Part I: Conformal Mappings
Mathematical Foundations of Contemporary Scientific Research
July 27, 2026
Arc of solid-state electronic detectors directly opposite the X-ray tube measures reduced photon intensity emerging through tissue.
Logarithmic ratio $\ln(I_0 / I)$ yields the line integral of total attenuation along ray path $L_{\theta,s}$.
Reconstructs the continuous 2D slice density $\mu(x,y,z_k)$ at slice position $z_k$ from 1D ray projections $p(\theta,s)$ for $\theta \in [0,\pi)$:
Unfiltered back-projection creates $1/r$ spatial blurring. High-pass ramp filtering $h(s)$ (Riesz transform) cancels low-frequency artifacts, recovering sharp tissue boundaries.
Radon Inversion, the Fourier Slice Theorem, and iterative CT reconstruction form a dedicated full lecture on inverse problems. In this course, we treat FBP as an established upstream provider of the 3D scalar field.
The continuous attenuation field $\mu(x,y,z)$ is sampled at discrete voxel lattice locations $(x_i, y_j, z_k) \in \mathbb{Z}^3$, defining the discrete scalar field mapping:
Values are calibrated into Hounsfield Units (HU) relative to water ($\text{HU}=0$) and air ($\text{HU}=-1000$):
High-field superconducting magnet produces static magnetic field $\vec{B}_0$ (1.5T or 3.0T) along the longitudinal $Z$-axis, aligning nuclear magnetic moments of hydrogen protons ($^1\text{H}$) in tissue water.
Protons precess around $\vec{B}_0$ at the characteristic Larmor precession frequency:
Radiofrequency (RF) coils transmit an oscillating magnetic field $B_1(t)$ tuned precisely to the Larmor frequency $\omega_0$. Protons absorb resonant energy, tipping net magnetization vector $\vec{M}$ into the transverse $XY$-plane.
Creates a coherent, spinning transverse magnetization vector $M_{xy}$ that induces measurable RF signals in receiver coils.
Governs trajectory of magnetization vector $\vec{M}$ returning to thermal equilibrium:
Longitudinal $T_1$ (Spin-Lattice):
$M_z(t) = M_0(1 - e^{-t/T_1})$
Transverse $T_2$ (Spin-Spin):
$M_{xy}(t) = M_{xy}(0) e^{-t/T_2}$
Varying water binding in tissues creates unique $(T_1, T_2)$ parameters for gray matter, white matter, and CSF.
Gradient coils $(G_x, G_y, G_z)$ encode spatial positions into RF signal frequencies in $k$-space. 3D Inverse Fast Fourier Transform (IFFT) yields the discrete voxel grid mapping:
Reconstructed from 3D spatial frequency domain ($k$-space) data $S(k_x, k_y, k_z)$:
Anatomy is modeled as a continuous scalar function $\mu \colon \Omega \to \mathbf{R}_{\ge 0}$ over a compact domain $\Omega \subset \mathbf{R}^3$. Scanners partition $\Omega$ into closed cuboid voxel cells $V_{i,j,k} \subset \Omega$:
A diagnosing physician cannot interpret dense 3D scalar arrays directly. To isolate and visualize distinct anatomical structures—such as the colon wall, cerebral cortex, or tumor boundaries—we must extract explicit geometric surface representations called level sets. Because the true physical field $\mu$ is unobservable, we compute the level set $\widehat{\mathcal{M}}_c$ of the continuous interpolant $\tilde{\mu}$ as the surrogate proxy.
If $c$ is a regular value ($\nabla \mu(p) \neq 0, \forall p \in \mu^{-1}(c)$), the Preimage Theorem mathematically guarantees $\mu^{-1}(c)$ is a smooth 2D manifold. We computationally analyze level sets of $\mathcal{I}$ because interpolation of the discrete grid values $\mathcal{I}(i,j,k)$ provides the only evaluable manifold approximation $\widehat{\mathcal{M}}_c \approx \mu^{-1}(c)$.
Anatomical boundaries (e.g. brain cortex, colon wall) are curved 2D Riemannian manifolds $\mathcal{M} \subset \mathbf{R}^3$. Direct 2D planar projections cause severe angular shear, distorting local shapes and intersection angles. Non-conformal projections stretch circular functional regions into elongated ellipses, corrupting diagnostic metrics.
