Skip to main content

Formula sheet

The main results of each module, with links to their derivations. Each result holds only under its module’s assumptions; read those before using it. Every formula here is checked by a test in tests/.

m01 · Bioelectric sources

Symbols: \(I\) point-source current, \(\sigma\) conductivity, \(\mathbf p\) dipole moment, \(\tau\) double-layer strength, \(\Omega\) solid angle, \(V_m\) transmembrane voltage, \(\beta\) intracellular conductance, \(I_i\) intracellular current, \(i_m\) membrane current, \(s\) distance from the front, \(h\) electrode height. Model: an infinite homogeneous isotropic conductor, quasi-static fields, a thin strand or sheet.

Point current source (derivation)

\[ \phi(\mathbf r) = \frac{I}{4\pi\sigma\,\lvert\mathbf r - \mathbf r'\rvert} \]

Current dipole (derivation)

\[ \phi(\mathbf r) = \frac{\mathbf p\cdot(\mathbf r - \mathbf r_0)}{4\pi\sigma\,\lvert\mathbf r - \mathbf r_0\rvert^3} \]

Uniform double layer (derivation)

\[ \phi = \frac{\tau\,\Omega}{4\pi\sigma} \]

Core conductor: membrane current (derivation)

\[ I_i = -\beta\,\frac{\partial V_m}{\partial x}, \qquad i_m = -\frac{\partial I_i}{\partial x} = \beta\,\frac{\partial^2 V_m}{\partial x^2} \]

Dipole moment of a front (derivation)

\[ p = \int s\,i_m\,ds = \beta\,\Delta V \]

A front as a line of dipoles (derivation)

\[ \phi(s) = \frac{p\,s}{2\pi\sigma\,(s^2 + h^2)} \]

m02 · Volume conduction and Poisson’s equation

Symbols: \(\phi\) potential, \(\sigma\) conductivity (\(\sigma_1\), \(\sigma_2\) either side of an interface), \(I_v\) current source density, \(I\) point-source current, \(k\) reflection coefficient, \(r\), \(r'\) distances to a source and its image, \(h\) grid spacing, \(\psi\) potential from unit current injected at an electrode, \(\mathbf p\) dipole moment. Model: quasi-static fields, isotropic piecewise-uniform conductors.

Poisson’s equation (derivation)

\[ \nabla\cdot\left(\sigma\,\nabla\phi\right) = -I_v \]

Interface conditions (derivation)

\[ \phi_1 = \phi_2, \qquad \sigma_1\,\frac{\partial\phi_1}{\partial n} = \sigma_2\,\frac{\partial\phi_2}{\partial n} \]

Method of images, flat interface (derivation)

\[ k = \frac{\sigma_1 - \sigma_2}{\sigma_1 + \sigma_2}, \qquad \phi_1 = \frac{I}{4\pi\sigma_1}\left(\frac1r + \frac{k}{r'}\right), \qquad \phi_2 = \frac{I}{2\pi(\sigma_1 + \sigma_2)\,r} \]

Flux-conserving finite differences (derivation)

\[ \frac{1}{h^2}\sum_{\text{4 neighbors } j} \sigma_{ij}\left(\phi_j - \phi_i\right) = -I_{v,i}, \qquad \sigma_{ij} = \frac{2\sigma_i\sigma_j}{\sigma_i + \sigma_j} \]

Reciprocity and the lead field (derivation)

\[ V = \mathbf p\cdot\nabla\psi(\mathbf r_0) \]

m03 · Unipolar, bipolar and omnipolar electrograms

Symbols: \(\sigma\) conductivity, \(i_m\) membrane current per area, \(V_m\) transmembrane voltage, \(h\) electrode height, \(w\) front width, \(c\) conduction velocity, \(d\) spacing, \(\theta\) angle between bipole and propagation, \(k\) spatial frequency, \(r_e\) electrode radius, \(s\) distance from the front. Model: a thin tissue sheet in an infinite homogeneous conductor and a planar front.

Unipolar potential, thin sheet in an infinite homogeneous conductor (derivation)

\[ \phi(\mathbf r, t) = \frac{1}{4\pi\sigma} \iint \frac{i_m(\mathbf r', t)}{\lVert \mathbf r - \mathbf r' \rVert}\, dA' \]

Membrane current from the voltage profile (derivation)

\[ i_m = \beta\,\frac{\partial^2 V_m}{\partial x^2} \]

Height filter (derivation)

\[ \hat\phi(k) = \frac{e^{-|k| h}}{2\sigma |k|}\, \hat i_m(k) \]

Unipolar electrogram for a planar arctangent front (derivation)

\[ U(s) = A\,\frac{s}{s^2 + a^2}, \qquad A = \frac{\beta\,\Delta V}{2\pi\sigma}, \qquad a = w + h \]

Unipolar slope (derivation)

\[ \frac{dU}{dt} = -c\,\frac{dU}{ds} \]

Bipolar electrogram (distal minus proximal) (derivation)

\[ V_{\text{bip}}(t) = \phi(\mathbf r_1, t) - \phi(\mathbf r_2, t), \qquad \mathbf r_{1,2} = \mathbf r_0 \pm \tfrac{d}{2}\,\hat u \]

Close bipole (derivation)

\[ V_{\text{bip}} \approx d\, \hat u \cdot \nabla \phi = d \cos\theta\, \frac{dU}{ds} \]

Disk electrode average (derivation)

\[ H_{\text{disk}}(k) = \frac{2 J_1(k r_e)}{k r_e} \]

Bipole transfer function (derivation)

\[ H_{\text{bip}}(k) = 2i \sin\!\left( \frac{k d \cos\theta}{2} \right), \qquad |H_{\text{bip}}| = 2 \left| \sin\!\left( \frac{k d \cos\theta}{2} \right) \right| \]

Far-field decay at distance R (derivation)

\[ \phi_{\text{monopole}} \propto \frac{1}{R},\quad V_{\text{bip}} \propto \frac{d}{R^2}; \qquad \phi_{\text{dipole}} \propto \frac{1}{R^2},\quad V_{\text{bip}} \propto \frac{d}{R^3} \]

Electric field from a clique (least squares) (derivation)

\[ \hat{\mathbf E}(t) = -\left( D D^{\mathsf T} \right)^{-1} D\, \mathbf b(t) \]

Omnipolar signal (derivation)

\[ v_\omega(t) = -d\, \hat\omega \cdot \hat{\mathbf E}(t) \]

Conduction velocity from one clique (derivation)

\[ c \approx \frac{\mathrm{pp}\left( dU/dt \right)}{\mathrm{pp}\left( \hat n \cdot \hat{\mathbf E} \right)} \]

m04 · Sampling, filtering and the EP recording chain

Symbols: \(f\) frequency, \(f_s\) sampling rate, \(T = 1/f_s\), \(N\) bits, \(V_{FS}\) full scale, \(q\) quantization step, \(f_c\) cutoff, \(n\) order, \(\tau_g\) group delay, \(f_0\) notch frequency, \(Q\) quality factor, \(\omega_0 = 2\pi f_0/f_s\). Model: ideal sampler, uniform rounding error, Butterworth filters, bilinear transform with prewarping.

Aliasing (derivation)

\[ f_{\text{alias}} = \left\lvert f - f_s\,\operatorname{round}\!\left(\frac{f}{f_s}\right)\right\rvert \]

Quantization noise (derivation)

\[ \mathrm{SQNR} = 10\log_{10}\frac{V_{FS}^2/2}{q^2/12} = 20\log_{10}\!\left(2^N\sqrt{3/2}\right) \approx 6.02N + 1.76\ \text{dB} \]

First-order high-pass (derivation)

\[ H(f) = \frac{jf/f_c}{1 + jf/f_c} \]

Butterworth magnitude (derivation)

\[ \lvert H(f)\rvert^2 = \frac{1}{1 + (f/f_c)^{2n}} \]

Group delay of a low-pass (derivation)

\[ \tau_g = \frac{1}{2\pi f_c}\ (n = 1), \qquad \tau_g = \frac{\sqrt2}{2\pi f_c}\ (n = 2) \]

Second-order notch (derivation)

\[ H(z) = g\,\frac{1 - 2\cos\omega_0\,z^{-1} + z^{-2}}{1 - 2g\cos\omega_0\,z^{-1} + (2g - 1)\,z^{-2}}, \qquad \omega_0 = \frac{2\pi f_0}{f_s}, \qquad g = \frac{1}{1 + \tan(\omega_0/2Q)} \]

m05 · Time–frequency tools

Symbols: \(f_s\) sampling rate, \(N\) samples, \(T\) record length, \(X\) Fourier transform, \(W\) window transform, \(K\) averaged segments, \(T_0\) cycle length, \(p\) one deflection, \(z\) analytic signal, \(\mathcal H\) Hilbert transform, \(\varphi\) phase, \(f_i\) instantaneous frequency, \(\sigma_t\), \(\sigma_f\) RMS duration and bandwidth, \(\psi\) wavelet, \(s\) scale, \(\omega_0\) wavelet frequency parameter. Model: stationary records unless stated; synthetic signals.

Frequency resolution (derivation)

\[ f_k = \frac{k f_s}{N} = \frac{k}{T}, \qquad \Delta f = \frac{1}{T} \]

Windowing and leakage (derivation)

\[ X_w(f) = \int X(\nu)\,W(f - \nu)\,d\nu \]

Welch averaging (derivation)

\[ \frac{\operatorname{std}\hat P(f)}{\operatorname{E}\hat P(f)} = \frac{1}{\sqrt K}, \qquad \Delta f = \frac{K}{T} \]

Spectrum of a pulse train (derivation)

\[ x(t) = \sum_k p(t - kT_0) \quad\Longrightarrow\quad X(f) = \frac{1}{T_0}\sum_m P\!\left(\frac{m}{T_0}\right)\delta\!\left(f - \frac{m}{T_0}\right) \]

Analytic signal and instantaneous frequency (derivation)

\[ z(t) = x(t) + j\,\mathcal H[x](t), \qquad \varphi(t) = \arg z(t), \qquad f_i(t) = \frac{1}{2\pi}\frac{d\varphi}{dt} \]

Time-frequency uncertainty (derivation)

\[ \sigma_t\,\sigma_f \ge \frac{1}{4\pi} \]

Morlet wavelet transform (derivation)

\[ \psi(t) = \pi^{-1/4}\,e^{j\omega_0 t}\,e^{-t^2/2}, \qquad W(s, \tau) = \frac1s\int x(t)\,\psi^*\!\left(\frac{t - \tau}{s}\right)dt, \qquad f = \frac{\omega_0}{2\pi s} \]

m06 · Triangle meshes, Laplace–Beltrami operators and geodesics

Symbols: \(V\), \(E\), \(F\) vertices, edges, faces; \(g\) genus; \(u_i\) vertex value; \(T\) activation time; \(A_f\), \(\mathbf N_f\) face area and unit normal; \(\mathbf e_i\) edge opposite vertex \(i\); \(\mathbf v\) conduction velocity; \(\alpha_{ij}\), \(\beta_{ij}\) angles opposite edge \(ij\); \(A_i\) vertex area; \(L_C\) cotangent matrix; \(\mathbf A\) diagonal of vertex areas; \(\lambda\), \(\phi\) eigenpair; \(R\) sphere radius; \(l\) degree; \(t\) diffusion time; \(\boldsymbol\delta_s\) unit source. Model: piecewise-linear functions on triangle meshes; test surfaces are flat lattices and icospheres.

