Enrico Martin

118 Chapter 6 Results After removal of duplicates, a total of 8463 citations were identified in PubMed and Embase databases. One hundred potentially relevant articles were selected through title/abstract screening. After full-text screening 43 studies were selected for qualitative synthesis ( Figure 1 ). These studies included data on 1875 patients and 2939 lesions. Amongst the included studies were 12 studies on MRI characteristics, 21 studies on FDG-PET characteristics, 7 studies on both MRI and FDG-PET, and 3 studies on liquid biopsies ( Table 1 ). Twenty-eight studies included NF1 patients only. In the remaining studies the percentage of NF1 patients ranged from 12-65%. The proportion of MPNST compared to BPNST varied from 2:1 to 1:29. Thirty-five studies were included for quantitative synthesis. Diagnostic accuracies of characteristics not included in quantitative synthesis are shown in Supplementary Table 2 . Conventional MRI characteristics Sixteen studies describing a total of 12 conventional MRI characteristics were included for quantitative synthesis. 16,21,36–41,22,29–35 These studies included a total of 1041 tumors in 925 patients (48% NF1). Eight studies included in meta-analyses were at high risk of bias, mainly due to the inclusion of histologically confirmed lesions only or the exclusion of patients who had received treatment prior to imaging ( Supplementary File 1 ). 21,22,29,30,33,34,36,41 Nine studies reported on ill-defined margins. 22,30,32,34–36,38–40 Pooled sensitivity and specificity were 0.52 (95% CI: 0.40-0.65) and 0.94 (95% CI: 0.88-0.98) respectively. Pooled pLR was 11.03 (3.83-31.62) and nLR was 0.51 (0.36-0.66). The forest plot and the 95% prediction region in the SROC plot ( Figure 2 & Supplementary Figure 1 ) demonstrated moderate heterogeneity between studies. Sensitivity was higher in studies with a smaller total sample of lesions ( Supplementary Table 3 ). Specificity was lower in studies with a higher proportion of MPNSTs, those that included symptomatic lesions only or histologically proven lesions only. Five studies reported on perilesional edema. 22,29,38–40 Pooled sensitivity and specificity were 0.65 (95% CI: 0.38-0.87) and 0.95 (95% CI: 0.83-1.00) respectively. Pooled pLR was 3415.18 (3.15-5948.77) and nLR was 0.38 (0.12-0.69). There was moderate heterogeneity between studies. Sensitivity was higher in studies with a smaller proportion of MPNST and when only NF1 patients were included and was lower when only histologically proven lesions were included.