An isometric mapping preserves all geodesic distances exactly. The metric tensor under mapping remains identical to the identity matrix:
By Gauss's Theorema Egregium, a curved surface can be mapped isometrically to a flat plane if and only if its Gaussian curvature is zero ($K = 0$).
Since anatomical structures (e.g. cortex, colon) possess non-zero Gaussian curvature ($K \neq 0$), isometric flattening is physically impossible without introducing cuts or tears.
Conformal maps preserve local angles and map infinitesimal circles to circles. The metric tensor scales isotropically by a conformal factor $e^{2\lambda}$:
Unlike isometric maps, conformal mappings accommodate arbitrary curvature without introducing artificial cuts or tears. By preserving local inner angles, they prevent shape distortion in critical areas, making them ideal for mapping cortical sulci or colon folds.
Authalic mappings preserve local area elements exactly across the domain. The determinant of the metric tensor remains constant:
While authalic mappings are essential for counting cell densities or computing integrals over surfaces, they introduce severe angular shear.
This shear distorts local shape profiles (e.g. circles to ellipses), rendering them poor for shape-based visual diagnosis.
Extract a 2D surface $\mathcal{M}$ from the volume by defining an implicit level set:
where $c$ is a designated threshold separating distinct tissue types (e.g. bone vs. soft tissue) and $\tilde{\mu}$ is the continuous interpolant proxy of the discrete voxel intensities.
Holomorphic functions $f = u+iv$ are harmonic ($\Delta u = 0$). Critical points of Dirichlet energy $E_D(f) = \frac{1}{2} \int_{\mathcal{M}} |\nabla f|^2 dA$ yield the Laplace-Beltrami PDE $\Delta_{\mathcal{M}} f = 0$.
Evaluated over vertex $i$'s 1-ring neighborhood $\mathcal{N}(i)$ on a triangular mesh:
Pinning boundary vertices yields sparse linear system $Lu = 0$.
For a triangular face $T = (i,j,k)$, any scalar function $u$ is interpolated using hat basis functions $\phi_i$:
The gradient $\nabla \phi_i$ on triangle $T$ is constant and perpendicular to the opposite edge $jk$.
Integrating the inner product of basis gradients over triangle $T$ yields the cotangent of the interior angle $\gamma_k$:
Summing over the two triangles sharing edge $ij$ yields the cotangent weight $w_{ij} = \frac{1}{2}(\cot \alpha_{ij} + \cot \beta_{ij})$.
The Earth is modeled as a sphere $\mathbb{S}^2$ with constant Gaussian curvature $K=1$. The flat map lies in the Euclidean plane $\mathbf{R}^2$ with constant Gaussian curvature $K=0$.
Therefore, every map projection must distort distance, area, or angle. For maritime navigation, preserving local angles (course headings) is paramount, naturally requiring mappings to be strictly conformal. A different angle will take you nowhere!
Maps a point $P = (X, Y, Z)$ on the Riemann sphere $\mathbb{S}^2$ conformally to $z = x + iy$ on the complex plane $\mathbb{C}$. By intersecting the plane exactly at the equator, the lower hemisphere maps to the interior of the unit disk $\mathbf{D} = \{z \in \mathbb{C} : |z| < 1\}$.
The projection preserves angles but distorts distances, quantified by the scalar conformal factor: $ds^2_{\mathbb{S}^2} = \lambda(x, y) ds^2_{\mathbf{R}^2}$.
We have seen that conformal mappings naturally arise in both Medical Imaging and Cartography. By the Uniformization Theorem, any simply connected Riemann surface is conformally equivalent to either the Riemann Sphere $\hat{\mathbb{C}}$, the Complex Plane $\mathbb{C}$, or the Upper Half Plane $\mathbb{H}$.
For the remainder of this lecture, we will explicitly characterize the algebraic structure of all conformal automorphisms mapping these canonical domains to themselves.
In the context of the canonical domains ($\hat{\mathbb{C}}, \mathbb{C}, \mathbb{H}$), the geometric property of conformality is inextricably linked to complex analyticity.
Therefore, characterizing the conformal automorphisms of these spaces is mathematically identical to finding all their biholomorphisms (bijective holomorphic mappings).
Let $f: \mathbf{R}^2 \to \mathbf{R}^2$ be a differentiable mapping given by $f(x,y) = (u(x,y), v(x,y))$. Geometrically, preserving angles and orientation requires the Jacobian matrix $J_f$ to act purely as a scaled rotation at every point.