Euler’s formula (derivation)

\[ \chi = V - E + F = 2 - 2g, \qquad 2E = 3F \]

Gradient on a face and conduction velocity (derivation)

\[ \nabla u\big|_f = \frac{1}{2A_f}\sum_{i \in f} u_i\,(\mathbf N_f \times \mathbf e_i), \qquad \mathbf v = \frac{\nabla T}{\lvert\nabla T\rvert^2} \]

Cotangent Laplacian (derivation)

\[ (\Delta u)_i \approx \frac{1}{A_i}\sum_{j \in \mathcal N(i)} \frac{\cot\alpha_{ij} + \cot\beta_{ij}}{2}\,(u_j - u_i) \]

Laplace–Beltrami spectrum (derivation)

\[ -\Delta \phi = \lambda\,\phi, \qquad \lambda_l = \frac{l(l+1)}{R^2}\ \ (l = 0, 1, 2, \dots,\ \text{multiplicity } 2l + 1), \qquad -L_C\,\boldsymbol\phi = \lambda\,\mathbf A\,\boldsymbol\phi \]

Heat method for geodesic distance (derivation)

\[ (\mathbf A - t\,L_C)\,\mathbf u = \boldsymbol\delta_s, \qquad \mathbf X = -\frac{\nabla u}{\lvert\nabla u\rvert}, \qquad L_C\,\boldsymbol\phi = \nabla\!\cdot\mathbf X \]

m07 · Inverse problems and regularization

Symbols: \(\mathbf x\) sources, \(\mathbf b\) data, \(A\) forward matrix, \(\mathbf e\) noise (\(\sigma\) per sample); \(s_i\), \(\mathbf u_i\), \(\mathbf v_i\) singular values and vectors; \(\kappa\) condition number; \(\hat\phi(k, z)\) Fourier coefficient at wavenumber \(k\) and height \(z\); \(h\) height; \(\varepsilon\) relative noise; \(f_i\) filter factors; \(k_c\) truncation index; \(\lambda\) regularization parameter; \(\tau\) safety factor; \(\sigma_x\), \(\sigma_e\) prior and noise spreads. Model: linear forward model with white noise; continuation across a source-free layer as the worked example.

Linear model and condition number (derivation)

\[ \mathbf b = A\mathbf x + \mathbf e, \qquad \frac{\lVert\delta\mathbf x\rVert}{\lVert\mathbf x\rVert} \le \kappa(A)\,\frac{\lVert\delta\mathbf b\rVert}{\lVert\mathbf b\rVert}, \qquad \kappa(A) = \frac{s_1}{s_n} \]

Continuation across a source-free layer (derivation)

\[ \hat\phi(k, h) = e^{-\lvert k\rvert h}\,\hat\phi(k, 0), \qquad \lvert k\rvert < k_{\max} = \frac{\ln(1/\varepsilon)}{h} \]

Naive inverse in the SVD (derivation)

\[ \mathbf x_{\text{naive}} = \sum_{i} \frac{\mathbf u_i^\top\mathbf b}{s_i}\,\mathbf v_i = \mathbf x + \sum_{i} \frac{\mathbf u_i^\top\mathbf e}{s_i}\,\mathbf v_i \]

Filter factors: truncated SVD and Tikhonov (derivation)

\[ \mathbf x_{\text{filt}} = \sum_i f_i\,\frac{\mathbf u_i^\top\mathbf b}{s_i}\,\mathbf v_i, \qquad f_i^{\text{TSVD}} = \begin{cases} 1, & i \le k_c \\ 0, & i > k_c \end{cases}, \qquad f_i^{\text{Tikh}} = \frac{s_i^2}{s_i^2 + \lambda^2} \]

Expected error: bias and noise (derivation)

\[ \operatorname{E}\lVert\mathbf x_{\text{filt}} - \mathbf x\rVert^2 = \sum_i (1 - f_i)^2\,(\mathbf v_i^\top\mathbf x)^2 + \sigma^2 \sum_i \frac{f_i^2}{s_i^2} \]

Choosing λ: discrepancy and GCV (derivation)

\[ \text{discrepancy: } \lVert A\mathbf x_\lambda - \mathbf b\rVert = \tau\lVert\mathbf e\rVert; \qquad \text{GCV: } \min_\lambda \frac{\lVert A\mathbf x_\lambda - \mathbf b\rVert^2}{\left[m - \sum_i f_i\right]^2} \]

Bayesian reading of Tikhonov (derivation)

\[ \mathbf x \sim \mathcal N(0, \sigma_x^2 I),\ \mathbf e \sim \mathcal N(0, \sigma_e^2 I) \;\Longrightarrow\; \operatorname{E}[\mathbf x \mid \mathbf b] = (A^\top A + \lambda^2 I)^{-1}A^\top\mathbf b, \quad \operatorname{Cov}[\mathbf x \mid \mathbf b] = \sigma_e^2 (A^\top A + \lambda^2 I)^{-1}, \quad \lambda = \frac{\sigma_e}{\sigma_x} \]

m08 · Action potentials and ionic models

Symbols: \(V\) voltage; \(C_m\) capacitance; \(I_{\text{ion}}\), \(I_{\text{stim}}\) currents; \(R\), \(T\), \(F\), \(z\) gas constant, temperature, Faraday constant, valence; \(m\), \(h\) gates; \(\alpha\), \(\beta\) rates; FitzHugh–Nagumo \(v\), \(w\), \(I\), \(a\), \(b\), \(\varepsilon\); Mitchell–Schaeffer \(v\), \(h\), \(\tau_{\text{in}}\), \(\tau_{\text{out}}\), \(\tau_{\text{open}}\), \(\tau_{\text{close}}\), \(v_{\text{gate}}\); DI diastolic interval; APD action potential duration. Model: single cells, no coupling; FitzHugh–Nagumo and Mitchell–Schaeffer are dimensionless in voltage.

Membrane equation (derivation)

\[ C_m \frac{dV}{dt} = -I_{\text{ion}} + I_{\text{stim}} \]

Nernst potential (derivation)

\[ E_X = \frac{RT}{zF}\ln\frac{[X]_o}{[X]_i} \]

Gating (derivation)

\[ \frac{dm}{dt} = \alpha_m(V)(1 - m) - \beta_m(V)\,m, \qquad m_\infty = \frac{\alpha_m}{\alpha_m + \beta_m}, \quad \tau_m = \frac{1}{\alpha_m + \beta_m}, \qquad I_{\text{Na}} = \bar g_{\text{Na}}\,m^3 h\,(V - E_{\text{Na}}) \]

FitzHugh–Nagumo model (derivation)

\[ \frac{dv}{dt} = v - \frac{v^3}{3} - w + I, \qquad \frac{dw}{dt} = \varepsilon\,(v + a - b\,w) \]

Onset of repetitive firing (derivation)

\[ 1 - v_*^2 - \varepsilon b = 0 \;\Longrightarrow\; v_* = -\sqrt{1 - \varepsilon b}, \qquad I_H = \frac{v_*^3}{3} - v_* + \frac{v_* + a}{b} \]

Mitchell–Schaeffer model (derivation)

\[ \frac{dv}{dt} = \frac{h\,v^2(1 - v)}{\tau_{\text{in}}} - \frac{v}{\tau_{\text{out}}} + J_{\text{stim}}, \qquad \frac{dh}{dt} = \begin{cases} (1 - h)/\tau_{\text{open}}, & v < v_{\text{gate}} \\ -h/\tau_{\text{close}}, & v \ge v_{\text{gate}} \end{cases} \]

APD and restitution (closed form) (derivation)

\[ \text{APD} \approx \tau_{\text{close}}\ln\frac{h_0}{h_{\min}}, \qquad h_{\min} = \frac{4\tau_{\text{in}}}{\tau_{\text{out}}}, \qquad h_0 = 1 - (1 - h_{\min})\,e^{-\text{DI}/\tau_{\text{open}}} \]

m09 · Propagation: cable theory, monodomain and bidomain, anisotropy

Symbols: \(V\) transmembrane voltage; \(\varphi_e\) extracellular potential; \(r_m\), \(r_i\), \(r_e\), \(c_m\) cable constants; \(\lambda\) length constant; \(\tau_m\) membrane time constant; \(\sigma_i\), \(\sigma_e\), \(\sigma_m\) conductivities; \(\chi\) area per volume; \(C_m\) capacitance; \(D\) diffusion coefficient; CV velocity, L and T along and across fibers; \(\theta\) angle to fibers; \(\kappa\) curvature; \(c_0\) plane-wave velocity. Model: continuum tissue; Mitchell–Schaeffer kinetics in the simulations.

Passive cable (derivation)

\[ \lambda^2\frac{\partial^2 V}{\partial x^2} = \tau_m\frac{\partial V}{\partial t} + V, \qquad \lambda = \sqrt{\frac{r_m}{r_i + r_e}}, \quad \tau_m = r_m c_m \]

Bidomain equations (derivation)

\[ \nabla\cdot\big(\sigma_i\nabla(V + \varphi_e)\big) = \chi\Big(C_m\frac{\partial V}{\partial t} + I_{\text{ion}}\Big), \qquad \nabla\cdot\big((\sigma_i + \sigma_e)\nabla\varphi_e\big) = -\nabla\cdot(\sigma_i\nabla V) \]

Monodomain equation (derivation)

\[ \chi\Big(C_m\frac{\partial V}{\partial t} + I_{\text{ion}}\Big) = \nabla\cdot(\sigma_m\nabla V), \qquad \sigma_m = \sigma_i(\sigma_i + \sigma_e)^{-1}\sigma_e \]

Velocity scaling (derivation)

\[ \text{CV} \propto \sqrt{D}, \qquad \frac{\text{CV}_L}{\text{CV}_T} = \sqrt{\frac{\sigma_{m,L}}{\sigma_{m,T}}} \]

Anisotropic spread from a point (derivation)

\[ T(\mathbf x) = \sqrt{\frac{x_L^2}{\text{CV}_L^2} + \frac{x_T^2}{\text{CV}_T^2}}, \qquad c(\theta) = \frac{\text{CV}_L\,\text{CV}_T}{\sqrt{\text{CV}_T^2\cos^2\theta + \text{CV}_L^2\sin^2\theta}} \]

Curvature (derivation)

\[ c(\kappa) \approx c_0 - D\,\kappa \]

m10 · Arrhythmia mechanisms

Symbols: CV velocity; ERP refractory period; \(\lambda\) wavelength; \(L\) circuit length; CL cycle length; APD, DI action potential duration and diastolic interval; \(f\) restitution function; \(\delta\) deviation from the steady state; \(\tau_{\text{open}}\), \(\tau_{\text{close}}\), \(h_{\min}\) m08 constants; \(\mathbf x_0\) focus; \(T\) activation time. Model: uniform tissue; one-beat memory for the restitution map.

Wavelength (derivation)

\[ \lambda = \text{CV}\times\text{ERP} \]

Reentrant circuit (derivation)

\[ \text{CL} = \frac{L}{\text{CV}}, \qquad L > \lambda, \qquad \text{gap} = L - \lambda = \text{CV}\,(\text{CL} - \text{ERP}) \]

Restitution map (derivation)

\[ \text{APD}_{n+1} = f(\text{CL} - \text{APD}_n), \qquad \delta_{n+1} = -f'(\text{DI}^*)\,\delta_n \]

Alternans onset (closed form) (derivation)

\[ \text{DI}_c = \tau_{\text{open}}\ln\!\Big[(1 - h_{\min})\Big(1 + \frac{\tau_{\text{close}}}{\tau_{\text{open}}}\Big)\Big] \]

Focal activation (derivation)

\[ T(\mathbf x) = T_0 + \frac{\lvert\mathbf x - \mathbf x_0\rvert}{\text{CV}}, \qquad \text{span} = \frac{\max T - \min T}{\text{CL}} < 1 \]

m11 · The EP study

Symbols: \(t_A\), \(t_H\), \(t_V\) local activation times; AH, HV intervals; NCT nodal conduction time; RT recovery time; \(\text{ERP}_N\) nodal refractory period; \(\text{NCT}_{\min}\), \(\text{NCT}_{\max}\), \(\tau\) recovery-curve constants; CL pacing cycle length; TCL tachycardia cycle length; PPI post-pacing interval; \(d\) distance to the circuit; CV velocity. Model: one-variable AV nodal model; uniform conduction for the PPI.

