Page 50 - Read Online

Dasgupta et al. J Transl Genet Genom 2018;2:15. I https://doi.org/10.20517/jtgg.2018.21 Page 11 of 15

medulloblastoma (n = 36) as having no or minimal enhancement as opposed to 7% for other subgroups.
[24]
Similarly, Lastowska et al. reported 6 of 10 (60%) group 4 tumors with weak or no enhancement. In a
[28]
large study of adult medulloblastoma, Zhao et al. , reported minimal or no enhancement in 22 of 44
[29]
(50%) patients with group 4 tumors. Mata-Mbemba et al. reported minimal enhancement of the primary
[31]
tumor in 55% of patients with group 4 medulloblastoma (n = 49). Dasgupta et al. reported minimal or no
enhancement in 30% of group 4 medulloblastoma (n = 23). Even for tumors showing partial contrast uptake,
intervening non-enhancing areas gave rise to a “patchy” enhancement pattern.

Other imaging features
[25]
The MRS profile of group 4 tumors is quite similar to group 3, but significantly different from SHH-
subgroup medulloblastoma, as described earlier. Similar to group 3 medulloblastoma, the majority (70%)
[31]
of group 4 tumors do not show any peri-tumoral edema . The location, morphology, and imaging
characteristics of metastases from group 4 medulloblastoma are quite distinct [29,30] . Metastatic deposits from
group 4 tumors are more nodular and associated with a “mismatching pattern”, i.e., lack of or minimal
contrast enhancement but restricted diffusion. The presence of metastases in the suprasellar region and/or
infundibular recess of the third ventricle is a highly specific marker of group 4 medulloblastoma [29,30] .

PREDICTORS OF MOLECULAR SUBGROUPS
The presence of subgroup-specific imaging features has prompted researchers to develop models and
nomograms for accurate pre-operative prediction of molecular subgroups in medulloblastoma. In a large
[28]
cohort involving 125 patients , researchers from Beijing Tian Tan Hospital in China reported location
(midline vermis/IVth ventricle, cerebellar hemisphere, and CPA; P < 0.0001) and pattern of enhancement;
P = 0.0048) to be independent predictors of molecular subgrouping in adult medulloblastoma. With the
logistic regression model based on location and pattern of enhancement, 79% of adult medulloblastomas
were accurately and appropriately classified, including WNT (24%), SHH (86%), and group 4 (91%)
medulloblastoma (R-squared goodness of fit = 0.669). The accuracy of the logistic regression model improved
significantly after incorporating anatomic localization patterns (horizontal location, brainstem contact and
vertical location) with pattern of enhancement. With the modified model, 92% of adult medulloblastomas
could be accurately and appropriately classified, including WNT (65%), SHH (95%), and group 4 (98%)
medulloblastoma (R-squared goodness of fit = 0.795).

Using a step-wise, multi-variable, multi-nomial, logistic regression model in a large single-institution cohort
[29]
(n = 119), researchers from Hospital for Sick Children at the University of Toronto , demonstrated the
lateralized cerebellar location for SHH-subgroup with an adjusted odds ratio (aOR) of 9 (P < 0.0001); minimal
enhancement of primary tumor for group 4 medulloblastoma (aOR = 5.2, P < 0.0001); CPA location for WNT-
pathway medulloblastoma (aOR = 1.4, P < 0.03); ependymal metastases with mismatching pattern for group 4
tumor (aOR = 2.8, P < 0.001); and spinal leptomeningeal metastases for group 3 tumors (aOR = 1.9, P < 0.01)
were independent predictors of molecular subgrouping. Specifically, the presence of a metastasis in the third
ventricular infundibular recess showing mismatching pattern was significantly associated with group 4
disease (P < 0.02).

[31]
In parallel, researchers from Tata Memorial Centre in India recently reported on the radiogenomics of
medulloblastoma in their single-institution cohort of 111 patients with known molecular subgroup affiliation.
Amongst their panel of 19 pre-specified imaging features, 11 were differentially distributed across the four
subgroups with statistical significance (P < 0.05) on univariate analysis. Two-thirds of patients (n = 76)
were chosen randomly from individual subgroups to form the training cohort (TC), while the remaining
one-third (n = 35) constituted the validation cohort (VC). Multi-nomial logistic regression analysis was
performed in the TC to identify few imaging features with highest discrimination of one subgroup from
the other three subgroups to construct subgroup-specific binary nomograms. Using receiver operating