Enrico Martin

129 Diagnostic accuracy of non-invasive tests lesions rises. Subgroup analyses in this study should however be interpreted with caution as many were performed on a small number of studies. Most studies were also retrospective of nature further diminishing quality of evidence. Despite these limitations using a Bayesian approach, the quantification of diagnostic accuracy and uncertainty of common MRI and FDG-PET characteristics were reliable even when total number of studies or patients was small and there was heterogeneity in thresholds. 26 Unfortunately many features of interest, such as delayed scanning in FDG-PET and functional MRI are thus far infrequently studied which excluded them from meta-analyses. Yet these features seem promising, possibly providing higher accuracies compared to features analyzed in meta-analyses. Based on the findings of this study, future research should investigate several knowledge gaps. First, the MRI characteristics found in this study should be validated in a large series of patients to distinguish a patient group at high- risk for malignant transformation which minimizes the need for further diagnostics in low-risk patients. Second, only symptomatic or growing lesions should undergo imaging. The value of cystic changes, heterogeneity on T1 and T2 weighted images, large tumor size, and intratumoral lobulation should be studied for additional value too. DWI and ADC imaging seem of interest as well and might be of particular interest in the sporadic patient population. Schwannoma’s are the most common form of BPNST in sporadic patients and cannot be reliable distinguished on FDG-PET as schwannomas commonly have high levels of FDG uptake. 70 Also, schwannoma’s with cystic changes are common (ancient schwannoma’s) and may exhibit heterogeneous features. 31 MRI characteristics need to be assessed between sporadic and NF1 patients to explore possible variations in diagnostic accuracy which may necessitate different diagnostic guidelines. In NF1 the use of a SUVmax threshold of ≥3.5 should be replicated in a large database of patients who underwent scans that adhere to EARL criteria. Additionally, late scanning and other semiquantitative parameters should be evaluated in the same population to find one with higher specificity. Altogether, these findings may enable proper diagnostic algorithms to arise for evaluating MRI scans and using distinct threshold values of FDG-PET characteristics in NF1 populations. This way unnecessary imaging, biopsies, and harmful resections will diminish. In sporadic patients, suspect lesions should then undergo biopsy based on MRI findings. In NF1 patients, suspect lesions should be evaluated with additional FDG-PET imaging. Lesions with SUVmax >3.5 or high T/L ratio should have a PET-guided biopsy. Whenever biopsies of suspect lesions are negative one may consider nerve-sparing resection or a wait-and-scan approach. Furthermore, the use of radiomics and deep learning has not yet been studied in nerve sheath tumors, but may be useful when studies are performed correctly including sufficient MPNST images. It may even help stratifying low and high-grade MPNSTs. 71–73 The search for an ideal liquid biopsy should be stimulated as well since its use may diminish the need for FDG-PET scans and decrease radiation exposure in the NF population who is already prone to tumorigenesis. 6