Open-Motion SDK: how should the real-time BFI output be interpreted?

[dblizinski]

Posted on behalf of Yueh-Chi, who raised this with the team — sharing here so the answer is public and searchable.

While reviewing the OpenMotion SDK, I want to confirm how the real-time BFI output should be interpreted. From my reading, per-frame speckle contrast K is computed from the CMOS histogram statistics, then converted to BFI by linear scaling against the per-camera C_min and C_max:

BFI = 10 × [1 − (K − C_min) / (C_max − C_min)]

1. Is this the intended formula for the real-time BFI output?

2. This differs from the speckle-contrast flow model we usually use, where measured contrast relates to the field autocorrelation through the exposure-time integral

K²(ρ,T) = (2β/T) ∫₀ᵀ (1 − τ/T) [g₁(ρ,τ)]² dτ

with g₁ modeled via the correlation diffusion equation and flow entering through αD_B (often simplified to flow ∝ 1/K²). Is the SDK BFI intended mainly as a normalized 0–10 visualization/index value, rather than a model-based physical flow estimate?

3. For quantitative analysis, would you recommend using the SDK-reported BFI directly, or working from the underlying contrast values and applying a separate speckle-contrast model depending on the application?

4. Is my understanding correct that C_min and C_max are fixed per-module/per-camera calibration constants loaded from the console/device calibration, rather than estimated per scan?

Thanks for the help — I want to make sure we interpret the SDK output correctly in our phantom and pressure-test analyses.

[bahartl]

1. Yes, and for individuals not already aware, this can be confirmed in our SciencePipeline.md document.

2. We've included extensive discussion regarding this in our 2024 publication in Neurophotonics, please specifically see section "2.5 Optical and TCD Data Processing". Briefly, we acknowledge that there are many models for quantifying static and dynamic optical properties in tissue that have been proposed throughout the diffuse optics literature. We chose a method agnostic to specific models, as choosing one and its associated assumptions is not needed to demonstrate the sensitivity of our device to changes in blood flow and volume. Instead, we defined our index using the simplest linear formula, which accurately captures the general behavior of small blood flow variations. This linear definition is also robust against experimental noise, avoiding the roughly twofold noise amplification seen in non-linear models that could otherwise obscure subtle features in the waveforms.

3. The reported BFI is the metric we chose to use for a real-time displayed value as well as quantitative analysis based on our extensive experience with pulsed-SCOS in the clinical setting. However, we understand that as research users, individuals may want to process the data using their own novel methods and models. The raw calculated contrast is openly available for such purposes.

4. Presently, yes, these are fixed for each individual camera. In the future we do hope to move towards a dynamic assessment of these values.