What it is — Quantitative phase imaging (QPI) measures the phase retardation a specimen imposes on a transmitted wavefront. That retardation is the optical path difference: the refractive-index contrast between the specimen and its medium, integrated along the optical axis. Nothing is stained and nothing is excited, so the same cells can be watched for hours or days without photobleaching or phototoxicity. The family is wide — off-axis and in-line digital holographic microscopy, quadriwave lateral shearing interferometry, phase from defocus, differential phase contrast, Fourier ptychography — but the measurand is the same in all of them, and it has units. That is the whole argument for the modality: the output is a physical quantity, not an intensity that means whatever the staining protocol made it mean.
Dry mass is the headline, and it is conditional — Barer's relation makes refractive index linear in the concentration of non-aqueous content, so summing phase over a cell's footprint and scaling by the refraction increment gives its dry mass. This is why QPI is used for growth rate: mass accumulation is measured directly, in picograms, on living unlabelled cells. But the simple estimator holds only when the medium's refractive index matches the intracellular solute. When it does not, what is measured is relative dry mass — and the gap is not academic. For a ten-micron cell in a medium only two thousandths of a refractive-index unit above the intracellular value, the dry mass is underestimated by tens of picograms. The medium's refractive index is therefore a measurement parameter, not an environmental detail, and a pipeline that does not capture it cannot state which of the two quantities it reported. This is Measure Where It Matters in units of mass.
2D phase and 3D refractive index are different measurements — A single phase image collapses index and thickness into one number, and they cannot be separated without a prior: the index of an object cannot be computed from one phase image without knowing its shape, nor its shape without knowing its index. For suspended cells and beads a sphere assumption closes the problem and yields an absolute index and radius; for an adherent cell of varying density it does not. Optical diffraction tomography — holotomography — removes the prior by illuminating from many angles and inverting the scattering, returning a true refractive-index volume. Conflating the two is the most common error in reading a QPI result, and the cost of the tomographic version is axial: rotating the illumination rather than the sample leaves a cone of spatial frequencies unmeasured, so axial resolution is structurally worse than lateral and the elongation artefacts that follow are geometry, not noise.
Where it breaks — The tempting summary is that label-free removes batch variance, and it is half true: there is no staining protocol to drift. What replaces it is optical variance. Background tilt and curvature from the optics, residual defocus, aberration, and medium composition all enter the phase map as signal, and every one of them is indistinguishable from biology in a single image. Phase unwrapping adds a failure mode with no counterpart in fluorescence: where the true phase exceeds one wavelength the recovered map can wrap, and an unwrapping error is a smooth, plausible-looking surface with a whole cycle missing from part of it — an artefact that survives every downstream step and biases mass in one direction. These are the Uncertainty, Drift, and Failure Modes in Bioimage Analysis this modality generates, and they are optical rather than biological, which is precisely why they are missed by QC written for stained images.
The standards position — Better than it looks in one respect and worse in another. The container problem is largely solved: OME-Zarr carries phase and refractive-index volumes, and the dominant vendor format is proprietary in layout but built on HDF5, so it is readable without vendor software. What does not exist is agreed reporting: no minimum-information checklist for a phase measurement, no convention for stating medium refractive index, refraction increment, wavelength and unwrapping method alongside a mass figure, and no reference material against which an instrument's phase accuracy is routinely verified. A dry mass without those parameters is not reproducible by anyone else, which makes this a working case of Validation Without a Ground Truth rather than an exception to it. The absence is stated plainly here instead of borrowing a checklist written for a different measurement.
increment is a number, not a measurement. Whether it is absolute or relative is a property of the imaging conditions, not of the analysis — and it cannot be recovered after the fact.
How we run it — the Label-Free / QPI pipeline, stage by stage — the nine-stage imaging backbone reshaped for this modality, with the tools, sub-steps and deliverable at each stage.