Gradient operators for edge detection: Gonzalez & Woods describe edge detection as computing the first derivative of the image. The Sobel operator combines smoothing in one direction with differentiation in the perpendicular direction — it detects edges while suppressing noise along them. For membrane detection, oriented versions of these operators are applied at 8-16 angles to capture membranes at any orientation.
The Z-resolution problem: Pawley's Confocal Handbook notes that axial (Z) resolution is always worse than lateral — typically 500-700 nm versus 200 nm. Cell membranes are approximately 7-10 nm thick, so they are always below the diffraction limit and appear as diffuse boundaries in the image. What you see as a "membrane" in a fluorescence image is actually the PSF-broadened representation of the membrane — several hundred nanometers wide in the image, not the nanometer-thin physical structure. This inflation is even more severe in Z, where the "missing cone" in the optical transfer function further blurs axial membrane contrast.
Why membrane detection is harder than nuclear detection: Nuclei are bright blobs on dark background — high contrast, favorable geometry. Membranes are thin lines between cells — low contrast (membrane signal often comparable to cytoplasmic background), unfavorable geometry (1 pixel wide vs. 50 pixels across for a nucleus), and variable orientation. The signal-to-noise ratio per pixel is typically much lower for membranes than nuclei, making thresholding less reliable. This is why gradient-based methods (which enhance edges) outperform simple intensity thresholding for membrane structures.
Dobrucki's misrepresentation warning: A 7 nm membrane appears as a ~250 nm structure in the image — a 35-fold inflation. The measured "membrane intensity" integrates signal from the membrane itself plus any cytoplasmic or extracellular fluorescence within the PSF footprint. This means membrane measurements are inherently contaminated by non-membrane signal, and the degree of contamination depends on the optical system. Deconvolution can partially mitigate this but cannot fully separate a 7 nm structure from its 250 nm image.
Membranes are the hardest structures to detect because they're incredibly thin — just 7-10 nanometers — but appear as ~250 nm wide lines in the image due to the optics. Edge-detection filters (Sobel-type operators) applied at multiple angles enhance these thin structures above the background. The fundamental challenge is that membrane signal is always mixed with nearby cytoplasmic signal because the microscope cannot resolve structures this thin.