AH and HV intervals (derivation)

\[ \text{AH} = t_H - t_A, \qquad \text{HV} = t_V - t_H \]

AV nodal recovery curve (derivation)

\[ \text{NCT}(\text{RT}) = \text{NCT}_{\min} + (\text{NCT}_{\max} - \text{NCT}_{\min})\,e^{-(\text{RT} - \text{ERP}_N)/\tau}, \qquad \text{RT} \ge \text{ERP}_N \]

Wenckebach point (derivation)

\[ \text{CL}_W = \text{ERP}_N + \text{NCT}_{\max}, \qquad \text{stable if } \frac{\text{NCT}_{\max} - \text{NCT}_{\min}}{\tau} < 1 \]

Post-pacing interval (derivation)

\[ \text{PPI} - \text{TCL} \approx \frac{2d}{\text{CV}}, \qquad \text{PPI} \approx \text{TCL} \text{ on the circuit} \]

m12 · The surface ECG and the forward problem

Symbols: \(\mathbf p\) heart vector; \(\sigma\) conductivity; \(R\) torso radius; \(\hat{\mathbf r}\) surface direction; \(\theta\) angle to \(\mathbf p\); \(\mathbf r_0\) source position; \(I\) point current; \(\gamma\) angle to \(\mathbf r_0\); \(P_l\) Legendre polynomials; \(\Phi_R\), \(\Phi_L\), \(\Phi_F\) limb potentials; I, II, III, aVR, aVL, aVF limb leads; WCT Wilson central terminal; \(\mathbf L\) lead vector; \(\alpha\) frontal axis. Model: homogeneous, insulated spherical torso; limb electrodes at Einthoven’s equilateral positions.

Centric dipole in a sphere (derivation)

\[ \varphi(r, \theta) = \frac{p\cos\theta}{4\pi\sigma}\left(\frac{1}{r^2} + \frac{2r}{R^3}\right), \qquad \varphi(R, \theta) = \frac{3\,\mathbf p\cdot\hat{\mathbf r}}{4\pi\sigma R^2} \]

Off-center source in a sphere (derivation)

\[ \varphi(R, \gamma) = \frac{I}{4\pi\sigma}\sum_{l \ge 1}\frac{2l + 1}{l}\,\frac{r_0^l}{R^{l+1}}\,P_l(\cos\gamma) + \text{const} \]

Lead vector (derivation)

\[ V = \mathbf L\cdot\mathbf p, \qquad \mathbf L_{ab} = \frac{3}{4\pi\sigma R^2}\left(\hat{\mathbf a} - \hat{\mathbf b}\right) \text{ for a centric dipole} \]

Einthoven’s law (derivation)

\[ \text{I} = \Phi_L - \Phi_R, \quad \text{II} = \Phi_F - \Phi_R, \quad \text{III} = \Phi_F - \Phi_L \;\Longrightarrow\; \text{I} + \text{III} = \text{II} \]

Augmented leads and WCT (derivation)

\[ \text{WCT} = \frac{\Phi_R + \Phi_L + \Phi_F}{3}, \qquad \text{aVR} = -\frac{\text{I} + \text{II}}{2}, \quad \text{aVL} = \frac{\text{I} - \text{III}}{2}, \quad \text{aVF} = \frac{\text{II} + \text{III}}{2} \]

Frontal axis (derivation)

\[ \alpha = \operatorname{atan2}\!\left(\frac{2\,\text{aVF}}{\sqrt3},\ \text{I}\right) \]

Forward problem (derivation)

\[ V(t) = \mathbf L\cdot\mathbf p(t), \qquad \mathbf p(t) = \int \text{dipole density of the activation fronts at time } t \]

m13 · Catheter design

Symbols: \(a\) electrode radius; \(\sigma\) conductivity; \(R_s\) spreading resistance; \(\theta\) bipole angle to propagation; \(A_0\) aligned amplitude; \(h\) height above the activated layer; \(k\) wavenumber; \(\Delta\) spacing; \(E\) energy above Nyquist; \(R\) basket radius; \(n\) splines; \(\vartheta\) polar angle; \(\varphi_i\) sensor angle; \(\mathbf F\) force; \(\alpha\), \(\beta\) axial and lateral sensitivity; \(\sigma_s\) sensor noise. Model: homogeneous medium; planar front; ideal spherical basket; linear force sensors with independent noise.

Spreading resistance (derivation)

\[ R_s^{\text{disk}} = \frac{1}{4\sigma a}, \qquad R_s^{\text{hemisphere}} = \frac{1}{2\pi\sigma a}, \qquad R_s^{\text{sphere}} = \frac{1}{4\pi\sigma a} \]

Bipole direction (derivation)

\[ A(\theta) = A_0\,\lvert\cos\theta\rvert, \qquad \max\bigl(\lvert\cos\theta\rvert, \lvert\sin\theta\rvert\bigr) \ge \tfrac{1}{\sqrt2}, \qquad \sqrt{A(\theta)^2 + A(\theta - \tfrac{\pi}{2})^2} = A_0 \]

Spacing, height and aliasing (derivation)

\[ g(x) = \frac{h/\pi}{x^2 + h^2}, \quad \hat g(k) = e^{-\lvert k\rvert h}, \qquad E = \frac{\int_{\pi/\Delta}^{\infty} e^{-2kh}\,dk}{\int_0^{\infty} e^{-2kh}\,dk} = e^{-2\pi h/\Delta}, \qquad \Delta_{\max} = \frac{2\pi h}{\ln(1/E)} \]

Basket spline spacing (derivation)

\[ s(\vartheta) = \frac{2\pi R\sin\vartheta}{n} \]

Contact-force estimate (derivation)

\[ s_i = \alpha F_z + \beta\,(F_x\cos\varphi_i + F_y\sin\varphi_i) + \text{noise}, \qquad \mathbf s = C\mathbf F, \qquad \hat{\mathbf F} = (C^\top C)^{-1}C^\top\mathbf s \]

Contact-force noise (derivation)

\[ \operatorname{cov}(\hat{\mathbf F}) = \sigma_s^2\,(C^\top C)^{-1}, \qquad \operatorname{SD}(\hat F_x) = \operatorname{SD}(\hat F_y) = \sqrt{\tfrac{2}{3}}\,\frac{\sigma_s}{\beta}, \qquad \operatorname{SD}(\hat F_z) = \frac{\sigma_s}{\sqrt3\,\alpha} \]

m14 · Biopotential front ends

Symbols: \(v_d\), \(v_{cm}\) differential and common-mode voltages; \(A_d\), \(A_{cm}\) their gains; \(R_1\)–\(R_4\) difference-stage resistors; \(G\) stage gain; \(t\) tolerance; \(G_1\) input-stage gain; \(Z_1\), \(Z_2\) electrode impedances; \(Z_{cm}\) common-mode input impedance; \(i_d\) displacement current; \(Z_{RL}\) right-leg impedance; \(G_{RL}\) loop gain; \(k\) Boltzmann’s constant; \(T\) temperature; \(R_s\) source resistance; \(e_n\), \(i_n\) amplifier noise densities; \(B\) bandwidth; \(\Phi_i\) limb potentials; \(R_W\) Wilson resistors; \(N\) bits; \(m\) samples averaged. Model: ideal op amps; resistive electrodes; white noise.

Differential and common mode (derivation)

\[ v_d = v_+ - v_-, \quad v_{cm} = \frac{v_+ + v_-}{2}, \qquad v_{out} = A_d v_d + A_{cm} v_{cm}, \qquad \text{CMRR} = \left\lvert\frac{A_d}{A_{cm}}\right\rvert \]

Difference amplifier (derivation)

\[ v_{out} = v_+\frac{R_4}{R_3 + R_4}\left(1 + \frac{R_2}{R_1}\right) - v_-\frac{R_2}{R_1}, \qquad \text{CMRR}_{\text{worst}} \approx \frac{1 + G}{4t} \]

Instrumentation amplifier (derivation)

\[ G_1 = 1 + \frac{2R}{R_G}, \qquad \text{CMRR}_{\text{total}} = G_1 \times \text{CMRR}_{\text{difference stage}} \]

Electrode impedance mismatch (derivation)

\[ v_{d,\text{error}} = v_{cm}\left(\frac{Z_{cm}}{Z_{cm} + Z_2} - \frac{Z_{cm}}{Z_{cm} + Z_1}\right) \approx v_{cm}\frac{Z_1 - Z_2}{Z_{cm}}, \qquad \text{CMRR}_{\text{imbalance}} = \left\lvert\frac{Z_{cm}}{\Delta Z}\right\rvert \]

Driven right leg (derivation)

\[ v_{cm} = i_d Z_{RL} - G_{RL}\,v_{cm} \;\Longrightarrow\; v_{cm} = \frac{i_d Z_{RL}}{1 + G_{RL}} \]

Noise (derivation)

\[ e_{\text{total}}^2 = 4kTR_s + e_n^2 + (i_n R_s)^2, \qquad v_{\text{rms}} = e_{\text{total}}\sqrt{B}, \qquad R_{s,\text{opt}} = \frac{e_n}{i_n} \]

Wilson central terminal (derivation)

\[ \Phi_{\text{WCT}} = \frac{\sum_i \Phi_i/(R_W + Z_i)}{\sum_i 1/(R_W + Z_i)} \]

Converter and oversampling (derivation)

\[ q_{\text{in}} = \frac{V_{FS}}{G\,2^N}, \qquad N \ge \log_2\frac{\text{span}}{\text{step}}, \qquad \Delta\text{SNR} = 10\log_{10} m \text{ dB} \]

m15 · Patient safety

Symbols: \(f\) frequency; \(C\) isolation capacitance; \(V\) rms voltage; \(I\) rms current; \(I_{\lim}\) a limit; \(a\) electrode radius; \(r\) distance; \(A\) body cross-section; \(J\) current density; \(V_{in}\) input voltage; \(V_c\) clamp voltage; \(R\) series resistance; \(T\) pulse duration; \(V_b\), \(R_b\) body voltage and resistance; \(f_c\) corner; \(n\) order; \(A_{RF}\) RF amplitude; \(\beta\) nonlinearity. Model: capacitive barriers on one mains phase; hemispherical electrode in a uniform medium; rectangular pulses; Butterworth filters; weak square-law nonlinearity.

Leakage through isolation (derivation)

\[ I = 2\pi f C V, \qquad I_{\text{total}} = \Bigl\lvert \sum_k I_k e^{j\phi_k} \Bigr\rvert \le \sum_k I_k, \qquad C_{\max} = \frac{I_{\lim}}{2\pi f V} \]

Current density (derivation)

\[ J(r) = \frac{I}{2\pi r^2}, \qquad \frac{J(a)}{I/A} = \frac{A}{2\pi a^2} \]

Defibrillation protection (derivation)

\[ I = \frac{V_{in} - V_c}{R}, \qquad E_R = \frac{(V_{in} - V_c)^2\,T}{R}, \qquad \frac{V_{in} I}{V_b^2/R_b} \;\text{(share of the body's power taken by one input)} \]

RF filtering (derivation)

\[ \lvert H(f)\rvert = \frac{1}{\sqrt{1 + (f/f_c)^{2n}}} \approx \left(\frac{f_c}{f}\right)^n, \qquad n \ge \frac{\log(A_{RF}/V_{\text{allowed}})}{\log(f/f_c)} \]

Demodulation (derivation)

\[ y = x + \beta x^2, \quad x = A\cos\omega t \;\Longrightarrow\; \overline{y} = \frac{\beta A^2}{2} \]

m16 · Magnetic localization

Symbols: \(\mathbf m\) dipole moment; \(\mathbf r\), \(r\), \(\hat{\mathbf r}\) displacement, distance, direction; \(\mathbf B\) field; \(\mu_0\) permeability; \(f\) frequency; \(N\) turns; \(A\) area; \(\hat{\mathbf n}\) sensor axis; \(\mathbf p\) sensor position; \(\mathbf s_k\), \(\mathbf m_k\) generator coils; \(V_k\) readings; \(\theta\) pose; \(J\) Jacobian; \(\sigma\) noise; \(\mathbf q\), \(\alpha\) conductor position and response. Model: point dipoles; quasi-static fields; independent Gaussian reading noise.

