Reconstruction of natural images from responses of primate retinal ganglion cells

  1. Nora Brackbill  Is a corresponding author
  2. Colleen Rhoades
  3. Alexandra Kling
  4. Nishal P Shah
  5. Alexander Sher
  6. Alan M Litke
  7. EJ Chichilnisky
  1. Department of Physics, Stanford University, United States
  2. Department of Bioengineering, Stanford University, United States
  3. Department of Neurosurgery, Stanford School of Medicine, United States
  4. Department of Ophthalmology, Stanford University, United States
  5. Hansen Experimental Physics Laboratory, Stanford University, United States
  6. Department of Electrical Engineering, Stanford University, United States
  7. Santa Cruz Institute for Particle Physics, University of California, Santa Cruz, United States
13 figures and 1 additional file

Figures

Linear reconstruction from ON and OFF parasol cell responses.

(A) Visual stimulus: static images from the ImageNet database were displayed for 100 ms, with 400 ms of gray between. The thin black rectangle indicates the central image region shown in C and E. (B)…

Figure 1—source data 1

Linear reconstruction from ON and OFF parasol cell responses.

This zip file contains the code and data for Figure 1D and F, which show the distribution of reconstruction scores across recordings, as well as the relationship between reconstruction performance and receptive field (RF) size.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig1-data1-v2.zip
Visual representation across retinas.

(A) Distribution of correlation between reconstructed images from different recordings, across 150 images and 66 pairs of recordings. (B) Example image. (C) Across 12 recordings, reconstructed …

Figure 2—source data 1

Visual representation across retinas.

This zip file contains the code and data for Figure 2A, which shows the similarity of reconstructed images across separate recordings.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig2-data1-v2.zip
Effect of the population on the visual message.

(A) The reconstruction filter of a single cell as more neighboring cells are included in the reconstruction. Left: receptive fields (RFs) of cells in reconstruction, with the primary cell indicated …

Figure 3—source data 1

Effect of the population on the visual message.

This zip file contains the code and data for Figure 3D and E, which show how the visual message changes depending on other RGCs. This includes the widths and profiles of the reconstruction filters.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig3-data1-v2.zip
Effect of visual message on reconstruction.

(A) Receptive field (RF, left) and reconstruction filter (right) contours for two sample recordings. (B) Reconstruction of an image (top) using the full, fitted filters (middle) and using scaled RFs …

Figure 4—source data 1

Effect of the visual message on reconstruction.

This zip file contains the code and data for Figure 4C, D and E, which compare the full and receptive field (RF) reconstructions. This includes the coverage values for the RFs, the filters, and the expanded RFs, as well as the full and RF reconstruction scores, and the power spectra of the full and RF reconstructions.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig4-data1-v2.zip
Effect of other cell types on the visual message.

(A) Average reconstruction filters for ON parasol (top row), OFF parasol (second row), ON midget (third row), and OFF midget (bottom row) cells for one recording. Left to right: including all cell …

Figure 5—source data 1

Effect of other cell types on the visual message.

This zip file contains the code and data for Figure 5B, which compares the magnitude and width of the filters when other cell types are included in the reconstruction.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig5-data1-v2.zip
Contributions of ON and OFF parasol cells.

(A,B) Example images, responses, and reconstructions from ON and OFF parasol cells. Top left: original image. Top right: Parasol cell mosaics shaded by their response value (ON - blue, middle, OFF - …

Figure 6—source data 1

Contributions of ON and OFF parasol cells.

This zip file contains the code and data for Figure 6C and D, which compare the reconstructions from ON and OFF parasol cell responses. This data includes the performance scores for reconstructions from ON and OFF parasol cell responses, as well as the binned true and estimated pixel values.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig6-data1-v2.zip
Contributions of the parasol and midget cell classes.

(A) Example image and reconstructions for parasol and midget cells. Top left: original image. Top right: reconstruction with parasol and midget cells (gray). Bottom left: reconstruction with only …

Figure 7—source data 1

Contributions of the parasol and midget cell classes.

This zip file contains the code and data for Figure 7C, D and E, which compare the reconstructions from parasol and midget cell responses. This data includes the performance scores for reconstructions from parasol and midget cell responses, as well as the power spectra of the resulting images, and the time required to reach 95% reconstruction performance.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig7-data1-v2.zip
Effect of noise correlations.

The change in reconstruction performance (Δρ) when using shuffled data for three scenarios: one 150 ms window, fifteen 10 ms windows, and one 10 ms window. Black bars show median +/- interquartile …

Figure 8—source data 1

Effect of noise correlations.

This zip file contains the code and data for Figure 8, which shows the effects of noise correlations on reconstruction performance.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig8-data1-v2.zip
Nonlinear reconstruction.

(A) Average pixel value in receptive field center vs. original response (blue) and transformed response (orange). (B) Distribution (across n = 2225 cells from 15 recordings) of the change in RMSE of …

Figure 9—source data 1

Nonlinear reconstruction.

This zip file contains the code and data for Figures 9B, C and D, which show the effects of using a static nonlinear transformation, and of including nonlinear interaction terms.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig9-data1-v2.zip
Comparison to simulated spikes.

(A) Average reconstruction filters calculated from spikes simulated using linear-nonlinear models (left) or recorded (right). (B) Images reconstructed from simulated (left) or recorded (middle) …

Figure 10—source data 1

Comparison to simulated spikes.

This zip file contains the code and data for Figure 10C, which compares reconstruction using recorded and simulated RGC responses.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig10-data1-v2.zip
Spatiotemporal reconstruction.

(A) Examples of the spatial components extracted from the spatiotemporal reconstruction filter (top) and the static spatial reconstruction filters (bottom) for an ON (left) and OFF (right) parasol …

Figure 11—source data 1

Spatiotemporal reconstruction.

This zip file contains code and data for Figure 11B, which compares static and spatiotemporal reconstruction filters.

https://cdn.elifesciences.org/articles/58516/elife-58516-fig11-data1-v2.zip
Verification of data sufficiency.

(A) Performance of reconstructions from parasol cell responses as a function of the amount of training data, for 19 recordings (colors). (B) Fraction of performance of reconstructions from parasol …

Selection of analysis region.

Reconstruction performance on a sample image (top) is measured by comparing the regions inside the contours shown on the reconstructions in the second row. These contours were obtained using the …

Additional files

Download links