Fig. 6: LF-denoising preserves temporal causality for neural analysis with high fidelity. | Nature Communications

Fig. 6: LF-denoising preserves temporal causality for neural analysis with high fidelity.

From: Leveraging spatial-angular redundancy for self-supervised denoising of 3D fluorescence imaging without temporal dependency

Fig. 6

a Left, system concept of 2pSAM with 13 scanning angles. Right, training strategy of LF-denoising on 2pSAM data. b Experiment on the Drosophila neural activity in response to odor stimuli. Created in BioRender. Fan, J. (2025) https://BioRender.com/cp0dmbdc Visualizations of a Drosophila brain expressing a genetically calcium indicator, comparing low-SNR raw image and LF-denoising enhanced image. d Functional traces extracted from raw images and LF-denoising enhanced images. e Average local peaks extracted from functional traces using different methods. f Boxplot showing full widths at 75% maximum of local peaks. The boxplot formats: center line, median; box limits, lower and upper quartiles; whiskers, 1.5-fold interquartile range. n = 20 represents the number of local peaks. Asterisks represent significance levels tested with two-sided independent t test. P values are 5.36 × 10−9, 4.22 × 10–2 and 6.62 × 10–2 from left to right. g Enlarged views showing the functional traces near the moment of odor stimuli. Arrows indicate errors induced by previous methods. h Average low-dimensional latent features of neural activity in olfactory brain regions after each category of odor release. Each trace represents the average odor response of an odor identity. Trace colors indicate the time relative to odor releases. Top, “Dist.” denotes the maximum distance between the three start points and the average start point of all trials. Smaller distances indicate greater temporal accuracy. i Causality analysis on data before (upper row) and after LF-denoising (lower row). Left, matrix of Granger causality between each pair of six brain regions. Causality with P < 0.05 (Granger F test) is shown. Right, summary of causality between regions. Red arrows indicate inferred causality direction. j Three neurons in the olfactory brain region, shown in LF-denoising enhanced image. k Matrices of Granger causality between each pair of three neurons from raw data, DeepCAD-RT, SRDTrans and LF-denoising enhanced data. Causality with P < 0.01 (Granger F test) is shown. Cross denotes causality significant in raw data but lost after denoising. Scale bars, 40 μm (c, i, j).

Back to article page