Image arithmetic as signal combination: Gonzalez & Woods describe arithmetic operations on images as fundamental tools for combining information. Subtraction reveals change (difference images). Division normalizes (ratio images). These same operations applied per-cell rather than per-pixel serve the same purpose: revealing information that neither operand contains alone.
The power of ratios — lessons from fluorescence: Dobrucki describes ratiometric dyes for calcium and pH measurement. The ratio of emission at two wavelengths is independent of local dye concentration, eliminating artifacts from uneven dye loading, photobleaching, and focus variation. The same principle applies to biomarker ratios: the ratio of two markers measured in the same cell cancels cell-to-cell variation in staining efficiency, section thickness, and illumination. What remains is the relative biological expression level.
Signal combination theory: Vetterli's signal processing framework shows that combining signals (addition, subtraction, multiplication) creates new signals whose frequency content is related to but different from the inputs. In biological terms, a ratio measurement has different statistical properties than either individual measurement — it may be more normally distributed, have less batch-to-batch variation, or provide better separation between cell populations.
When ratios fail: Division by zero or near-zero values produces extreme outliers. If the denominator measurement can be very small (e.g., in cells with low CK expression), the ratio becomes unstable. Practical solutions include adding a small constant to the denominator, using log-ratios (which compress the dynamic range), or filtering out cells with denominator values below a minimum threshold before computing the ratio.
Ratios are powerful because they cancel out confounding factors. Just as ratiometric calcium dyes eliminate artifacts from uneven dye loading, biomarker ratios eliminate artifacts from uneven staining. The PD-L1/CK ratio is independent of how thickly the section was cut or how strongly it was stained — it captures relative expression. But beware of dividing by very small numbers, which produces extreme, unreliable values.
