Simple Summary
Distinguishing between local relapse and damage from radiation-induced necrosis after brain radiotherapy remains a significant challenge. This research aims to improve the accuracy of diagnostic imaging, specifically perfusion MRI. The study evaluates new ways to analyze perfusion MRIs, integrating a technique that looks at the entire affected area in three dimensions and an automated method for setting measurement thresholds. By finding more reliable ways to interpret these images, the authors hope to help clinicians make more precise diagnoses. This could lead to better-tailored treatments for patients, preventing unnecessary or inappropriate therapies and ultimately improving patient outcomes.
Abstract
Background: Distinguishing tumor recurrence from radiation necrosis after radiotherapy for brain metastases remains a major diagnostic challenge. Perfusion MRI, particularly the measurement of relative cerebral blood volume (rCBV), is a commonly used technique to differentiate between these two entities. However, variations in the placement of regions of interest (ROIs) affect diagnostic accuracy. This study compares the diagnostic performance of different cerebral perfusion methods, including a novel volumetric 3D ROI method and automatic thresholding, to differentiate tumor recurrence from radiation necrosis. Methods: We retrospectively analyzed data from 23 patients, including 25 brain metastases treated with stereotactic radiotherapy, who were suspected of local recurrence and had histological confirmation via biopsy or surgical resection. Each patient underwent perfusion MRI before surgery. The diagnostic performance of the different ROI methods (manual and 3D) was evaluated using the area under the ROC curve (AUC), as well as sensitivity and specificity measures. An automatic thresholding method was also applied, generating tumor sub-volumes with predefined cut-off values to determine the rCBV threshold most specific for differentiating relapse from necrosis. Results: The 3D ROI method, considering the whole lesion and a healthy ROI in the head of the caudate nucleus, demonstrated superior diagnostic performance (AUC = 0.65), outperforming manual methods (AUC = 0.53). Robustness was moderate, with an intraclass correlation coefficient of 0.60 between Syngo.via and IntelliSpace. Conclusions: The 3D ROI method shows promise in improving diagnostic accuracy in distinguishing tumor recurrence from radiation necrosis. Further studies with standardized protocols and larger populations are needed to validate these results.