We identified 171 patients of medulloblastoma with pre-operative imaging and known molecular subgroup affiliation at our institute between 2007 and 2018 that comprise the present study cohort. Baseline patient, disease, and treatment-related characteristics across individual molecular subgroups are summarized in Table 1. Briefly, the median age of the study cohort was 9 years with an inter-quartile range (IQR) of 5-16 years with male gender preponderance (3.6:1 ratio). SHH was the most prevalent molecular subgroup (n=60, 35%) in our study cohort, driven by large number of adult medulloblastomas (n=40), followed by Group 4 (n=42, 24.5%), Group 3 (n=35, 20.5%), and WNT-pathway (n=34, 20%) medulloblastoma respectively. The distribution of various semantic MRI features across the four molecular subgroups is described in online supplementary Table S1.
Clinical outcomes
By the time of this analysis, 55 patients had experienced disease recurrence/progression resulting in a total of 44 deaths, including one death due to treatment-related toxicity (without evidence of disease). The median time to relapse (recurrence/progression) was 19 months (IQR= 10-33 months) for the entire study cohort but was significantly different across molecular subgroups. Only one patient with WNT-pathway medulloblastoma had documented recurrence/progression at 9-months from index diagnosis. Time to relapse was significantly longer (p=0.015) in patients with SHH (median 25 months, IQR=12-53 months) and Group 4 (median 23 months, IQR= 16-42 months) medulloblastoma compared to Group 3 tumors (median 10 months, IQR=5-20 months). The pattern of failure was also quite different across molecular subgroups with local tumor-bed recurrence being the predominant mode of first failure in SHH-pathway medulloblastoma, while neuraxial failure with leptomeningeal dissemination being more common in Group 3 and Group 4 tumors (online supplementary Table S2). At a median follow-up of 45 months (IQR=19-65 months), 5-year Kaplan-Meier estimates of RFS and OS with 95% confidence interval (CI) were 64% (95%CI:60-69%) and 71% (95%CI: 67-75%) respectively for the entire study cohort (online supplementary file S3). As expected, molecular subgrouping demonstrated prognostic significance for RFS (p<0.001) and OS (p<0.001) on stratified Kaplan-Meier analysis, with WNT-pathway medulloblastoma having the best outcomes while Group 3 medulloblastoma showing the worst survival (online supplementary file S4).
Prognostic impact of MRI features
Univariate analysis demonstrating prognostic impact of specific semantic imaging features on RFS and OS in all patients across molecular subgroups as well as within induvial subgroups is summarized in Tables 2 and 3, respectively. Six semantic imaging features significantly impacted upon outcomes - non-central tumor location on vertical axis, absence of brainstem involvement, ≤80% solid tumor area with contrast-uptake, heterogenous pattern of contrast-enhancement, presence of necrosis, and calcification - were all associated with significantly inferior RFS on univariate analysis (online supplementary file S5). Given that only a single WNT patient had experienced relapse, specific imaging features within WNT-subgroup medulloblastoma could not be further correlated with RFS. Within SHH-subgroup medulloblastoma, only non-central tumor location on vertical axis predicted significantly inferior RFS. In Group 3 medulloblastoma, five imaging features - non-central tumor location on vertical axis, absence of brainstem involvement, ≤80% solid tumor area with contrast-uptake, presence of necrosis, and calcification - were all associated with significantly worse RFS. Heterogenous pattern of contrast-enhancement was the only factor predicting significantly inferior RFS within Group 4 medulloblastoma. Similarly, six semantic imaging features - non-central tumor location on vertical axis, absence of brainstem involvement, heterogenous pattern of contrast-enhancement, T2-weighted heterogeneity, presence of necrosis, and calcification - were all associated with worse OS on univariate analysis (Figure 1). No deaths were encountered in WNT-subgroup medulloblastoma precluding any correlation of specific imaging features with survival. Within SHH-subgroup medulloblastoma, only non-central tumor location on vertical axis predicted significantly inferior OS. In Group 3 tumors, six semantic imaging features - non-central tumor location on vertical axis, absence of brainstem involvement, ≤80% solid tumor area with contrast-uptake, infiltrative tumor margin, presence of necrosis, and calcification - were all associated with significantly worse OS. Heterogenous pattern of contrast-enhancement and T2-weighted heterogeneity predicted inferior OS within Group 4 medulloblastoma.
Multivariate Cox regression analysis of semantic imaging features for RFS and OS in all patients across all four molecular subgroups is described in Table 4. For the entire study cohort, three imaging features - tumor location on vertical axis (p=0.029), brainstem involvement (p=0.031), and calcification (p=0.013) were associated with independent prognostic significance for RFS. The same three imaging features - tumor location on vertical axis (p=0.011), brainstem involvement (p=0.022), and calcification (p=0.021) were also found to be significant and independent prognostic factors for OS. None of the semantic MRI features were significant within WNT-pathway medulloblastoma and only one imaging feature was significant within SHH-subgroup medulloblastoma on univariate precluding any multivariate analysis, which was limited to Group 3 and Group 4 tumors. For Group 3 tumors, percentage of solid tumor area with contrast-uptake demonstrated independent statistical significance for RFS (p=0.009) and OS (p=0.02), respectively, while calcification (p=0.049) achieved statistical significance for RFS but not OS. None of the imaging features that correlated significantly with RFS and/or OS in Group 4 medulloblastoma on univariate analysis demonstrated any independent prognostic impact on Cox regression analysis.