Given a topological space $X$ and a path $p : [0, 1] \to X$ such that $p(0) \neq p(1)$, does there always exist an injective path $p' : [0,1] \to \operatorname{im}p$ such that $p'(0) = p(1)$ and $p'(1) = p(1)$? For intuitive examples of paths we could "cut out" the loops which occur in $p$, but are there pathological examples of paths in which this technique cannot be applied? Otherwise, is there a Theorem that guarantees the existence of $p'$, even for only certain "nice" classes of paths $p$?
For any $X$ that satisfies this property, any subspace of $X$ is path-connected if and only if every two points are joined by an injective path.
What about the special case where $X = \mathbb{R}^n$?