Substituting a single matrix $\begin{pmatrix} a & -b \\ b & a \end{pmatrix}$ for the scaled rotation yields a direct algebraic equivalence.
Because $u$ and $v$ satisfy the Cauchy-Riemann equations, the function $f(z) = u + iv$ is strictly holomorphic. The scaling factor $\lambda > 0$ ensures $f'(z) \neq 0$.
The conformal automorphisms of the projective space $\mathbf{P}^1 = \mathbf{C} \cup \{\infty\}$ are exactly the fractional linear transformations, also known as Möbius transformations.
This group is isomorphic to the projective special linear group $PSL(2, \mathbf{C})$. The condition $ad - bc \neq 0$ ensures the mapping is invertible and non-degenerate. Möbius transformations are uniquely determined by their action on any three distinct points in $\mathbf{P}^1$.
Any conformal automorphism of $\mathbf{P}^1$ is a meromorphic function globally defined on the extended complex plane. Because the mapping is bijective, it can possess exactly one pole, and this pole must be simple.
Let $f$ have a simple pole at $z_0 \neq \infty$. Its Laurent expansion is $f(z) = \frac{c}{z-z_0} + g(z)$, where $g(z)$ is entire because $f$ has no other poles. Since $f$ is bijective, $z_0$ is the unique pole, meaning $f$ has no pole at $\infty$. Consequently, $g(z)$ must be bounded as $z \to \infty$. By Liouville's Theorem, $g(z)$ is a constant $k$.
Thus, $f(z) = \frac{c}{z-z_0} + k = \frac{kz + (c - kz_0)}{z-z_0}$, which is exactly a Möbius transformation. If $z_0 = \infty$, the principal part is $cz$, and $f(z) - cz$ is an entire bounded function, thus a constant $d$, yielding $f(z) = cz + d$.
The conformal automorphisms of the complex plane $\mathbf{C}$ are the affine transformations.
This group is a subgroup of the full Möbius group. It contains all Möbius transformations that strictly fix the point at infinity. Geometrically, these mappings represent translations, rotations, and dilations.
Let $f \in \operatorname{Aut}(\mathbf{C})$. To preserve bijectivity, $f$ must be an entire function with a well-defined behavior at infinity. If $\infty$ were a removable singularity, $f$ would be bounded and thus constant by Liouville's theorem. If $\infty$ were an essential singularity, the Casorati-Weierstrass theorem dictates the image of $\{z : |z| > R\}$ is dense in $\mathbf{C}$. By the Open Mapping Theorem, the image of $\{z : |z| < R\}$ is a non-empty open set. Their intersection is therefore non-empty, violating injectivity.
Therefore, the isolated singularity at $\infty$ is strictly a pole. Because $f(\infty) = \infty$, the map extends to an automorphism of the projective space $\mathbf{P}^1$.
By the previous theorem, every such automorphism is a Möbius transformation $f(z) = \frac{az+b}{cz+d}$. For this transformation to map $\infty \mapsto \infty$, we must strictly have $c = 0$, immediately reducing the function to the affine form $f(z) = az + b$.
The conformal automorphisms of the upper half plane $\mathbf{H} = \{z \in \mathbf{C} : \operatorname{Im}(z) > 0\}$ correspond naturally to the real projective transformations.
By scaling the coefficients such that $ad - bc = 1$, we obtain an explicit isomorphism to the projective special linear group $PSL(2, \mathbf{R})$.
To rigorously prove that every conformal automorphism takes this form, we must show that any $f \in \operatorname{Aut}(\mathbf{H})$ is necessarily a fractional linear transformation. We establish this by extending the domain of $f$.
Let $f \colon \mathbf{H} \to \mathbf{H}$ be a biholomorphism. Because $f$ maps the real axis $\mathbf{R}$ bijectively to itself, the Schwarz Reflection Principle allows us to analytically continue $f$ across $\mathbf{R}$ into the lower half plane. This symmetric extension produces a bijective meromorphic function on the entire projective space $\mathbf{P}^1$.
By our previous characterization, any such global automorphism must be a Möbius transformation $f(z) = \frac{az+b}{cz+d}$. For this transformation to preserve the extended real axis $\hat{\mathbf{R}}$, its coefficients must be real (up to a global complex scalar), yielding $a,b,c,d \in \mathbf{R}$.
