SANS Dataset: Soft Matter & Fundus Imaging

• SANS dataset is comprised of two distinct collections: one with absolute-scale neutron scattering of HA-lysozyme complexes and another featuring ISS-acquired fundus images for SANS assessment.
• The soft matter dataset uses controlled variables such as pH and ionic strength to reveal rod-like and cluster scattering regimes with quantitative modeling of protein–polyelectrolyte interactions.
• The fundus image dataset employs standardized preprocessing and synthetic bicubic degradation to evaluate super-resolution performance through PSNR and SSIM metrics.

The SANS dataset refers to two distinct, domain-specific datasets: (1) a Small-Angle Neutron Scattering (SANS) dataset characterizing polyelectrolyte–protein complexes, as reported in Morfin et al. (Boué et al., 2012), and (2) a fundus photograph dataset for visual assessment of Spaceflight-Associated Neuro-ocular Syndrome (SANS), as described in Hossain et al. (Hossain et al., 2023). Each serves as a prototypical data source in its respective field—soft matter structural biophysics and biomedical image super-resolution—under highly controlled or constrained acquisition protocols.

1. SANS Dataset in Soft Matter Biophysics

The SANS dataset of Morfin et al. (Boué et al., 2012) comprises absolute-scale neutron scattering measurements of complexes formed by hyaluronan (HA), a polyelectrolyte, and lysozyme, a basic protein, across variable protein concentrations, chain lengths, and ionic conditions. HA was examined in two molecular weight regimes: HA1 (Mw ≈ 6 000 Da, Lc ≈ 15 nm) and HA2 (Mw ≈ 500 000 Da, Lc ≈ 1 250 nm). Lysozyme concentrations ranged from 3.32 to 40 g L⁻¹, with a constant HA concentration of 10 g L⁻¹.

Scattering experiments were conducted at three neutron sources with the configurations detailed in Table 1. The Q-range spanned from 3.5×10⁻³ to 3.5×10⁻¹ Å⁻¹, and the data were corrected for transmission and detector efficiency using water as a standard, with absolute scaling in units of cm⁻¹. Experimental conditions included pH variation (4.7 and 7.4), buffer ionic strength modulation (up to 170 mM with NaCl), and ambient temperature (20 ± 2 °C).

2. SANS Dataset in Space Medicine Fundus Imaging

The term SANS also denotes the Spaceflight-Associated Neuro-ocular Syndrome fundus image dataset, employed for super-resolution assessment in Hossain et al. (Hossain et al., 2023). This collection comprises 276 color retinal images acquired aboard the International Space Station (ISS) with a non-mydriatic portable fundus camera under operational bandwidth constraints. Image data are divided into an 80/20 train/test split with 5-fold cross-validation on the training set, yielding 220 training and 56 test images.

Images were originally captured at high-resolution (exact pixel dimensions not specified) and then rescaled to 512×512 px via center cropping or padding and bicubic interpolation. Synthetic bicubic degradation was applied to simulate data compression for Earth transmission, downsampling by factors s = 2 and s = 4. Labels comprise a binary designation ("control" or "SANS") in a clinically assessed subset of 56 images; no pixel-wise or demographic metadata are provided. Images are stored as 8-bit/channel RGB PNGs.

3. Experimental Design and Preprocessing

Soft Materials SANS Dataset: Acquisition utilized the PACE (LLB-Saclay), D11, and D22 (ILL-Grenoble) beamlines (see Table 1). All samples were prepared in D₂O-based buffers at fixed HA concentration. pH and salt content were systematically varied to modulate protein charge and ionic screening. Detector absolute intensities were referenced to water, and Q-resolution (ΔQ/Q) was typically 5–10 %. No explicit error propagation for I(Q) was reported beyond point-to-point variations (~3–5 %).

Spaceflight Fundus SANS Dataset: All images underwent uniform pre-processing to 512×512 px for neural network input conformity (Hossain et al., 2023). Bicubic convolution and downsampling generated LR images according to

$I_{LR} = (I_{HQ} * k) \downarrow_s$

where $k$ is the bicubic kernel and $\downarrow_s$ denotes subsampling. Images were saved losslessly as PNG; no additional noise was simulated.