Dipole field (derivation)

\[ \mathbf B(\mathbf r) = \frac{\mu_0}{4\pi}\,\frac{3(\mathbf m\cdot\hat{\mathbf r})\hat{\mathbf r} - \mathbf m}{r^3} \]

Sensor coil voltage (derivation)

\[ V = 2\pi f\,N A\,\hat{\mathbf n}\cdot\mathbf B \]

Readings and pose (derivation)

\[ V_k = \hat{\mathbf n}\cdot\mathbf B_k(\mathbf p - \mathbf s_k), \qquad \hat\theta = \arg\min_\theta \sum_k \bigl(V_k^{\text{measured}} - V_k(\theta)\bigr)^2, \quad \theta = (\mathbf p, \hat{\mathbf n}) \]

Precision (derivation)

\[ \operatorname{cov}(\hat\theta) \approx \sigma^2\,(J^\top J)^{-1} \]

Range sensitivity (derivation)

\[ \lvert\mathbf B\rvert = \frac{2m}{r^3} \;\Longrightarrow\; \delta r = \frac{r\,\delta B}{3\lvert\mathbf B\rvert} \propto r^4\,\delta B \]

Degrees of freedom (derivation)

\[ V_k\bigl(R_{\hat{\mathbf n}}(\psi)\,\hat{\mathbf n}\bigr) = V_k(\hat{\mathbf n}) \;\Longrightarrow\; \operatorname{rank} \frac{\partial \mathbf V}{\partial(\mathbf p, \boldsymbol\omega)} = 5 \]

Distortion (derivation)

\[ \mathbf m_{q,k} = \alpha\,\mathbf B_k(\mathbf q), \qquad V_k = \hat{\mathbf n}\cdot\bigl[\mathbf B_k(\mathbf p) + \mathbf B(\mathbf m_{q,k};\, \mathbf p - \mathbf q)\bigr], \qquad \delta\theta \approx (J^\top J)^{-1}J^\top\,\delta\mathbf V \]

m17 · Impedance localization

Symbols: \(I\) current; \(\sigma\) conductivity; \(k = I/(4\pi\sigma)\); \(\mathbf a\), \(\mathbf b\) patches; \(\phi\) potential; \(d\) patch half-separation; \(x\) axial position; \(E_0\) central field; \(\boldsymbol\phi\) three potentials; \(J\) local Jacobian; \(M\) metric; \(\ell\) known spacing; \(\sigma_{\text{ref}}\), \(\sigma_{\text{local}}\) conductivities; \(\delta\phi\) voltage noise. Model: point patches in an unbounded homogeneous medium unless stated; locally linear fields for scaling.

Patch pair (derivation)

\[ \phi(\mathbf r) = k\left(\frac{1}{\lvert\mathbf r - \mathbf a\rvert} - \frac{1}{\lvert\mathbf r - \mathbf b\rvert}\right), \qquad k = \frac{I}{4\pi\sigma} \]

Axis nonlinearity (derivation)

\[ \phi(x) = \frac{2kx}{d^2 - x^2}, \qquad E_0 = \frac{2k}{d^2}, \qquad \frac{\phi(x)/E_0 - x}{x} = \frac{x^2}{d^2 - x^2} \approx \frac{x^2}{d^2} \]

Linear map (derivation)

\[ \hat x_i = \frac{\phi_i(\mathbf r)}{E_0}, \qquad i = 1, 2, 3 \]

Field scaling (derivation)

\[ \ell^2 = \lvert\Delta\mathbf x\rvert^2 = \Delta\boldsymbol\phi^\top M\,\Delta\boldsymbol\phi, \qquad M = (JJ^\top)^{-1} \]

Interface (derivation)

\[ \frac{E_{\text{local}}}{E_{\text{ref}}} = \frac{\sigma_{\text{ref}}}{\sigma_{\text{local}}} \]

Precision (derivation)

\[ \delta x = \frac{\delta\phi}{\lvert E\rvert}, \qquad \lvert E_0\rvert = \frac{2k}{d^2} \]

m18 · Hybrid and dielectric localization

Symbols: \(\mathbf u\) impedance coordinates; \(\mathbf x\) true coordinates; \(A\), \(\mathbf b\) affine map; \(\varphi\) radial basis function of width \(s\); \(\mathbf w_j\) weights; \(\lambda\) regularization; \(h\) gap to the nearest reference point; \(v_1\), \(v_2\) variances; \(\sigma\) conductivity; \(\varepsilon_0\), \(\varepsilon_r\) permittivities; \(\omega\) angular frequency; \(\tau\) relaxation time; \(\varepsilon_\infty\), \(\Delta\varepsilon\) dispersion parameters. Model: smooth distortion of impedance coordinates; independent unbiased estimates; one Debye relaxation per tissue.

Affine correction (derivation)

\[ (\hat A, \hat{\mathbf b}) = \arg\min_{A,\mathbf b}\sum_{i=1}^N \lVert \mathbf x_i - A\mathbf u_i - \mathbf b\rVert^2 \]

Smooth correction (derivation)

\[ \mathbf x(\mathbf u) = A\mathbf u + \mathbf b + \sum_{j=1}^N \mathbf w_j\,\varphi(\lVert\mathbf u - \mathbf u_j\rVert), \qquad \varphi(r) = e^{-r^2/(2s^2)}, \qquad \begin{bmatrix}\Phi + \lambda I & P\\ P^\top & 0\end{bmatrix}\begin{bmatrix}W\\ C\end{bmatrix} = \begin{bmatrix}X\\ 0\end{bmatrix} \]

Coverage (derivation)

\[ \lvert f(x) - f_{\text{lin}}(x)\rvert \le \frac{h^2}{8}\max\lvert f''\rvert \]

Fusion (derivation)

\[ \hat{\mathbf x} = \frac{\hat{\mathbf x}_1/v_1 + \hat{\mathbf x}_2/v_2}{1/v_1 + 1/v_2}, \qquad \operatorname{var}(\hat{\mathbf x}) = \frac{1}{1/v_1 + 1/v_2} \le \min(v_1, v_2) \]

Complex conductivity (derivation)

\[ \sigma^*(\omega) = \sigma + j\omega\varepsilon_0\varepsilon_r(\omega), \qquad \varepsilon_r(\omega) = \varepsilon_\infty + \frac{\Delta\varepsilon}{1 + j\omega\tau}, \qquad \arg Z = -\arg\sigma^* \]

m19 · Motion and drift

Symbols: \(z\) displacement from exhale; \(b\) excursion; \(T\) period; \(n\) shape; \(\delta\) gating window; \(D\) duty cycle; \(\sigma_x\), \(\sigma_r\) SDs of point and reference motion; \(\rho\) correlation; \(k\) slope; \(t_i\), \(x_i\) landmark visits; \(v\) drift rate; \(\sigma\) noise; \(W\) averaging window; \(f\) frequency. Model: periodic breathing; jointly Gaussian motions; linear drift; white noise.

Breathing (derivation)

\[ z(t) = -b\cos^{2n}\!\left(\frac{\pi t}{T}\right) \]

Gating (derivation)

\[ D = 1 - \frac{2}{\pi}\arccos\!\left[\left(\frac{\delta}{b}\right)^{1/(2n)}\right], \qquad \lvert z\rvert \le \delta \text{ for every accepted point} \]

Reference subtraction (derivation)

\[ \operatorname{SD}(x - r) = \sqrt{\sigma_x^2 + \sigma_r^2 - 2\rho\sigma_x\sigma_r} < \sigma_x \iff \rho > \frac{\sigma_r}{2\sigma_x} \]

Surrogate regression (derivation)

\[ \operatorname{SD}(x - kr - c) = \sigma_x\sqrt{1 - \rho^2} \]

Drift (derivation)

\[ \hat v = \frac{\sum_i (t_i - \bar t)(x_i - \bar x)}{\sum_i (t_i - \bar t)^2}, \qquad \operatorname{SE}(\hat v) = \frac{\sigma}{\sqrt{\sum_i (t_i - \bar t)^2}} \]

Averaging (derivation)

\[ \lvert H(f)\rvert = \left\lvert\frac{\sin(\pi f W)}{\pi f W}\right\rvert, \qquad \text{delay} = \frac{W}{2} \]

m20 · Geometry from catheter positions

Symbols: \(P\) positions; \(H(P)\) convex hull; \(B_r\) ball of radius \(r\); \(\oplus\), \(\ominus\) dilation and erosion; \(\mathbf x_i\) vertices with neighbors \(N(i)\); \(\lambda\), \(\mu\) smoothing steps; \(n\) polygon vertices, \(a = 1 - \cos(2\pi/n)\); \(\mathbf v_a\), \(\mathbf v_b\), \(\mathbf v_c\) triangle vertices; \(V\) volume; \(A\), \(B\) surfaces; \(d\) distance. Model: catheter positions near the wall; closed, outward-oriented meshes.

Convex hull (derivation)

\[ H(P) = \Bigl\{\textstyle\sum_i w_i\mathbf p_i : w_i \ge 0,\ \sum_i w_i = 1\Bigr\}, \qquad d(\mathbf w, \partial H) = \text{depth of the concavity at a wall point } \mathbf w \]

Inflate, fill, shrink (derivation)

\[ S_r(P) = \operatorname{fill}\bigl(P \oplus B_r\bigr) \ominus B_r \]

Smoothing (derivation)

\[ \mathbf x_i \leftarrow \mathbf x_i + \lambda\Bigl(\tfrac{1}{\lvert N(i)\rvert}\textstyle\sum_{j\in N(i)}\mathbf x_j - \mathbf x_i\Bigr), \qquad \rho \leftarrow (1 - \lambda a)\rho, \qquad \text{Taubin: } \rho \leftarrow (1 - \lambda a)(1 - \mu a)\rho \]

Volume and Hausdorff distance (derivation)

\[ V = \frac16\sum_{\text{triangles}}\mathbf v_a\cdot(\mathbf v_b\times\mathbf v_c), \qquad d_H(A, B) = \max\Bigl\{\sup_{\mathbf a\in A} d(\mathbf a, B),\ \sup_{\mathbf b\in B} d(\mathbf b, A)\Bigr\} \]

Mean surface distance (derivation)

\[ \bar d(A, B) = \frac{\sum_{\mathbf a\in A} d(\mathbf a, B) + \sum_{\mathbf b\in B} d(\mathbf b, A)}{\lvert A\rvert + \lvert B\rvert} \]

m21 · Image integration

Symbols: \(t\) threshold; \(\omega_0\), \(\omega_1\) class fractions; \(\mu_0\), \(\mu_1\) class means; \(\sigma_B^2\), \(\sigma_W^2\), \(\sigma_T^2\) between-class, within-class and total variance; \(A\), \(B\) segmentations; \(D\) Dice; \(J\) Jaccard; \(\mathbf p_i\), \(\mathbf q_i\) points; \(R\) rotation; \(\mathbf t\) translation; \(s\) scale; \(H = UDV^\top\) cross-covariance; \(S\) sign matrix; \(\mathbf n_i\) normals; \(\boldsymbol\omega\) small rotation; \(J_s\) point-to-plane Jacobian. Model: two-class intensity images; rigid or similarity transforms between image and map.

Otsu’s threshold (derivation)

\[ \sigma_T^2 = \sigma_W^2(t) + \sigma_B^2(t), \qquad \sigma_B^2(t) = \omega_0\omega_1(\mu_0 - \mu_1)^2, \qquad t^* = \arg\max_t \sigma_B^2(t) \]

Dice and Jaccard (derivation)

\[ D(A, B) = \frac{2\lvert A\cap B\rvert}{\lvert A\rvert + \lvert B\rvert}, \qquad J(A, B) = \frac{\lvert A\cap B\rvert}{\lvert A\cup B\rvert}, \qquad D = \frac{2J}{1 + J} \]

Landmark least squares (derivation)

\[ R = U S V^\top, \quad S = \operatorname{diag}\bigl(1, \ldots, 1, \operatorname{sign}\det(UV^\top)\bigr), \quad s = \frac{\operatorname{tr}(DS)}{\frac1N\sum_i\lVert\mathbf p_i - \bar{\mathbf p}\rVert^2}, \quad \mathbf t = \bar{\mathbf q} - sR\bar{\mathbf p} \]

Iterative closest point (derivation)

\[ j(i) = \arg\min_j \lVert R_k\mathbf p_i + \mathbf t_k - \mathbf q_j\rVert, \qquad (R_{k+1}, \mathbf t_{k+1}) = \arg\min_{R,\mathbf t}\sum_i\lVert R\mathbf p_i + \mathbf t - \mathbf q_{j(i)}\rVert^2 \]

Point-to-plane linearization (derivation)

\[ r_i = \mathbf n_i\cdot(\boldsymbol\omega\times\mathbf p_i + \mathbf t) = (\mathbf p_i\times\mathbf n_i)\cdot\boldsymbol\omega + \mathbf n_i\cdot\mathbf t, \qquad J_s^\top J_s \text{ has a zero eigenvalue for every motion that slides the surface along itself} \]

m22 · Electrogram annotation

Symbols: \(t_i\) local time; \(t_{\mathrm{ref}}\) reference time; \(\sigma_a\), \(\sigma_r\) annotation and reference SDs; \(U\) unipolar potential with strength \(A\); \(s\) front distance; \(c\) conduction velocity; \(a = w + h\); \(B\) bipolar potential; \(d\) spacing; \(\theta\) bipole angle; \(t_{\mathrm{mid}}\) midpoint activation; \(f\) onset fraction; \(\sigma_n\) noise SD; \(x\) signal; \(T\) cycle length; \(w_0\), \(W\) window start and width; \(l\) local signal; \(\phi\) far-field template; \(g\) gain. Model: the m03 arctangent front; planar propagation; a stable reference and cycle.

