Breast Cancer Nuclei images for DL Training + ZeroCostDL4Mic StarDist Model

Training dataset: Paired microscopy images (fluorescence) and corresponding masks Microscopy data type: Fluorescence microscopy and masks obtained via manual correction of automatic segmentation with pre-trained StarDist model (see https://github.com/qupath/models/tree/main/stardist) Cells were imaged using a 20x objective with a 1x camera adapter was used in conjunction with a pco.edge 4.2 4MP camera on Pannoramic SCAN 150 scanner. Cell type: FFPE tissue sections were sliced from all cancer-containing paraffin blocks File format: .tif (8-bit for fluorescence and 16-bit for the masks) StarDist Model: The StarDist model was generated using the ZeroCostDL4Mic platform (Chamier et al., 2021). This custom StarDist model was trained for 100 epochs using 80 manually annotated paired images (image dimensions: (257, 257)) with a batch size of 2, an augmentation factor of 10 and a mae loss function. The StarDist “Versatile fluorescent nuclei” model was used as a training starting point. Key python packages used include TensorFlow (v 2.2.0), Keras (v 1.1.2), CSBdeep (v 0.7.2), NumPy (v 1.21.6), Cuda (v 11..1.105). The training was accelerated using a Tesla P100GPU. The model weights can be used in the ZeroCostDL4Mic StarDist 2D notebook or in the StarDist Fiji plugin. a QuPath-compatible model is also provided.