A radiomic signature based on combined [¹⁸F]-DOPA PET and MRI to differentiate progression from radiation necrosis in patients with gliomas: a pilot study

Aim: Positron Emission Tomography (PET) imaging with radiolabeled amino acids can help in differentiating delayed complications of radiotherapy, such as cerebral radionecrosis, from true progression, but remains challenging. Recently, radiomic have opened new perspectives for image analysis by mining data undetectable from naked eyes from images to provide more accurate diagnoses. The aim of this pilot study was 1) develop a data processing pipeline in order to 2) identify a radiomic signature from combined L-3,4-dihydroxy-6-[¹⁸F]fluoro-phenyl-alanine ([¹⁸F]-DOPA) and Magnetic Resonance Imaging (MRI) to differentiate cerebral radionecrosis from true progression. Method: Twenty-nine [¹⁸F]-DOPA PETs and MRIs in twenty-two patients with histopathologically-proven gliomas were retrospectively analysed after inconclusive MRI to differentiate cerebral radionecrosis from progression. Reference standard was obtained histopathologically whenever available or based on clinical and imaging follow-up. After delineation of tumor volumes, 95 radiomic features were extracted from PETs static images (acquired from 10 to 30 min after injection) and MRIs (on contrast-enhanced-T1/T1 images). The most discriminant features were selected with a random forest algorithm and used to create a radiomic signature to predict the tumoral status ('viable tumor' vs. 'no viable tumor'). Results: Out of the 29 lesions analysed, 23 were concluded as 'viable tumor' and 6 as 'no viable tumor' according to the reference standard. Discriminant features selected were: PET.Compacity, PET.std and PET.GLRLM-LRLGE. No feature was selected from MRI images. A radiomic signature based on those three features could differentiate 'viable tumor' from 'no viable tumor' with high diagnosis performance (AUC = 1), even in regards to the usual T/N ratio used in routine (AUC 0.87; p=0.78). Conclusion: In this study, we established a data processing pipeline to identify a radiomic signature from combined [¹⁸F]-DOPA PET and MRI. In regards to the first results, radiomic analysis might help to differentiate cerebral radionecrosis from true progression. A larger study will be performed including an external validation set. (author)