Local activation time (derivation)

\[ \mathrm{LAT}_i = t_i - t_{\mathrm{ref}}, \qquad \operatorname{Var}(\mathrm{LAT}_i - \mathrm{LAT}_j) = 2\sigma_a^2 + \begin{cases} 0 & \text{same beat} \\ 2\sigma_r^2 & \text{different beats} \end{cases} \]

Steepest downstroke and bipolar peak (derivation)

\[ \frac{dU}{ds} = A\frac{a^2 - s^2}{(s^2 + a^2)^2} \text{ is largest at } s = 0, \qquad B(t_{\mathrm{mid}} + \tau) = B(t_{\mathrm{mid}} - \tau), \quad \lvert B\rvert \text{ peaks at } t_{\mathrm{mid}} \text{ if } d\cos\theta < 2\sqrt3\,a \]

Onset bias (derivation)

\[ t_{\mathrm{mid}} - t_{\mathrm{on}} = \frac{a}{c}\sqrt{x_f}, \qquad \frac{1 - x_f}{(1 + x_f)^2} = f \]

Threshold jitter (derivation)

\[ \sigma_t = \frac{\sigma_n}{\lvert x'(t_c)\rvert} \]

Window of interest (derivation)

\[ \mathrm{LAT}_W = w_0 + \bigl((t - t_{\mathrm{ref}} - w_0) \bmod T\bigr), \quad \text{kept only if } \mathrm{LAT}_W < w_0 + W \]

Far-field sharpness and template (derivation)

\[ \frac{\max\lvert dU/dt\rvert}{\operatorname{pp}(U)} = \frac{c}{a}, \qquad x = l + g\phi \;\Rightarrow\; \hat g = \frac{\langle x, \phi\rangle}{\langle \phi, \phi\rangle} = g + \frac{\langle l, \phi\rangle}{\langle \phi, \phi\rangle} \]

m23 · Voltage mapping

Symbols: \(U\) unipolar potential with strength \(A\); \(a = w + h\); \(\operatorname{pp}\) peak-to-peak; \(B\) bipolar potential; \(d\) spacing; \(\theta\) bipole angle; \(q\) fraction of aligned amplitude; \(n\) noise samples; \(\sigma\) noise SD; \(\phi\), \(\Phi\) standard normal density and distribution; \(t\) cutoff; \(m_s\), \(m_n\) medians; \(\sigma_s\), \(\sigma_n\) log SDs. Model: the m03 arctangent front; planar propagation; log-normal tissue voltages.

Amplitude (derivation)

\[ \operatorname{pp}(U) = \frac{A}{a}, \qquad \operatorname{pp}(B) \to \frac98\,\frac{A\,d\lvert\cos\theta\rvert}{a^2} \;\; (d\lvert\cos\theta\rvert \ll a), \qquad \operatorname{pp}(B) \to \frac{A}{a} \;\; (d\lvert\cos\theta\rvert \gg a) \]

Direction (derivation)

\[ P\bigl(\lvert\cos\theta\rvert < q\bigr) = \frac{2}{\pi}\arcsin q, \qquad \mathrm E\lvert\cos\theta\rvert = \frac2\pi, \qquad \max\bigl(\lvert\cos\theta\rvert, \lvert\sin\theta\rvert\bigr) \ge \frac{1}{\sqrt2} \]

Noise floor (derivation)

\[ \mathrm E[\operatorname{pp}] = 2\sigma\int_{-\infty}^{\infty} x\, n\,\phi(x)\,\Phi(x)^{n-1}\,dx \]

Cutoffs (derivation)

\[ \text{sensitivity} = \Phi\Bigl(\frac{\ln t - \ln m_s}{\sigma_s}\Bigr), \qquad \text{specificity} = 1 - \Phi\Bigl(\frac{\ln t - \ln m_n}{\sigma_n}\Bigr), \qquad t^* = \sqrt{m_s m_n} \text{ if } \sigma_s = \sigma_n \]

m24 · Interpolation, activation maps and isochrones

Symbols: \(T\) activation time; \(T_i\) at mapped point \(i\); \(h\) spacing; \(\lambda_i\) barycentric weights; \(d_i\) distance; \(p\) IDW power; \(L\) cotangent Laplacian; \(\tau\) isochrone time; \(\Delta\tau\) step; \(c\) conduction velocity; \(\Delta T\) delay across a block; \(\rho\) point density; \(r\) fill radius. Model: piecewise-linear meshes; activation from mapped points only.

Linear interpolation (derivation)

\[ T(\mathbf x) = \sum_{i=1}^{3}\lambda_i(\mathbf x)\,T_i, \qquad \lvert T - I_h T\rvert \le \frac{h^2}{8}\max\lvert T''\rvert \]

Inverse-distance weighting (derivation)

\[ T(\mathbf x) = \frac{\sum_i d_i^{-p}\,T_i}{\sum_i d_i^{-p}}, \qquad \min_i T_i \le T \le \max_i T_i, \qquad \nabla T(\mathbf x_i) = \mathbf 0 \text{ for } p > 1 \]

Harmonic interpolation (derivation)

\[ (L\,T)_j = 0 \text{ at unmapped vertices } j, \qquad T_i \text{ fixed at mapped vertices } i \]

Isochrones (derivation)

\[ \mathbf x = \mathbf x_i + \frac{\tau - T_i}{T_j - T_i}(\mathbf x_j - \mathbf x_i), \qquad \text{spacing between isochrones } \Delta x = c\,\Delta\tau \]

Block and coverage (derivation)

\[ c_{\text{apparent}} = \frac{h}{\Delta T}, \qquad f_{\text{covered}} = 1 - e^{-\rho\pi r^2} \]

m25 · Conduction velocity estimation

Symbols: \(T\) activation time; \(\mathbf g = \nabla T\); \(\mathbf v\) velocity; \(c\) speed; \(\sigma\) timing SD; \(\sigma_g\) gradient SD; \(h\) spacing; \(E\) edge matrix; \(S\) difference covariance pattern; \(X_c\) centered coordinates; \(n\) points; \(R\) fit radius; \(\phi\) potential; \(\phi_t\) its time derivative; \(a\) m03 length scale; \(d\) clique spacing. Model: independent timing errors; plane waves for the potential-based result.

Speed from a gradient (derivation)

\[ \mathbf v = \frac{\nabla T}{\lvert\nabla T\rvert^2}, \qquad c = \frac{1}{\lvert\nabla T\rvert}, \qquad \mathrm E\lvert\hat{\mathbf g}\rvert^2 = \lvert\mathbf g\rvert^2 + 2\sigma_g^2 \text{ (2D)} \]

Two points (derivation)

\[ \frac{\sigma_c}{c} = \frac{\sqrt2\,\sigma\,c}{h} \]

Triangle (derivation)

\[ E = \begin{pmatrix} (\mathbf x_2 - \mathbf x_1)^\top \\ (\mathbf x_3 - \mathbf x_1)^\top \end{pmatrix}, \qquad E\,\mathbf g = \begin{pmatrix} T_2 - T_1 \\ T_3 - T_1 \end{pmatrix}, \qquad \operatorname{Cov}(\hat{\mathbf g}) = \sigma^2 E^{-1} S E^{-\top}, \quad S = \begin{pmatrix} 2 & 1 \\ 1 & 2 \end{pmatrix} \]

Local plane fit (derivation)

\[ \hat{\mathbf g} = \arg\min_{\mathbf g, t_0}\sum_i\bigl(T_i - t_0 - \mathbf g\cdot\mathbf x_i\bigr)^2, \qquad \operatorname{Cov}(\hat{\mathbf g}) = \sigma^2(X_c^\top X_c)^{-1} \approx \Bigl(\frac{2\sigma}{R\sqrt n}\Bigr)^2 I \text{ for a disk} \]

Potential-based velocity (derivation)

\[ \mathbf v = -\frac{\phi_t\,\nabla\phi}{\lvert\nabla\phi\rvert^2}, \qquad \frac{\hat c}{c} = 1 + \frac{(d/2)^2}{a^2} \text{ for a cross-shaped clique at the front's arrival along an axis} \]

m26 · Map quality and uncertainty

Symbols: \(\mathbf x\) site; \(\mathbf y\) measured times; \(m\) prior mean; \(k\) kernel with variance \(s^2\) and length scale \(\ell\); \(K\) kernel matrix; \(\mathbf k_*\) covariances with a new site; \(\sigma_n\) per-point error; \(A = K + \sigma_n^2 I\); \(r_i\) leave-one-out residual; \(\sigma_a\), \(\sigma_r\), \(\sigma_x\) annotation, reference and position SDs; \(c\) speed; \(\rho\) point density; \(D\) nearest-point distance. Model: Gaussian-process interpolation with independent per-point errors; random point placement for the density result.

Gaussian-process map (derivation)

\[ \mu(\mathbf x) = m + \mathbf k_*^\top A^{-1}(\mathbf y - m), \qquad \sigma^2(\mathbf x) = s^2 - \mathbf k_*^\top A^{-1}\mathbf k_*, \qquad A = K + \sigma_n^2 I \]

Leave-one-out (derivation)

\[ r_i = y_i - \mu_{-i}(\mathbf x_i) = \frac{\bigl[A^{-1}(\mathbf y - m)\bigr]_i}{\bigl[A^{-1}\bigr]_{ii}}, \qquad \sigma^2_{-i} = \frac{1}{\bigl[A^{-1}\bigr]_{ii}} \]

Per-point error budget (derivation)

\[ \sigma_n^2 = \sigma_a^2 + \sigma_r^2 + \Bigl(\frac{\sigma_x}{c}\Bigr)^2 \]

Point density (derivation)

\[ \mathrm E[D] = \frac{1}{2\sqrt\rho}, \qquad \operatorname{median}(D) = \sqrt{\frac{\ln 2}{\pi\rho}} \]

Next point (derivation)

\[ \mathbf x_{n+1} = \arg\max_{\mathbf x}\,\sigma_n^2(\mathbf x) \]

m37 · AI on the surface ECG

Symbols: \(z_c\) class score, \(p_c\) probability, \(y_c\) label, \(w[k]\) kernel weight, \(f_s\) sampling rate, \(k_l\) and \(s_l\) kernel width and stride of layer \(l\), \(RR_i\) an RR interval, \(\pi\) prevalence, Se and Sp sensitivity and specificity. Model: a classifier that outputs class probabilities; the prior-shift result needs calibrated probabilities and unchanged class-conditional recordings.