Finally, evaluating the imaginary part yields $\operatorname{Im}(f(z)) = \frac{(ad-bc)\operatorname{Im}(z)}{|cz+d|^2}$. For $f$ to map the upper half plane onto itself (rather than the lower half plane), we must strictly have $ad - bc > 0$.
The upper half plane $\mathbf{H}$ and the open unit disk $\mathbf{D}$ are conformally equivalent domains. This equivalence establishes a strict group isomorphism between their respective automorphism groups.
A Möbius transformation is uniquely determined by its action on three distinct points.
First, construct a transformation $T$ mapping $(z_1, z_2, z_3)$ to $(0, 1, \infty)$. The explicit formula is:
Similarly, construct $S$ mapping $(w_1, w_2, w_3)$ to $(0, 1, \infty)$. The composition $f = S^{-1} \circ T$ maps $z_k \mapsto w_k$.
For uniqueness, assume $g$ also maps $z_k \mapsto w_k$. The composition $h = S \circ g \circ T^{-1}$ is a Möbius transformation $h(z) = \frac{az+b}{cz+d}$ fixing $0, 1,$ and $\infty$. Since $h(\infty) = \infty$, $c = 0$, so $h(z) = \frac{a}{d}z + \frac{b}{d}$. Since $h(0) = 0$, $b = 0$, yielding $h(z) = \frac{a}{d}z$. Finally, $h(1) = 1$ implies $a = d$. Thus, $h(z) = z$, the identity map. This strictly proves $g = S^{-1} \circ T = f$.
When flattening a 3D cortical surface onto a sphere, the mapping retains the three complex degrees of freedom of $PSL(2, \mathbf{C})$.
By fixing exactly three distinct anatomical landmarks on the cortex, the 3-point rigidity theorem ensures the resulting Möbius transformation is uniquely determined. This eliminates arbitrary rotational freedom.
The cross-ratio of four distinct points $(z_1, z_2, z_3, z_4)$ in $\mathbf{P}^1$ is defined as:
Define $T(z) = (z, z_2; z_3, z_4)$. By construction, $T$ is the unique Möbius transformation mapping $(z_2, z_3, z_4)$ to $(1, 0, \infty)$.
Let $f$ be an arbitrary Möbius transformation. The composition $T \circ f^{-1}$ is a Möbius transformation mapping $(f(z_2), f(z_3), f(z_4))$ to $(1, 0, \infty)$.
By the 3-point rigidity theorem, $T \circ f^{-1}$ is the unique map sending those three points to $(1, 0, \infty)$. Evaluating this map at $f(z_1)$ computes the cross-ratio of the images: $(T \circ f^{-1})(f(z_1)) = T(z_1) = (z_1, z_2; z_3, z_4)$. Thus, the cross-ratio is preserved.
In 2D X-ray fluoroscopy, the imaging process acts as a projective transformation on the 2D detector plane.
By placing four collinear radio-opaque markers on a surgical instrument, their cross-ratio remains strictly invariant on the resulting radiograph. This invariant allows the tracking system to identify the instrument independently of the C-arm camera angle.
A generalized circle in $\mathbf{P}^1$ is either a standard circle or a straight line (a circle passing through the point at infinity).
The general equation for a generalized circle in $\mathbf{C}$ is $A|z|^2 + B z + \bar{B} \bar{z} + C = 0$, where $A, C \in \mathbf{R}$ and $|B|^2 > AC$.
Since any Möbius transformation can be decomposed into translations $z \mapsto z + a$, dilations/rotations $z \mapsto k z$, and inversion $z \mapsto 1/z$, it suffices to check invariance under these generators.
Translations and dilations trivially map circles to circles and lines to lines. For inversion $w = 1/z$, substitute $z = 1/w$ into the equation and multiply by $|w|^2$ to obtain $C|w|^2 + \bar{B} w + B \bar{w} + A = 0$. This preserves the form, yielding a generalized circle.
The stereographic projection is exactly a Möbius transformation mapping $\mathbf{S}^2$ to the extended complex plane $\mathbf{P}^1$.
Due to the generalized circle-preserving property, geometric circles on the Earth's surface (such as lines of latitude) map exclusively to circles or straight lines on the flat cartographic projection.