Dataset of "Volumetric denoising enables efficient acquisition of volume electron microscopy"

This dataset contains raw volume electron microscopy (VEM) data supporting the findings in the manuscript "Volumetric denoising enables efficient acquisition of volume electron microscopy". The data includes both Serial Block-Face SEM (SBF-SEM) and Plasma-FIB SEM (PFIB-SEM) modalities, used to evaluate raw data, supervised, and self-supervised denoising strategies for accelerating VEM acquisition. 1. PFIB-SEM Data for Resolution Validation (Related to Figure 1) This subset contains authentic PFIB-SEM datasets acquired from mouse brain tissue at a 5x5x5nm isotropic voxel size. These volumes were used to validate the impact of pixel dwell time on effective spatial resolution via Sectional Fourier Shell Correlation (SFSC). PFIB_5x5x5nm_2us.tif to PFIB_5x5x5nm_0d25us.tif: A series of authentic datasets scanned at varying dwell times: 2 μs, 1.5 μs, 1 μs, 0.75 μs, 0.5 μs, and 0.25 μs. 2. Supervised Denoising Experiments (Related to Figure 3) Datasets acquired using SBF-SEM (Zeiss Gemini 300) to compare 2D and volumetric (3D) denoising performance against a shared ground truth. XY:15nm pixel size. SBEM2-Z50-fast.tif: The "Snapshot stack" (S-stack); acquired with 0.5 μs dwell time and 50 nm section thickness. SBEM2-Z50-slow.tif: The "Reference stack" (R-stack); acquired with 2 μs dwell time and 50 nm section thickness, serving as the ground truth. 3. Self-Supervised Denoising on Ultra-Thin Sections (Related to Figure 4) SBF-SEM datasets used to demonstrate how volumetric denoising enables reduced cutting thickness (improved axial resolution) by mitigating radiation damage through faster scanning. XY:15nm pixel size. SBEM3-Z20-300.tif: 20 nm cutting thickness (~300 slices). SBEM3-Z25-300.tif: 25 nm cutting thickness (~300 slices). SBEM3-Z50-300.tif: 50 nm cutting thickness (~300 slices). (All acquired with 0.5 μs dwell time in snapshot mode). 4. Downstream Segmentation Tasks (Related to Figure 5) Additional PFIB-SEM volumes and expert annotations used to evaluate the biological utility of denoised data in automated organelle and structure segmentation. PFIB_seg_0d5us_Raw_data_5x5x5nm.tif: Raw data scanned at 0.5 μs dwell time for segmentation testing. PFIB_seg_2us_Reference_5x5x5nm.tif: High-SNR reference data scanned at 2 μs dwell time. mitos_human_label.tif: Expert human annotations for mitochondria performed on the 2 μs reference stack. synapses_human_label.tif: Expert human annotations for synapses performed on the 2 μs reference stack.