As a novice microscopist processing my first super-resolution datasets, I often wondered: Have I optimized this image, or pushed it too far? Are these intricate patterns real, or just artifacts of overprocessing? Even today, with years of experience, I still ask myself: Am I extracting the maximum useful information, finding the ideal balance between resolution and fidelity?
When imaging simple, known structures like microtubules, it is straightforward to identify parameters that reveal smooth, continuous filaments. But to gain biological insights, we aim to probe more complex and less characterized systems, ideally in living cells. We want to push the limits of super-resolution techniques to see the nanoscale architecture of dynamic protein assemblies and organelles. And in these cases, the line between real and artifact is blurred. There are no easy answers. As seasoned microscopists told me early on: "With practice, your trained eye will recognize oversharpening. You'll develop an intuition." But while experience helps, we need to move beyond rules of thumb.
Fundamentally, we want to translate the human perception of image quality into quantifiable metrics. We want to teach the computer to "see" like a microscopist, finding the sweet spot between resolution and artifacts. This is the central challenge in super-resolution image processing. And the solution could accelerate discovery across the biological sciences.
When developing enhanced super-resolution radial fluctuation (eSRRF) microscopy, we implemented exactly this concept . Unlike previous algorithms like SRRF , eSRRF optimizes parameters based on quantitative measures computed from the data itself.
The optimization uses two image quality metrics from the NanoJ-SQUIRREL engine . The first, Fourier ring correlation, estimates resolution. This reveals how much structural detail is present. The second metric analyzes global patterns to quantify fidelity. This detects artifacts like rippling or checkerboards that arise from overprocessing.
The ideal parameters would maximize both - extreme resolution risks artifacts, while an overly smooth image loses information. To balance them, eSRRF computes their harmonic mean, termed the "Quality and Resolution" (QnR) factor. By sweeping parameters and maximizing QnR, the algorithm identifies settings that optimize the trade-off.
This data-driven parameter optimization is now built into the Fiji plugin NanoJ-eSRRF. With a click, users can see how choices affect resolution and fidelity, empowering them to make informed decisions. The computer provides an objective starting point to identify the sweet spot.
But this is not a black box approach. Image quality varies across the sample, so user guidance is still required. Filaments may be optimally rendered, but dimmer regions risk artifacts. Parameters ideal for one structure may not be for others. Subjectivity remains inherent to microscopy.
The goal of eSRRF is thus not automation, but augmentation. It complements human judgement with data-driven suggestions. Together, the two can achieve more than either alone. The computer provides objective metrics, but the microscopist brings context and experience.
 Laine, R.F. & Heil, H.S. et al. High-fidelity 3D live-cell nanoscopy through data-driven enhanced super-resolution radial fluctuation. Nat. Methods (2023) https://doi.org/10.1038/s41592-023-02057-w
 Gustafsson, N. et al. Fast live-cell conventional fluorophore nanoscopy with ImageJ through super-resolution radial fluctuations. Nat.Commun. 7 , 12471 (2016).
 Culley, S. et al. Quantitative mapping and minimization of super-resolution optical imaging artifacts. Nat. Methods 15 , 263–266 (2018).