P10.14.A DISTINCTION OF RADIONECROSIS AND TUMOR RECURRENCE BASED ON ANATOMICAL MRI COMPUTATIONAL ANALYSIS: A PROOF-OF-CONCEPT

Abstract BACKGROUND Most common malignant brain tumors are metastases (BM) and glioblastoma (GBM). Stereotactic and fractionated radiotherapy are modalities of choice for treatment of these malignancies. As a delayed complication of these treatments, radiation necrosis can occur several months and even years after the completion of radiotherapy. Distinguishing between tumor recurrence and radionecrosis solely through conventional MRI and functional MRI can be challenging at times. This challenge can result in delays in clinical decision-making or require utilization of additional imaging modalities such as PET for more accurate diagnosis. The aim of this study was to distinguish radionecrosis from tumor progression of post-radiotherapy brain tumors (BM and GBM) thanks to computer analysis comparing follow-up MRI exam to initial MRI exam, with DOPA PET as a gold standard. Developed method is a classification task at the patient level. MATERIAL AND METHODS This retrospective study included 16 patients with GBM and BM, whom had total of 17 lesions. Lesions were split into 3 subsets: 5 lesions of radionecrosis, 10 lesions of tumor progression, 2 lesions with mixed radionecrosis and progression. We analyzed 2 time points: baseline M0 exam - the first time showing enhancing part of the lesion on a post-radiotherapy MRI exam; follow-up M1 exam - MR time point exam on which the lesion clearly showed expansion. DOPA PET MRI was done at the time of M1 exam and served as a gold standard. Specific preprocessing steps were implemented. Firstly, automatic lesion segmentation and elastic co-registration were done, where M2 was aligned to M1 images. Then, rigid co-registration of M2 and DOPA was applied to the M1 images, after which center of the lesion was computed. The analysis of deformation vector field was performed. On the slice showing the maximum area of the lesion, the radius (mm) from the centre of mass of the lesion mask for M0 and M1 was calculated, and the absolute and relative variations for each 360º angle between M0 and M1 were studied and compared with the mean DOPA uptake per angle (Bq/ml). RESULTS Tumor recurrence and radionecrosis can be distinguished thanks to their growth patterns. The mean standard deviation on angular variation showed lower values for radionecrosis (4.69, IQR 3.37 - 4.49) compared to tumor progression (11.72, IQR 6.3 - 13.54), which goes in concordance with DOPA uptake on PET exam: lower in radionecrosis (4459.43 IQR 4466.55 - 4804