Let \(\mathcal {G}_1, \mathcal {G}_2\) be subsets of \(\mathcal {L}(\textbf{X})\) that contain all rank-one idempotent and nilpotent operators. For \(\varepsilon > 0\) and \(\textbf{T} \in \mathcal {L}(\textbf{X})\) , we denote by \(\sigma _{\varepsilon }(\textbf{T})\) and \(r_{\varepsilon }(\textbf{T})\) the pseudospectrum and the pseudospectral radius of \(\textbf{T}\) , respectively. In this work, we characterize the structure of surjective mappings \(\varphi : \mathcal {G}_1 \rightarrow \mathcal {G}_2\) satisfying \( r_\varepsilon (\varphi (\textbf{T}_1)\varphi (\textbf{T}_2)\varphi (\textbf{T}_1)) = r_\varepsilon (\textbf{T}_1 \textbf{T}_2 \textbf{T}_1) \quad {\text {or}} \quad \sigma _\varepsilon (\varphi (\textbf{T}_1)\varphi (\textbf{T}_2)\varphi (\textbf{T}_1)) = \sigma _\varepsilon (\textbf{T}_1 \textbf{T}_2 \textbf{T}_1), \) for all \(\textbf{T}_1, \textbf{T}_2 \in \mathcal {G}_1\) . Analogous descriptions are also established in the finite-dimensional setting without requiring the surjectivity of \(\varphi \) .