Writing a type trait for detecting Matrix Expressions in Eigen

I am trying (and failing) to write a type trait that detects Eigen expressions. In other words, I'd like to be able to detect things like A * A + B etc. where A and B are Eigen matrices/vectors. Currently I'm doing this:

template<typename T>
struct is_matrix_expression : std::false_type
{
};

template<typename Derived> // specialization
struct is_matrix_expression<typename Eigen::MatrixBase<Derived>> :
        std::true_type
{
};

Note that Eigen::MatrixBase<Derived> is a (template) base for all possible Eigen expressions (such as decltype(A * A + B) etc). However, the general template is being picked out, as it is a better match for something like decltype(A * A + B), and not the MatrixBase<Derived> specialization.

How can I somehow enforce the specialization to be picked out? Or, in other words, enable the specialization for all possible children of Eigen::MatrixBase<Derived>? I played a bit with SFINAE on std::is_base_of, but that requires an explicit type and not a template where the type of the expression (in this case Derived) is not known in advance.