4. Data Structure, Modeling, and Evaluation

Soft Materials SANS Dataset: Scattering curves (I(Q) vs. Q) reveal three regimes corresponding to protein concentration. Low concentration displays q⁻².¹ upturns from concentration inhomogeneity; intermediate regimes show q⁻¹ scaling indicative of rod-like complexes; high concentrations yield q⁻⁴ Porod scattering due to cluster formation, with superposed protein–protein correlation features at q ≈ 0.2 Å⁻¹. The form factor for rod-like particles is modeled as

$$I(Q) = \Delta\rho^2 V^2 \frac{2}{(Q L)^2}\left[\sin(Q L) - Q L \cos(Q L)\right]^2 P_\text{cross}(Q, R)$$

with simplifications in the limit $Q L \ll 1$. Fit-derived parameters (see Table 2) include rod length $L_{rod}$, radius $R_{rod}$, intra-rod protein spacing $a \approx 8 \pm 1$ nm, and inner charge neutrality ($(–)/(+)_\text{inner} \approx 1$). Goodness-of-fit is reported by visual correspondence, not by statistical measure.

Spaceflight Fundus SANS Dataset: Evaluation metrics for image super-resolution include Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), computed as

$$\mathrm{PSNR} = 10\log_{10} \left( \frac{\mathrm{MAX}^2}{\mathrm{MSE}} \right), \quad \mathrm{SSIM}(x, y) = \frac{(2\mu_x\mu_y + C_1)(2\sigma_{xy} + C_2)}{(\mu_x^2 + \mu_y^2 + C_1)(\sigma_x^2 + \sigma_y^2 + C_2)}$$

with results of 45.32 dB PSNR and 0.9793 SSIM for 2× upsampling, outperforming previous architectures on the SANS dataset (Hossain et al., 2023).

5. Structural and Clinical Interpretation

Soft Materials SANS Dataset: Structural analysis supports a "necklace" complex model in which lysozyme globules electrostatically and possibly via H-bonds bind at quasi-regular intervals along the HA backbone. Unlike the isolated HA chain, where rod-like scattering persists only up to the intrinsic persistence length ($L_p \approx 80 nm$), the complexed system exhibits rod-like correlations at much greater lengths, up to the full contour length for long-chain HA2 ($\sim 1 250 nm$), with rod radii corresponding to one or several intertwined HA chains. Salt addition suppresses cluster formation and high-q protein–protein peaks while preserving the q⁻¹ regime, indicating a non-purely electrostatic binding contribution (Boué et al., 2012).

Spaceflight Fundus SANS Dataset: The image-level binary SANS labeling reflects visible pathology, including optic-disc edema and choroidal folds, although ground-truthing is limited. The 276-image set, though of high photographic quality, is constrained by its small size, private status, lack of hardware metadata, and the reliance on synthetic, bicubic degradation for super-resolution assessment. This suggests limited generalizability beyond the ISS operational context.

6. Limitations and Data Access

The soft-matter SANS dataset (Boué et al., 2012) is comprehensively documented with high methodological transparency, yet no raw datasets are provided for external use. In contrast, the spaceflight fundus SANS dataset (Hossain et al., 2023) is retained privately by NASA, with only aggregate results published. Neither dataset includes pixel-level or spatial landmarks (for images) nor provides error estimates or ground-truth labels beyond those reported.

7. Tables: Instrument and Dataset Summary

Table 1. SANS Instrument Configurations (Morfin et al. (Boué et al., 2012))

Instrument	λ (Å)	Collimation (m)	Sample–Detector (m)	Diaphragm
PACE (LLB)	2.5	4.5	4.5	7 mm round
D11 (ILL)	4.6	5.5	1.1, 10.8, 13.5	7 × 10 mm²
D22 (ILL)	≈6.0	4–5 (typical)	17.5	7 × 10 mm²

Table 2. Best‐Fit Rod Parameters (Morfin et al. (Boué et al., 2012))

System	L₍rod₎ (nm)	R₍rod₎ (nm)	Protein spacing a (nm)	(–)/(+)₍inner₎
HA1+lysozyme	15 ± 2	1.1 ± 0.1	8 ± 1	≈1.0
HA2+lysozyme	>60	1.4 ± 0.1	8 ± 1	≈1.0

Table 3. Spaceflight SANS Fundus Dataset Summary (Hossain et al., 2023)

Total Images	Train/Test Split	Image Dimensions	Labeling
276	220/56	512×512 px, RGB	Binary (control/SANS)

• Morfin, I., et al. "Rodlike Complexes of a Polyelectrolyte (Hyaluronan) and a Protein (Lysozyme) observed by SANS" (Boué et al., 2012).
• Hossain, M. S., et al. "Revolutionizing Space Health (Swin-FSR): Advancing Super-Resolution of Fundus Images for SANS Visual Assessment Technology" (Hossain et al., 2023).