$\newcommand{\ko}{\mathit{ko}}
\newcommand{\MTSpin}{\mathit{MTSpin}}
\newcommand{\Z}{\mathbb Z}$
As Mike points out in his comment, it's not obvious what it means for a bordism invariant is
“topological” or “geometric.” The bordism invariants you mention can be described
topologically, and some bordism invariants which you might think of as topological also admit geometric
descriptions.
Pontrjagin numbers are oriented bordism invariants which look topological: take a cohomology class on your manifold
and evaluate it on the fundamental class. For example, there's an injective map $\Omega_4^{\mathrm{SO}}\to\mathbb
Z$ sending a closed, oriented $4$-manifold $X$ to $\langle p_1(X), [X]\rangle$; here $p_1$ is the first Pontrjagin
class of $X$.
However, there is an equivalent, “geometric” definition of this invariant: choose a connection on the
vector bundle $TX\to X$, and let $F$ be its curvature. Then one can make sense of $\mathrm{tr}(F\wedge
F)\in\Omega^4(X)$, and Chern-Weil theory proves that
$$
-\frac{1}{8\pi^2}\int_X \mathrm{tr}(F\wedge F) = \langle p_1(X), [X]\rangle.
$$
Certainly a connection is geometric data, so this invariant is both “topological” and
“geometric.”
The pin$^\pm$ bordism invariants you mention admit geometric interpretations (via $\eta$-invariants), but also
topological ones, though the topology is harder to see. First, let's reframe the above bordism invariant in terms
of Thom spectra: the characteristic class $p_1\in H^4(B\mathrm{SO})$ defines via the Thom isomorphism a cohomology
class in $\tilde H^4(M\mathrm{SO})$, hence a map $M\mathrm{SO}\to\Sigma^4 H\mathbb Z$, and upon taking $\pi_4$, we obtain
the map $\Omega_4^{\mathrm{SO}}\to\mathbb Z$ from above.
The point of this reformulation is that by replacing $H\mathbb Z$ (i.e. ordinary cohomology) with other cohomology
theories, we can describe the invariants you've mentioned above.
- As a warm-up, take the Arf invariant of a spin surface, which defines an isomorphism $\Omega_2^{\mathrm{Spin}}\to\mathbb
Z/2$. There are several different ways to define it, but here's one in line with our above description of $p_1$:
we have the Atiyah-Bott-Shapiro map $\mathit{ABS}\colon\MTSpin\to\ko$,1 and upon taking $\pi_2$, this
yields a map $\Omega_2^{\mathrm{Spin}}\to \pi_2\ko\cong\mathbb Z/2$. One can unwind this definition through the
Pontrjagin-Thom isomorphism and obtain a description of the Arf invariant through integration of $\ko$-cohomology
classes.
- Next, the Arf-Brown-Kervaire invariant. There is a splitting $\mathit{MTPin}^-\simeq\MTSpin\wedge
\Sigma^{-1}\mathit{MO}_1$, where $\mathit{MO}_1$ is the Thom spectrum of the tautological bundle $\sigma\to
B\mathrm O_1$. Therefore, smashing the Atiyah-Bott-Shapiro map with $\Sigma^{-1}\mathit{MO}_1$, we obtain a map
$$
\mathit{MTPin}^-\simeq\MTSpin\wedge \Sigma^{-1}\mathit{MO}_1\longrightarrow \ko\wedge\Sigma^{-1}\mathit{MO}_1.
$$
Taking $\pi_2$, we get a map
$$\Omega_2^{\mathrm{Pin}^-}\to \pi_2(\ko\wedge \Sigma^{-1}\mathit{MO}_1)\cong
\widetilde{\ko}_3(\mathit{MO}_1)\cong\Z/8,$$
and this is the Arf-Brown-Kervaire invariant.2 This can also be described in terms of a pushforward in
twisted $\ko$-theory.
- The same approach works $\Omega_4^{\mathrm{Pin}^+}\to\Z/16$, using this time the splitting
$\mathit{MTPin}^+\simeq\MTSpin\wedge\Sigma\mathit{MTO}_1$; here $\mathit{MTO}_1$ is the Thom spectrum of the
virtual bundle $-\sigma\to B\mathrm O_1$. Smashing the Atiyah-Bott-Shapiro map with $\Sigma\mathit{MTO}_1$ and
taking $\pi_4$ yields a map $\Omega_4^{\mathrm{Pin}^+}\to \widetilde{\ko}_3(\mathit{MTO}_1)\cong\Z/16$.
- The same approach works for $\Omega_2^{\mathrm{Pin}^-}(B\Z/2)\to\Z/4$:
$$\Omega_2^{\mathrm{Pin}^-}(B\Z/2)\cong\pi_2(\MTSpin\wedge \Sigma^{-1}\mathit{MO}_1\wedge (B\Z/2)_+),$$
which maps to $$\pi_2(\ko\wedge \Sigma^{-1}\mathit{MO}_1\wedge (B\Z/2)_+) = \widetilde{\ko}_3(\mathit{MO}_1\wedge
(B\Z/2)_+)\cong\Z/4.$$
If you're asking which bordism invariants come from cohomology classes, the answer is that characteristic classes
for $n$-dimensional $G$-bordism live in $H^n(BG;A)$, where $A$ is a coefficient group; a $G$-structure on a
manifold $M$ pulls back this class to $M$, and then we evaluate it on the fundamental class. In general, this won't
capture everything: for example, if two $G$-structures have the same underlying orientation, their values on any
cohomological bordism invariant will agree. Hence, for example, the Arf invariant isn't cohomological, as there are
different spin structures on a torus which induce the same orientation, but have different Arf invariants. This is
why topological descriptions of such bordism invariants use generalized cohomology.
1: Here, given a $G$-structure $X\to B\mathrm O$, $\mathit{MTG}$ means the Thom spectrum whose homotopy groups are
the bordism groups of manifolds with a $G$-structure on the tangent bundle, rather than on the stable normal
bundle. This distinction is irrelevant for spin bordism, but important for pin$^\pm$ bordism.
2: The Arf-Brown-Kervaire invariant depends on a choice of an 8th root of unity; depending on this choice, we might
need to compose with an automorphism of $\Z/8$ to obtain “the” Arf-Brown-Kervaire invariant. The same
caveat applies to the $\Z/16$ and $\Z/4$ invariants.
