Machine Learning-Based Estimation of Experimental Artifacts and Image Quality in Fluorescence Microscopy - Supporting Data

Supporting data for the reproduction of the results reported in Corbetta, E., Bocklitz, T., Machine learning based estimation of experimental artifacts and image quality in fluorescence microscopy (2024) [1].

MM-IQA_Images_png and MM-IQA_Images_tif

Folder containing all the supporting datasets of the publication.

• images_manual_inspection: semisynthetic dataset used for the manual inspection of the quality metrics.
• images_lda_training: semisynthetic dataset used to train the Linear Discriminant Analysis (LDA) model.
• images_lda_prediction: datasets predicted by the LDA model.
  • images_experimental: every subfolder is a dataset composed of measurements of a different sample. Images are from publicly available datasets from [2] and [3].
  • images_known_semisynthetic: knwon semisynthetic dataset used for prediction, assessment and interpretation of the trained model.

Tif files are the original data used for the study.

MM-IQA_Source_data

Folder containing all the supporting metadata of the publication.

• manual_inspection: quality metrics computed for the semisynthetic dataset used for the manual inspection.
  • manual_inspection_bg: indices for the selection of the background region in each sample.
  • manual_inspection_free_parameters: parameters used for the generation of the simulated artifacts.
  • manual_inspection_metrics: quality metrics computed for the dataset, used for the manual inspection.
  • manual_inspection_samples: free parameters associated to each image of the dataset for manual inspection.
• LDA_training: semisynthetic dataset used to train the Linear Discriminant Analysis (LDA) model.
  • lda_metrics_synthetic+semisynthetic_uniform_max: quality metrics computed for the training dataset, with maximum normalization of the images. (Not used in the manuscript)
  • lda_metrics_synthetic+semisynthetic_uniform_rescale01: quality metrics computed for the training dataset, with image values rescaled between 0 and 1.
  • parameters_all_degradations: parameters used for the generation of the simulated artifacts.
• LDA_prediction: datasets predicted by the LDA model.
  • experimental: quality metrics computed for measurements of different samples. Images are from publicly available datasets from [2] and [3].
  • known_semisynthetic: metadata for the knwon semi-synthetic dataset used for prediction, assessment and interpretation of the trained model:
    • known_semisynthetic_free_parameters: parameters used for the generation of the simulated artifacts.
    • known_semisynthetic_metrics_rescale01: quality metrics computed for the training dataset, with image values rescaled between 0 and 1.
    • known_semisynthetic_maxnorm_lda_results: lda prediction results, when metrics are maximum normalized to the training dataset.
    • known_semisynthetic_znorm_lda_results: lda prediction results, when metrics are z-score normalized to the training dataset.

Source data can be used to reproduce the results of the manuscript, using the codes shared in the public GitLab repository multi-marker-IQA.

How to use the source data

The following table describes which data can be used in the scripts provided in the public GitLab repository multi-marker-IQA.

MM-IQA_Scripts

• multi-marker-iqa-main: original GitLab repository for MM-IQA, version available at the date of manuscript publication.
• Notebooks_peer_review: additional notebooks generated during the peer-review process with the computation of metrics for natural images and correlation measures.