Abstract
BACKGROUND Post-treatment radiological deterioration of patients with an irradiated high-grade (WHO grade 3 and 4) glioma (HGG) may be the result of true progressive disease (PD) or treatment-induced effects (TIE). Differentiation between these two entities is of great importance, but remains a diagnostic challenge. This study assesses the diagnostic value
... read more
of conventional MRI characteristics to differentiate PD from TIE in treated HGGs. MATERIAL AND METHODS In this single-centre, retrospective cohort study, we included adult patients with a HGG, who were treated with radiotherapy and subsequently developed a new or increasing contrast-enhancing lesion on conventional follow-up MRI. TIE and PD were defined radiologically as stable/decreased for a minimum of six weeks or progressive according to the RANO criteria, and histologically as predominantly TIE without viable tumour or PD. Demographic and clinical data were retrieved. Twenty-one preselected MRI characteristics of the progressive lesions were assessed by two neuroradiologists. The statistical analysis included logistic regression to develop a) a full multivariable model b) a diagnostic model with model reduction, and a Cohen’s Kappa interrater reliability coefficient. RESULTS 210 patients (median age 61, IQR=54–68, 189 males) with 284 lesions were included, of which 141 (50%) had PD. Median time to PD was 2 (0.7–6.1) and to TIE 0.9 (0.7–3.5) months after RT. In multivariable modelling and after model reduction, the following determinants were significant diagnostic factors: Radiation dose (Odds ratio (OR)=0.68, p=0.017), longer time since radiotherapy (OR=3.56, p<0.0005), certain enhancement patterns (soap bubble enhancement: OR=2.63, p=0.003), isointense apparent diffusion coefficient-signal (OR=2.11, p=0.036), development of multiple new lesions (OR=1.68, p=0.088) and increased marginal enhancement (OR=2.04, p=0.027). ORs of >1 indicate higher odds of PD. The Hosmer & Lemeshow test showed a good calibration (p=0.947) and the area under the ROC-curve was 0.722 (95%-CI=0.66–0.78). Interrater reliability analysis between neuroradiologists revealed moderate to near-perfect agreement for the significantly predictive items, but poor agreement for others. CONCLUSION In patients with irradiated high-grade gliomas, several characteristics from conventional MRI are significant predictors for the discrimination between true progression and treatment-induced effects. Interrater reliability for these characteristics was variable. Conventional MRI characteristics from this study should be incorporated into a multimodal diagnostic model that includes advanced imaging techniques.
show less