magnetic resonance images

OBJECTIVE: To explore multi-aspect radiologic assessment of immunotherapy response in intracranial malignancies based on a semi-automatic segmentation technique, and to explore volumetric thresholds with good performance according to RECIST 1.1 thresholds. METHODS: Patients diagnosed with intracranial malignancies and treated with immunotherapy were included retrospectively. In all MR images, target lesions were measured using a semi-automatic segmentation technique that could intelligently generate visual diagrams including RECIST 1.1, total volume, and max. 3D diameter. The changes in parameters were calculated for each patient after immunotherapy. The ROC curve was used to analyze the sensitivity and specificity of the size change of the legion. This was useful to find new volumetric thresholds with better efficiency in response assessment. The changes in total volume were assessed by conventional volumetric thresholds, while RECIST 1.1 thresholds were for the max. 3D diameter. A chi-square test was used to compare the concordance and diagnostic correlation between the response assessment results of the three criteria. RESULTS: A total of 20 cases (average age, 58 years; range, 23 to 84 years) and 58 follow-up MR examinations after immunotherapy were included in the analysis. The P-value of the chi-square test between RECIST 1.1 and total volume is 0 (P <0.05), same as that in RECIST 1.1 and max. 3D diameter. The kappa value of the former two was 0.775, and the kappa value for the latter two was 0.742. The above results indicate a significant correlation and good concordance for all three criteria. In addition, we also found that the volumetric assessment had the best sensitivity and specificity for the immunotherapy response in intracranial malignancies, with a PR threshold of −64.9% and a PD threshold of 21.4%. CONCLUSIONS: Radiologic assessment of immunotherapy response in intracranial malignancy can be performed by multiple criteria based on semi-automatic segmentation technique on MR images, such as total volume, max. 3D diameter and RECIST 1.1. In addition, new volumetric thresholds with good sensitivity and specificity were found by volumetric assessment.