Softmax (derivation)

\[ p_c = \frac{e^{z_c}}{\sum_{j=1}^{C} e^{z_j}} \]

Cross-entropy gradient (derivation)

\[ \frac{\partial L}{\partial z_c} = p_c - y_c \]

Logistic regression (derivation)

\[ p = \sigma(\beta_0 + \boldsymbol\beta^{\mathsf T}\mathbf x), \qquad \sigma(u) = \frac{1}{1 + e^{-u}} \]

Convolution layer (derivation)

\[ y[n] = b + \sum_{k=0}^{K-1} w[k]\, x[s n + k] \]

Kernel frequency response (derivation)

\[ H(f) = \sum_{k=0}^{K-1} w[k]\, e^{-i 2\pi f k / f_s} \]

Receptive field (derivation)

\[ r = 1 + \sum_{l=1}^{L} (k_l - 1) \prod_{j<l} s_j \]

RR-interval features (derivation)

\[ \overline{RR}, \qquad \mathrm{CV} = \frac{s_{RR}}{\overline{RR}}, \qquad \mathrm{nRMSSD} = \frac{1}{\overline{RR}} \sqrt{\frac{1}{n-1}\sum_{i=1}^{n-1} \left(RR_{i+1} - RR_i\right)^2 } \]

Screening yield (derivation)

\[ N_{+} = N\left[\pi\,\mathrm{Se} + (1-\pi)(1-\mathrm{Sp})\right], \qquad \frac{N_{+}}{N\pi\,\mathrm{Se}} = \frac{1}{\mathrm{PPV}} \]

Prior-shift correction (derivation)

\[ p' = \frac{\dfrac{\pi'}{\pi}\, p}{\dfrac{\pi'}{\pi}\, p + \dfrac{1-\pi'}{1-\pi}\,(1-p)} \]

m38 · Wearables and implantable monitors

Symbols: \(RR_i\) beat intervals, \(\Delta_i\) pulse delays, \(\mathrm{IBI}_i\) pulse intervals, \(q\) and \(p\) per-window flag rates in regular rhythm and AF, \(k\) windows required in a row, \(n\) windows per day, \(L\) windows per episode, \(N\) patients, \(\lambda\) alerts per patient per day, \(\tau\) review minutes, \(\kappa\) and \(r\) shares of true alerts kept and false alerts removed. Model: independent windows and alerts.

Heart rate (derivation)

\[ \mathrm{HR} = \frac{60}{\overline{\mathrm{IBI}}} \quad \text{beats per minute} \]

Pulse intervals (derivation)

\[ \mathrm{IBI}_i = RR_i + \left(\Delta_{i+1} - \Delta_i\right) \]

Pulse-interval variance (derivation)

\[ \operatorname{Var}(\mathrm{IBI}) = \operatorname{Var}(RR) + 2\sigma_\Delta^2\,(1 - \rho) \]

False notifications, k in a row (derivation)

\[ \mathbb E[\text{runs}] = q^{k}\left[1 + (n - k)(1 - q)\right] \]

Episode caught (derivation)

\[ a_m = \sum_{j=0}^{k-1} p^{j}(1 - p)\, a_{m-j-1}, \qquad P(\text{episode caught}) = 1 - a_L \]

Review load (derivation)

\[ A = N\lambda, \qquad A_{\text{false}} = N\lambda\,(1 - \mathrm{PPV}), \qquad H = \frac{N\lambda\,\tau}{60}\ \text{hours} \]

PPV after a triage filter (derivation)

\[ \mathrm{PPV}' = \frac{\kappa\,\mathrm{PPV}}{\kappa\,\mathrm{PPV} + (1 - r)(1 - \mathrm{PPV})} \]

m39 · AI on electrograms and maps

Symbols: \(A\) mesh adjacency matrix, \(I\) identity, \(\tilde D\) degree matrix of \(A + I\), \(H\) vertex features, \(W\) shared weights, \(\sigma\) nonlinearity, \(\mathbf v\) eigenvector with \(v_i \propto \sqrt{\tilde d_i}\), \(P\) and \(T\) predicted and true regions, \(R\) and \(r\) disc radii, \(d\) centre distance, \(p_t\) probability of the true class, \(\alpha\) and \(\gamma\) focal-loss weight and focusing parameters, \(\pi_k\), \(\mu_k\), \(\sigma_k\) mixture weights, means and standard deviations, \(\gamma_{ik}\) responsibilities.

Graph convolution layer (derivation)

\[ H' = \sigma\!\left(\hat A H W\right), \qquad \hat A = \tilde D^{-1/2}\,(A + I)\,\tilde D^{-1/2} \]

Fixed point of the normalized adjacency (derivation)

\[ \hat A\, \mathbf v = \mathbf v, \qquad v_i \propto \sqrt{\tilde d_i} \]

Dice coefficient (derivation)

\[ \mathrm{Dice} = \frac{2\,\lvert P \cap T \rvert}{\lvert P \rvert + \lvert T \rvert} \]

Overlap of two discs (lens area) (derivation)

\[ \lvert P \cap T\rvert = r^2 \cos^{-1}\!\frac{d^2 + r^2 - R^2}{2dr} + R^2 \cos^{-1}\!\frac{d^2 + R^2 - r^2}{2dR} - \tfrac12\sqrt{(-d + r + R)(d + r - R)(d - r + R)(d + r + R)} \]

Focal loss (derivation)

\[ \mathrm{FL}(p_t) = -\alpha_t\,(1 - p_t)^{\gamma}\,\log p_t \]

Gaussian mixture responsibilities (derivation)

\[ \gamma_{ik} = \frac{\pi_k\, \mathcal N(x_i \mid \mu_k, \sigma_k)}{\sum_{j=1}^{K} \pi_j\, \mathcal N(x_i \mid \mu_j, \sigma_j)} \]

m40 · AI for cardiac imaging and substrate

Symbols: \(P\) and \(T\) predicted and true regions, \(\partial P\) and \(\partial T\) their surfaces, \(J\) Jaccard index, \(d(x, \partial T)\) distance from a point to a surface, \(\tau\) tolerance, \(\delta\) boundary offset, \(S\) and \(V\) surface area and volume, \(R\) sphere radius, \(t\) wall thickness, \(t_i\), \(e_i\) and \(r_i\) follow-up time, event indicator and risk score of patient \(i\).

Dice and Jaccard (derivation)

\[ \mathrm{Dice} = \frac{2J}{1 + J}, \qquad J = \frac{\mathrm{Dice}}{2 - \mathrm{Dice}} \]

Hausdorff distance (derivation)

\[ \mathrm{HD}(P, T) = \max\left\{ \max_{p \in \partial P} d(p, \partial T),\; \max_{q \in \partial T} d(q, \partial P) \right\} \]

Average symmetric surface distance (derivation)

\[ \mathrm{ASSD} = \frac{\int_{\partial P} d(p, \partial T)\, dp + \int_{\partial T} d(q, \partial P)\, dq}{\lvert\partial P\rvert + \lvert\partial T\rvert} \]

Normalized surface Dice (derivation)

\[ \mathrm{NSD}_\tau = \frac{\lvert\{p \in \partial P : d(p, \partial T) \le \tau\}\rvert + \lvert\{q \in \partial T : d(q, \partial P) \le \tau\}\rvert}{\lvert\partial P\rvert + \lvert\partial T\rvert} \]

Dice after a uniform boundary offset (derivation)

\[ \mathrm{Dice} \approx 1 - \frac{\lvert\delta\rvert\, S}{2V} \]

Concentric spheres (derivation)

\[ \mathrm{Dice} = \frac{2\min(R, R+\delta)^3}{R^3 + (R+\delta)^3} \approx 1 - \frac{3\lvert\delta\rvert}{2R} \]

Shifted wall (derivation)

\[ \mathrm{Dice} = \max\!\left(0,\; 1 - \frac{\lvert\delta\rvert}{t}\right) \]

Harrell’s concordance index (derivation)

\[ C = \frac{\sum_{i,j} \mathbb 1[t_i < t_j]\, e_i \left(\mathbb 1[r_i > r_j] + \tfrac12\, \mathbb 1[r_i = r_j]\right)}{\sum_{i,j} \mathbb 1[t_i < t_j]\, e_i} \]

m41 · Locating arrhythmia origins from the ECG

Symbols: \(\Omega_{ej}\) solid angle of surface element \(j\) seen from electrode \(e\), \(\tau_j\) activation times, \(c\) potential scale, \(T\) integration window, \(A = [X\;\mathbf 1]\) features with an intercept column, \(R\) origin coordinates, \(\mathbf u\), \(\mathbf v\) unit vectors on a sphere of radius \(R\), \(a_l\), \(b_l\) two ECGs in lead \(l\), \(K\) and \(\mathbf k_*\) kernel matrices, \(\ell\), \(\sigma_f\), \(\sigma_n\) kernel length scale, signal scale and noise, \(f^*\) best score so far, \(\xi\) margin, \(\Phi\) and \(\varphi\) standard normal distribution and density.

QRS integral in a double-layer model (derivation)

\[ x_e = \int_0^T \phi_e(t)\, dt = c \sum_j \Omega_{ej}\,(T - \tau_j)_+ = -c \sum_j \Omega_{ej}\, \tau_j \quad \text{when } T \ge \max_j \tau_j \]

Least-squares localization (derivation)

\[ \begin{bmatrix} W \\ \mathbf b^\top \end{bmatrix} = (A^\top A)^{-1} A^\top R \]

Error along a spherical surface (derivation)

\[ d = R \arccos(\mathbf u \cdot \mathbf v), \qquad \lvert \mathbf u - \mathbf v \rvert R = 2R\sin\frac{d}{2R} \le d \]

Pace-map similarity (derivation)

\[ \rho = \frac{1}{12} \sum_{l=1}^{12} \frac{\sum_t (a_l(t) - \bar a_l)(b_l(t) - \bar b_l)}{\sqrt{\sum_t (a_l(t) - \bar a_l)^2 \sum_t (b_l(t) - \bar b_l)^2}} \]

Gaussian-process posterior (derivation)

\[ \mu(\mathbf s) = m + \mathbf k_*^\top (K + \sigma_n^2 I)^{-1} (\mathbf f - m), \qquad \sigma^2(\mathbf s) = \sigma_f^2 - \mathbf k_*^\top (K + \sigma_n^2 I)^{-1} \mathbf k_* \]

Expected improvement (derivation)

\[ \mathrm{EI}(\mathbf s) = \Delta\, \Phi\!\left(\frac{\Delta}{\sigma}\right) + \sigma\, \varphi\!\left(\frac{\Delta}{\sigma}\right), \qquad \Delta = \mu(\mathbf s) - f^* - \xi \]

m42 · Physics-informed machine learning

Symbols: \(T\) activation time (ms), \(s = 1/v\) slowness (ms/mm), \(\mathbf x_i\), \(t_i\) measurement points and times, \(\mathbf x_c\) collocation points, \(\lambda\) physics weight (ms²), \(\mathbf c_k\), \(h\), \(a_k\) basis centres, width and coefficients, \(J\) Jacobian, \(\mathbf r\) residual vector, \(\mu\) damping, \(T_m\) ensemble members, \(k\) a scale factor.

Eikonal residual (derivation)

\[ r(\mathbf x) = s(\mathbf x)\,\lvert \nabla T(\mathbf x) \rvert - 1 \]

Physics-informed loss (derivation)

\[ L = \frac{1}{N_d} \sum_{i=1}^{N_d} \big(T(\mathbf x_i) - t_i\big)^2 + \frac{\lambda}{N_c} \sum_{c=1}^{N_c} r(\mathbf x_c)^2 \]

Radial-basis map and its gradient (derivation)

\[ T(\mathbf x) = a_0 + \sum_{k} a_k\, e^{-\lvert \mathbf x - \mathbf c_k\rvert^2/2h^2}, \qquad \nabla T(\mathbf x) = -\sum_k a_k\, \frac{\mathbf x - \mathbf c_k}{h^2}\, e^{-\lvert \mathbf x - \mathbf c_k\rvert^2/2h^2} \]

Levenberg–Marquardt step (derivation)

\[ \left(J^\top J + \mu\, \mathrm{diag}(J^\top J)\right) \boldsymbol\delta = -J^\top \mathbf r, \qquad J = \frac{\partial \mathbf r}{\partial \mathbf p} \]

Ensemble mean and spread (derivation)

\[ \bar T(\mathbf x) = \frac{1}{M}\sum_{m=1}^{M} T_m(\mathbf x), \qquad \sigma^2(\mathbf x) = \frac{1}{M-1}\sum_{m=1}^{M} \big(T_m(\mathbf x) - \bar T(\mathbf x)\big)^2 \]

Slowness scaling of activation times (derivation)

\[ T\big(\mathbf x;\, k\,s(\cdot)\big) = k\, T\big(\mathbf x;\, s(\cdot)\big) \]

m43 · Foundation models and large language models

Symbols: \(N\) batch size, \(S_{ik}\) similarity of anchor \(i\) and candidate \(k\) (unit embeddings), \(\tau\) temperature, \(P_{ik}\) row softmax of \(S/\tau\), \(Q, K, V\) queries, keys and values, \(d\) key length, \(\operatorname{tf}\) term count in a document, \(\operatorname{df}\) number of documents containing the term, \(\hat p = k/n\) observed accuracy, \(z\) normal quantile (1.96 for 95%), \(h\) interval half-width. Model: independent questions for the intervals; independent unit-variance query and key entries for the scaling.

