The analysis cannot extract information that the digitization didn't capture. The rule sounds obvious and is easy to forget. Pawley describes imaging as a chain of specimen preparation, contrast formation, photon collection, digitization, processing and measurement, and each step constrains the next.
In digital pathology the scanner is the digitization step. Its spatial resolution (pixel size), spectral resolution (number of channels), bit depthLoading... and noise define the information content every later step works from.
So the scanning modality should match the analytical need, a point Combs and Shroff make. BrightfieldLoading... whole-slide scanning (3-channel RGB at 20x to 40x) is sufficient for H&E and standard IHCLoading.... Multiplex IFLoading... with spectral unmixingLoading... needs multispectral scanning (5 to 9 narrowband channels), and thick sections need confocal-like optical sectioning. Scanning with more capability than the analysis requires is slower and more expensive for nothing; scanning with less limits the results.
Then there is the arithmetic. A single 40x scan is about 100,000 × 50,000 pixels × 3 channels, or 15 billion pixel values, and a 500-slide study holds about 7.5 terabytes of raw image data. At about 500,000 cells per slide, single-cell analysis yields 250 million cells with dozens of measurements each. Nobody reviews that by hand, which is why digital pathology and tissue cytometryLoading... are natural partners.
Digital pathology turns glass slides into data a computer can process. The scan's pixel size sets what the analysis can detect, and features smaller than a pixel are invisible to it. A typical study produces terabytes of images and millions of cell measurements, far beyond what manual analysis can handle.
