Super-Resolution Microscopy via Optical Fluctuation Imaging
Implements SOFI — a computational super-resolution technique extracting sub-diffraction spatial information from temporal fluorescence fluctuations of quantum dots. First successful SOFI implementation in Chile. Resolves ~120 nm features below the 232 nm diffraction limit.
Business Context
Conventional optical microscopy cannot resolve features smaller than approximately 232 nm — the Abbe diffraction limit. Super-resolution techniques like PALM and STORM break this barrier through single-molecule localization, but they require extreme emitter sparsity, special photoswitchable probes, and tens of thousands of frames — making them slow, expensive, and incompatible with many biological specimens and live-cell imaging.
Strategic Value
SOFI offers a fundamentally different approach to super-resolution: instead of localizing individual molecules, it exploits the temporal statistics of independently blinking emitters to narrow the effective point spread function computationally. It works with densely labeled samples, standard widefield hardware, and requires far fewer frames (500-1000 vs 10000-50000 for localization methods). This implementation — the first successful SOFI in Chile, developed at SCIAN-Lab/BNI in collaboration with the University of Göttingen — computes cumulants from orders 2 through 6, achieving up to 2.45x resolution improvement. The modernized Python/FastAPI web application preserves the full computational pipeline (cumulant computation, Fourier interpolation, Wiener and Richardson-Lucy deconvolution, nth-root linearization) while making it accessible through a browser.
The Challenge
Optical microscopy is limited by diffraction: d_min = 0.61λ/NA ≈ 232 nm for 532 nm excitation with 1.4 NA objective. PALM/STORM require single-molecule sparsity conditions. SOFI offers an alternative using temporal statistics of blinking emitters.
Our Approach
nth-order cumulant computation narrows effective PSF by √n. Cumulants orders 2-6 with consecutive time lags to eliminate shot noise bias. Synthetic quantum dot simulator with power-law on/off statistics. Fourier-domain zero-padding for sub-pixel enhancement, Wiener and Richardson-Lucy deconvolution, nth-root linearization correction.
Key Performance Indicators
| KPI | Baseline | Result | Impact |
|---|---|---|---|
| Resolution Improvement | 232 nm diffraction limit | ~120 nm (6th order = 2.45x improvement) | Sub-diffraction imaging |
| Historical Significance | No SOFI implementation in Chile | First successful SOFI in Chile | National capability established |
Architecture
sofi qdots
Breaking the Diffraction Limit
Conventional optical microscopy cannot resolve features smaller than ~232 nm — the Abbe diffraction limit (d_min = 0.61λ/NA for 532 nm excitation with a 1.4 NA objective). Techniques like PALM and STORM break this barrier through single-molecule localization, but they require extreme emitter sparsity (only a few fluorophores active per frame), special photoswitchable probes, and thousands of frames — making them slow and incompatible with many biological samples.
SOFI offers a fundamentally different approach. Instead of localizing individual molecules, it exploits the temporal statistics of independently blinking emitters to narrow the effective point spread function computationally. It works with densely labeled samples, standard widefield hardware, and requires far fewer frames (~500–1,000 vs ~10,000–50,000 for localization methods).
The Mathematics
The fluorescence intensity at each pixel is a sum of blinking emitter contributions convolved with the microscope’s PSF:
F(r,t) = Σₖ εₖ · sₖ(t) · U(r - rₖ) + noise
The key mathematical insight: the nth-order cumulant of this signal simplifies to:
Cₙ(r) = Σₖ εₖⁿ · κₙ[sₖ] · Uⁿ(r - rₖ)
Because cumulants are additive for independent variables, cross-terms between different emitters vanish — a property that moments do not share. The PSF appears raised to the nth power, narrowing the effective width by √n:
| Order | Resolution Gain | Effective Resolution |
|---|---|---|
| 2nd | 1.41x | ~164 nm |
| 4th | 2.00x | ~116 nm |
| 6th | 2.45x | ~95 nm |
The Processing Pipeline
The implementation computes cumulants from orders 2 through 6 using consecutive time lags to eliminate shot noise bias. The 4th-order cumulant requires subtracting three pair-partition products; the 6th-order requires removing 40 partition terms — the combinatorial complexity grows rapidly.
A synthetic quantum dot simulator generates realistic blinking with power-law on/off statistics P(t) ~ t⁻ᵅ (α ≈ 1.5), producing Levy-type statistics where the mean dwell time diverges. This heavy tail means quantum dots can stay dark for very long periods — which is why SOFI needs hundreds of frames for reliable statistics.
Fourier-domain zero-padding provides sub-pixel grid enhancement before cumulant computation. Two deconvolution strategies sharpen the result further: Wiener filtering in the frequency domain with noise regularization, and Richardson-Lucy iterative maximum-likelihood deconvolution preserving non-negativity. An nth-root linearization corrects the εⁿ brightness nonlinearity that would otherwise create extreme contrast ratios between emitters (a 2:1 brightness ratio becomes 64:1 at 6th order).
Historical Significance
This was the first successful SOFI implementation in Chile, developed at SCIAN-Lab (BNI, Universidad de Chile) in collaboration with the III Physics Institute, University of Gottingen, Germany (2012–2014). The original MATLAB codebase was modernized as a Python/FastAPI web application with interactive visualization, WebSocket progress streaming, and 38+ automated tests — preserving the full computational pipeline while making it accessible through a browser.
Technology Stack
Application Screenshots
Technical Diagrams
sofi cumulant orders
sofi cumulant vs moment
sofi pipeline
sofi vs palm