InfoNCE contrastive loss (derivation)

\[ L = -\frac{1}{N}\sum_{i=1}^{N} \ln \frac{\exp(S_{ii}/\tau)}{\sum_{k=1}^{N} \exp(S_{ik}/\tau)} \]

Its gradient with respect to the similarities (derivation)

\[ \frac{\partial L}{\partial S_{ik}} = \frac{P_{ik} - \delta_{ik}}{N\tau} \]

Scaled dot-product attention (derivation)

\[ \operatorname{Attention}(Q, K, V) = \operatorname{softmax}\!\left(\frac{QK^\top}{\sqrt d}\right) V \]

Variance of a query-key score (derivation)

\[ \operatorname{Var}\!\left(\mathbf q^\top \mathbf k\right) = d, \qquad \operatorname{Var}\!\left(\mathbf q^\top \mathbf k / \sqrt d\right) = 1 \]

TF-IDF weight (smoothed) (derivation)

\[ w_{t,d} = \operatorname{tf}_{t,d}\left[\ln\frac{1 + N}{1 + \operatorname{df}_t} + 1\right] \]

Cosine similarity (derivation)

\[ \cos(\mathbf a, \mathbf b) = \frac{\mathbf a^\top \mathbf b}{\lVert\mathbf a\rVert\,\lVert\mathbf b\rVert} \]

Wilson score interval (derivation)

\[ \frac{\hat p + \dfrac{z^2}{2n} \pm z\sqrt{\dfrac{\hat p(1-\hat p)}{n} + \dfrac{z^2}{4n^2}}}{1 + \dfrac{z^2}{n}} \]

Questions needed for an interval of half-width h (derivation)

\[ n \approx \frac{z^2\,p(1-p)}{h^2} \]

m44 · Agentic AI in EP

Symbols: \(n\) steps in a task, \(p_i\) or \(p\) step success, \(c\) share of failed steps a check catches, \(p_r\) success of the retry, \(N\) tasks, \(r\) share of wrong outputs a reviewer catches, \(h\) share of hard tasks, \(p_e\), \(p_h\) step success in easy and hard tasks, \(\bar p\) mean step success. Model: steps fail independently (except in the mixture), one retry per caught failure.

Task success of a chain of independent steps (derivation)

\[ P_{\text{task}} = \prod_{i=1}^{n} p_i \;=\; p^n \ \text{ if all } p_i = p \]

With a check and one retry (derivation)

\[ P_{\text{task}} = \big[\,p + (1-p)\,c\,p_r\,\big]^n \;=\; \big[1 - (1-p)(1-c)\big]^n \ \text{ if } p_r = 1 \]

Wrong outputs that get past a reviewer (derivation)

\[ E = N\,(1 - P_{\text{task}})\,(1 - r) \]

Correlated errors: hard and easy tasks (derivation)

\[ P_{\text{task}} = (1-h)\,p_e^{\,n} + h\,p_h^{\,n} \;\ge\; \bar p^{\,n}, \qquad \bar p = (1-h)\,p_e + h\,p_h \]

m45 · Evidence and regulation of AI in EP

Symbols: \(Se\), \(Sp\) expected sensitivity and specificity, \(d\) interval half-width, \(\pi\) prevalence, \(z\) normal quantile (1.96 for 95%), TP and FP true and false positives among \(n\) patients, \(p_t\) threshold probability, \(x_t\) standardized daily statistic, \(\mu_0\) its in-control mean, \(k\) allowance, \(h\) alarm threshold. Model: independent patients and days; normal approximation for the sample size.

Patients needed for a sensitivity or specificity interval (derivation)

\[ n_{Se} = \frac{z^2\,Se\,(1-Se)}{d^2\,\pi}, \qquad n_{Sp} = \frac{z^2\,Sp\,(1-Sp)}{d^2\,(1-\pi)} \]

Net benefit of a decision rule (derivation)

\[ \mathrm{NB}(p_t) = \frac{\mathrm{TP}}{n} - \frac{\mathrm{FP}}{n}\cdot\frac{p_t}{1 - p_t} \]

CUSUM drift alarm (derivation)

\[ S_t = \max\bigl(0,\; S_{t-1} + x_t - \mu_0 - k\bigr), \quad S_0 = 0, \qquad \text{alarm when } S_t > h \]

m46 · Trustworthy AI in EP

Symbols: \(F\) model output, \(\mathbf x\) input, \(\mathbf x'\) baseline, \(\hat p_k\) predicted probability of class \(k\), \(n\) calibration examples, \(\alpha\) error rate, \(\hat q\) conformal threshold, \(\pi\) prevalence, TPR, FPR, PPV true-positive rate, false-positive rate and positive predictive value, \(\mathbf f\) features with training mean \(\boldsymbol\mu\) and covariance \(\Sigma\), \(k\) number of features, \(L\) loss, \(\varepsilon\) max-norm change, \(z = \mathbf w^\top\mathbf x + b\) linear score. Model: exchangeable calibration and test data for the conformal guarantee; Gaussian features for the chi-square reference.

Integrated gradients and completeness (derivation)

\[ \mathrm{IG}_i(\mathbf x) = (x_i - x'_i)\int_0^1 \frac{\partial F\bigl(\mathbf x' + \alpha(\mathbf x - \mathbf x')\bigr)}{\partial x_i}\,d\alpha, \qquad \sum_i \mathrm{IG}_i(\mathbf x) = F(\mathbf x) - F(\mathbf x') \]

Split conformal prediction sets (derivation)

\[ \hat q = s_{(\lceil (n+1)(1-\alpha)\rceil)}, \qquad C(\mathbf x) = \{k : 1 - \hat p_k(\mathbf x) \le \hat q\}, \qquad \Pr\bigl(y \in C(\mathbf x)\bigr) \ge 1 - \alpha \]

False-positive rate from prevalence, PPV and TPR (derivation)

\[ \mathrm{FPR} = \frac{\pi}{1-\pi}\cdot\frac{1-\mathrm{PPV}}{\mathrm{PPV}}\cdot\mathrm{TPR} \]

Mahalanobis unfamiliarity score (derivation)

\[ d^2(\mathbf x) = \bigl(\mathbf f(\mathbf x) - \boldsymbol\mu\bigr)^{\top}\Sigma^{-1}\bigl(\mathbf f(\mathbf x) - \boldsymbol\mu\bigr) \;\sim\; \chi^2_k \ \text{ for Gaussian features} \]

Fast gradient sign attack on a linear score (derivation)

\[ \mathbf x_{\text{adv}} = \mathbf x + \varepsilon\,\operatorname{sign}\bigl(\nabla_{\mathbf x} L\bigr), \qquad \lvert\Delta z\rvert \le \varepsilon\lVert\mathbf w\rVert_1, \qquad \varepsilon^\ast = \frac{\lvert z\rvert}{\lVert\mathbf w\rVert_1} \]

m47 · Data and deployment engineering

Symbols: \(d\) stored integer (adu), \(b\) baseline (adu), \(g\) gain (adu per physical unit), \(v\) physical value, \(m\) labellers each right with probability \(p\), \(f_i\) size of record \(i\)’s equivalence class, \(k\) smallest class size, \(J\) number of classes, \(n\) records, \(r\) real value, \(q\) integer, \(S\) scale, \(Z\) zero point, \(b\) bits (in the SQNR), \(\overline{x^2}\) mean square of the signal. Model: independent labellers; an attacker who knows the person is in the data; rounding error uniform within one step.

Header conversion (derivation)

\[ v = \frac{d - b}{g}, \qquad \lvert v_{\text{stored}} - v \rvert \le \frac{1}{2g} \]

Majority of labellers (derivation)

\[ P_{\text{maj}}(m, p) = \sum_{j=(m+1)/2}^{m} \binom{m}{j}\, p^{j} (1 - p)^{m-j} \]

Re-identification risk (derivation)

\[ r_i = \frac{1}{f_i}, \qquad R_{\max} = \frac{1}{\min_i f_i} = \frac{1}{k}, \qquad \bar R = \frac{1}{n}\sum_{i=1}^{n} \frac{1}{f_i} = \frac{J}{n} \]

Affine integer quantization (derivation)

\[ r = S\,(q - Z), \qquad S = \frac{r_{\max} - r_{\min}}{q_{\max} - q_{\min}}, \qquad Z = q_{\min} - \operatorname{round}\!\left(\frac{r_{\min}}{S}\right) \]

Quantization noise and SQNR (derivation)

\[ \sigma_e^2 = \frac{S^2}{12}, \qquad \mathrm{SQNR} = 10 \log_{10} \frac{\overline{x^2}}{S^2/12}\ \text{dB} \;\approx\; 6.02\,b + 1.76\ \text{dB for a full-scale sine} \]

m48 · Imaging across the EP workflow

Symbols: \(\sigma\) blur standard deviation (mm), \(\Gamma\) FWHM (mm), \(d\) separation (mm), \(S_A, S_B\) mean signals, \(\sigma_n\) noise standard deviation, \(n\) samples averaged, \(\Delta\) pixel spacing (mm), \(\sigma_s\) smoothing standard deviation (mm), \(C\) contrast, \(w\) wall thickness (mm), \(k\) required CNR, Se sensitivity, Sp specificity, \(\pi\) prevalence, \(p_{\text{post}}\) post-test probability. Model: Gaussian blur; white noise; sensitivity and specificity that do not change with prevalence.

Point-spread function, FWHM and two-point resolution (derivation)

\[ h(x) = \frac{1}{\sqrt{2\pi}\,\sigma}\,e^{-x^2/(2\sigma^2)}, \qquad \Gamma = 2\sqrt{2\ln 2}\;\sigma \approx 2.355\,\sigma, \qquad \text{dip between two points} \iff d > 2\sigma \approx 0.85\,\Gamma \]

Contrast-to-noise ratio (derivation)

\[ \mathrm{CNR} = \frac{|S_A - S_B|}{\sigma_n}, \qquad \mathrm{CNR}_n = \sqrt{n}\;\mathrm{CNR}_1 \]

Smoothing white noise (derivation)

\[ \sigma_n' = \frac{\sigma_n\,\Delta}{2\sqrt{\pi}\,\sigma_s}, \qquad \sigma_{\text{tot}}^2 = \sigma^2 + \sigma_s^2 \]

Visibility of a thin wall (derivation)

\[ \mathrm{CNR}_{\text{wall}} = \frac{C}{\sigma_n}\operatorname{erf}\!\left(\frac{w}{2\sqrt2\,\sigma}\right), \qquad w_{\min} = 2\sqrt2\,\sigma\,\operatorname{erf}^{-1}\!\left(\frac{k\,\sigma_n}{C}\right) \]

Likelihood ratios (derivation)

\[ \mathrm{LR}^+ = \frac{\mathrm{Se}}{1 - \mathrm{Sp}}, \qquad \mathrm{LR}^- = \frac{1 - \mathrm{Se}}{\mathrm{Sp}}, \qquad \frac{p_{\text{post}}}{1 - p_{\text{post}}} = \frac{\pi}{1 - \pi}\,\mathrm{LR} \]

m49 · Fluoroscopy and radiation protection

Symbols: \(I_0\) incident and \(I\) transmitted intensity, \(\mu\) linear attenuation coefficient (1/mm), HVL half-value layer, \(C\) subject contrast of an object of thickness \(t\) with attenuation \(\mu_o\) in background \(\mu_b\), \(N\) photons per pixel, CNR contrast-to-noise ratio, \(K\) air kerma, \(A\) beam area, \(d\) distance from the focal spot, \(P_{KA}\) kerma-area product, \(H\) staff dose, \(f_p\) pulse rate, \(q\) dose per pulse, \(\tau\) time, \(r\) distance from the patient, \(n\) shielding in half-value layers. Model: narrow monoenergetic beam; Poisson photon counts; scatter from a point at the patient.

