Training dataset for analysis of slit diaphragm in Drosophila nephrocytes

Training dataset for U-Net segmentation of slit diaphragms in Drosophila nephrocytes This dataset contains annotated confocal fluorescence microscopy images and corresponding segmentation masks used to train a U-Net model for automated detection of slit membranes and membrane protein clusters in Drosophila nephrocytes stained with anti-SNS antibody. Dataset contents 21 manually annotated confocal image tiles (original source data) 63 annotation masks — three masks per image: slit membrane (_mem), protein clusters (_clusters), and uncertainty/ignore regions (_ignore; pixel value 255 excluded from training loss) 8 photometric augmentation variants per image: Gaussian blur, Poisson noise, radial vignetting, linear intensity gradient, intensity multiplication, additive brightness shift (+30), power/gamma transform, and spatially-varying random intensity field Elastic deformation variants for one tile (PN1-tile20; α=50, σ=5), generating 3 independent deformed source images Pre-assembled train/test split: 189 training images and 16 test images as used in the published model Folder structure original_images/ — 21 manually annotated source tiles (TIFF) original_masks/ — 63 corresponding annotation masks (TIFF) augmented/ — all augmented image variants organised by augmentation type train_test_split/ — ready-to-use training and test subsets as used for model training training_metadata/ — training curves, training report, and example predictions Dataset_v8_documentation.md — full dataset documentation including augmentation parameters and train/test split composition Intended use Use original_images/ + original_masks/ to retrain with a different augmentation strategy. Use train_test_split/ to reproduce the exact results reported in the paper. The corresponding pre-trained model weights (U-Net v8, best epoch 71, test loss 0.274) are available in a separate Zenodo record. Associated resources Model weights: [Zenodo DOI — 10.5281/zenodo.19828538] Analysis pipeline and inference code: https://github.com/MedD-Research/nephrocyte-slit-membrane