Attenuation and half-value layer (derivation)

\[ I = I_0 \exp\!\left(-\int \mu(s)\, ds\right), \qquad \mathrm{HVL} = \frac{\ln 2}{\mu} \]

Subject contrast (derivation)

\[ C = 1 - e^{-(\mu_o - \mu_b)\, t} \;\approx\; (\mu_o - \mu_b)\, t \ \text{ when small} \]

Quantum noise (derivation)

\[ \mathrm{CNR} = \frac{N_b - N_o}{\sqrt{N_b}} = C\sqrt{N}, \qquad N_{\text{needed}} = \left(\frac{\mathrm{CNR}^\ast}{C}\right)^2 \]

Kerma-area product along the beam (derivation)

\[ K(d) = K(d_0)\left(\frac{d_0}{d}\right)^2, \qquad A(d) = A(d_0)\left(\frac{d}{d_0}\right)^2 \;\Rightarrow\; P_{KA} = K\,A \ \text{ is the same at every } d \]

Staff dose: time, distance, shielding (derivation)

\[ H \;\propto\; \frac{f_p\, q\, \tau}{r^2}\; 2^{-n} \]

m50 · Intracardiac echocardiography

Symbols: \(t\) echo round-trip time, \(c\) speed of sound (\(c_{\text{assumed}}\) used by the scanner, \(c_{\text{true}}\) in the tissue), \(z\) depth, \(\hat z\) displayed depth, \(f\) frequency, \(\lambda\) wavelength, \(n\) cycles per pulse, \(\alpha\) attenuation slope in dB/(cm·MHz), \(B\) signal budget in dB, \(N\) elements spaced \(d\) apart at positions \(x_k\), \(\theta_0\) steering angle, \(D\) image depth, \(N_{\text{lines}}\) lines per frame. Model: straight-line propagation at one speed of sound; attenuation linear in frequency; far-field array factor of point elements.

Echo time to depth (derivation)

\[ z = \frac{c\,t}{2}, \qquad \hat z = z\,\frac{c_{\text{assumed}}}{c_{\text{true}}} \]

Axial resolution (derivation)

\[ \lambda = \frac{c}{f}, \qquad \Delta z_{\text{axial}} = \frac{n\,\lambda}{2} \]

Attenuation and reach (derivation)

\[ L = 2\,\alpha\,f\,z \ \text{dB}, \qquad z_{\max} = \frac{B}{2\,\alpha\,f} \]

Phased-array steering and grating lobes (derivation)

\[ \tau_k = \frac{x_k \sin\theta_0}{c}, \qquad \mathrm{AF}(\theta) = \frac{1}{N}\left|\sum_{k=0}^{N-1} e^{\,i 2\pi k \frac{d}{\lambda}(\sin\theta - \sin\theta_0)}\right|, \qquad \sin\theta_m = \sin\theta_0 + m\frac{\lambda}{d} \]

Frame rate (derivation)

\[ \mathrm{FR} = \frac{c}{2\,D\,N_{\text{lines}}} \]

m51 · Cardiac CT

Symbols: \(\mu\) linear attenuation coefficient (\(\mu_w\) water, \(\mu_a\) air), \(c\) iodine concentration in mg/mL, \((\mu/\rho)_{\mathrm I}\) iodine mass attenuation, \(f(x,y)\) image, \(p_\theta(s)\) projection at angle \(\theta\) and offset \(s\), \(P_\theta(k)\) its 1-D Fourier transform, \(F\) the image’s 2-D Fourier transform, \(q_\theta\) filtered projection, \(d\) detector spacing, \(N_0\) photons per ray, \(T_{\text{rot}}\) rotation time, \(N_{\text{src}}\) X-ray sources, \(v\) speed, \(b\) blur, \(D\) vessel diameter, \(w\) wall thickness, \(\sigma\) blur standard deviation. Model: monoenergetic beam; parallel-beam geometry; Poisson counts; constant speed during the window; Gaussian blur.

Hounsfield units (derivation)

\[ \mathrm{HU} = 1000\,\frac{\mu - \mu_w}{\mu_w - \mu_a} \]

Iodine contrast (derivation)

\[ \Delta\mathrm{HU} = 1000\,\frac{c \cdot 10^{-3}\,(\mu/\rho)_{\mathrm I}}{\mu_w} \]

Projection (Radon transform) (derivation)

\[ p_\theta(s) = \iint f(x, y)\,\delta(x\cos\theta + y\sin\theta - s)\,dx\,dy \]

Central-slice theorem (derivation)

\[ P_\theta(k) = \int p_\theta(s)\,e^{-i2\pi k s}\,ds = \iint f(x, y)\,e^{-i2\pi k (x\cos\theta + y\sin\theta)}\,dx\,dy = F(k\cos\theta,\,k\sin\theta) \]

Filtered back-projection (derivation)

\[ f(x, y) = \int_0^{\pi} q_\theta(x\cos\theta + y\sin\theta)\,d\theta, \qquad q_\theta = \mathcal{F}^{-1}\{\,|k|\,P_\theta(k)\,\} \]

Ram-Lak kernel (derivation)

\[ h[0] = \frac{1}{4d^2}, \qquad h[n] = -\frac{1}{\pi^2 n^2 d^2}\ \ (n \text{ odd}), \qquad h[n] = 0\ \ (n \text{ even},\ n \neq 0) \]

Quantum noise (derivation)

\[ \operatorname{Var}\!\left[-\ln\frac{N}{N_0}\right] \approx \frac{1}{\bar N} = \frac{e^{p}}{N_0} \]

Gating window and motion blur (derivation)

\[ \Delta t \approx \frac{T_{\text{rot}}}{2\,N_{\text{src}}}, \qquad b = v\,\Delta t \]

A smeared disk (derivation)

\[ I(x, y) = \frac{\left|\,[x - b/2,\ x + b/2] \cap [-c,\ c]\,\right|}{b}, \qquad c = \sqrt{D^2/4 - y^2} \]

A blurred thin wall (derivation)

\[ g(x) = \tfrac12\left[\operatorname{erf}\!\left(\frac{x + w/2}{\sqrt2\,\sigma}\right) - \operatorname{erf}\!\left(\frac{x - w/2}{\sqrt2\,\sigma}\right)\right], \qquad g(0) = \operatorname{erf}\!\left(\frac{w}{2\sqrt2\,\sigma}\right) \]

m52 · Cardiac MRI

Symbols: \(\gamma\) gyromagnetic ratio, \(B_0\) static field, \(f\) Larmor frequency, \(M_z\) longitudinal magnetization (\(M_0\) at equilibrium), \(T_1\) longitudinal relaxation time (\(T_{1m}\) myocardium, \(T_{1s}\) scar, \(T_{1,0}\) native), \(r_1\) relaxivity, \([\mathrm{Gd}]\) concentration, \(R_1 = 1/T_1\), ECV extracellular volume fraction, Hct hematocrit, TI inversion time, \(I\) image intensity, \(\theta\) threshold, \(B\) burden. Model: ideal inversion; readout ignored; gadolinium at equilibrium between blood and extracellular space.

Larmor relation (derivation)

\[ f = \frac{\gamma}{2\pi}\,B_0 \]

Inversion recovery (derivation)

\[ M_z(t) = M_0\left(1 - 2e^{-t/T_1}\right), \qquad M_z = 0 \ \text{at}\ t = T_1 \ln 2 \]

Relaxivity (derivation)

\[ \frac{1}{T_1} = \frac{1}{T_{1,0}} + r_1\,[\mathrm{Gd}] \]

Extracellular volume (derivation)

\[ \mathrm{ECV} = (1 - \mathrm{Hct})\,\frac{\Delta R_{1,\text{myo}}}{\Delta R_{1,\text{blood}}} \]

Inversion time for LGE (derivation)

\[ \mathrm{TI}^* = \frac{\ln(T_{1m}/T_{1s})}{1/T_{1s} - 1/T_{1m}} \]

Image intensity ratio and burden (derivation)

\[ \mathrm{IIR}_i = \frac{I_{\text{wall},i}}{\bar I_{\text{blood}}}, \qquad B(\theta) = \frac{1}{N}\sum_{i=1}^{N} \mathbf{1}\!\left[\mathrm{IIR}_i > \theta\right] \]

m53 · Lesion imaging and real-time MRI

Symbols: \(w\) lesion line width, \(g\) gap width, \(\sigma\) blur standard deviation, \(\theta\) threshold, \(B, G\) blurred box profiles, \(N_{\text{lines}}\) lines per frame, TR repetition time, \(R\) acceleration factor, \(v\) speed, \(d\) motion per frame. Model: straight line, rectangular gap, isotropic Gaussian blur; line-by-line acquisition.

A gap in a blurred lesion line (derivation)

\[ I(x, y) = B(y)\,\bigl[1 - G(x)\bigr], \qquad I_{\text{line}} = \operatorname{erf}\!\left(\frac{w}{2\sqrt2\,\sigma}\right), \qquad I_{\text{gap}} = I_{\text{line}}\left[1 - \operatorname{erf}\!\left(\frac{g}{2\sqrt2\,\sigma}\right)\right] \]

Real-time frame time (derivation)

\[ T_{\text{frame}} = \frac{N_{\text{lines}}\,\mathrm{TR}}{R}, \qquad \text{frame rate} = \frac{1}{T_{\text{frame}}}, \qquad d = v\,T_{\text{frame}} \]

m54 · Nuclear imaging

Symbols: \(A\) activity (\(A_0\) at injection, \(A'\) net injected and decay-corrected), \(\lambda\) decay constant, \(T_{1/2}\) half-life, \(C\) activity concentration, \(W\) body weight, \(w\) wall thickness, \(\sigma\) blur standard deviation, RC recovery coefficient, \(R\) true ratio of follow-up to baseline, \(R_m\) measured ratio, \(c\) per-scan coefficient of variation, \(\Phi\) standard normal distribution function. Model: physical decay only; 1 mL of tissue = 1 g; slab with Gaussian blur; independent log-normal noise per scan.

Radioactive decay (derivation)

\[ A(t) = A_0\,e^{-\lambda t}, \qquad \lambda = \frac{\ln 2}{T_{1/2}} \]

Standardized uptake value (derivation)

\[ \mathrm{SUV} = \frac{C}{A'/W} \]

Recovery coefficients of a wall (derivation)

\[ \mathrm{RC}_{\max} = \operatorname{erf}\!\left(\frac{w}{2\sqrt2\,\sigma}\right), \qquad \mathrm{RC}_{\text{mean}} = \operatorname{erf}(b) - \frac{1 - e^{-b^2}}{b\sqrt\pi}, \quad b = \frac{w}{\sqrt2\,\sigma} \]

Noise and a response rule (derivation)

\[ \ln R_m \sim \mathcal N\!\left(\ln R,\ 2\ln(1 + c^2)\right), \qquad P(\text{called a response}) = \Phi\!\left(\frac{\ln 0.75 - \ln R}{\sqrt{2\ln(1 + c^2)}}\right) \]

m55 · Capstone

Symbols: SOD source-to-object (isocenter) distance, SID source-to-detector distance, \(x, z\) lateral and head-foot positions, \(y\) depth beyond the isocenter, \(u, v\) detector coordinates, \(M\) magnification, \(N\) number of landmarks, FLE, FRE and TRE fiducial localization, fiducial registration and target registration errors, \(d_k\) distance of the target from principal axis \(k\) of the landmarks, \(f_k^2\) mean squared distance of the landmarks from that axis, \(\sigma_i\) RMS error of step \(i\). Model: point source and flat detector; rigid registration with independent isotropic landmark error (first order); independent zero-mean errors.

C-arm projection (derivation)

\[ u = \frac{\mathrm{SID}}{\mathrm{SOD} + y}\,x, \qquad v = \frac{\mathrm{SID}}{\mathrm{SOD} + y}\,z, \qquad M(y) = \frac{\mathrm{SID}}{\mathrm{SOD} + y} \]

Fiducial and target registration error (derivation)

\[ \langle \mathrm{FRE}^2 \rangle = \left(1 - \frac{2}{N}\right)\langle \mathrm{FLE}^2\rangle, \qquad \langle \mathrm{TRE}^2(\mathbf r) \rangle \approx \frac{\langle \mathrm{FLE}^2\rangle}{N}\left(1 + \frac13\sum_{k=1}^{3}\frac{d_k^2}{f_k^2}\right) \]

Error budget (derivation)

\[ \sigma_{\text{total}} = \sqrt{\sum_i \sigma_i^2} \]