Preprocessed MALDI-MS imaging dataset across tumor stages in a murine breast cancer model

Annotations are available for 9 of 38 specimen and classify the MALDI images by their correspondence to the HER2-positive target regions as match, approximate match or non-match. All images, including the histological references, were preprocessed for deep learning-based image analysis. For details on data acquisition please see the method section of the related publication or contact the corresponding author.

Performed preprocessing Preprocessing included grayscale conversion of the HER2-stained reference images and aspect-ratio-preserving resizing to 224 x 224 pixels using zero padding. MALDI-MSI intensities were capped at the 99.9%-quantile to exclude extreme outliers and scaled to 0-1. Folder structure and software The data is provided as '.pt' files containing PyTorch tensors.

The files were generated in Python and can be loaded using the PyTorch package via torch.load('filename.pt', weights_only=False) Dataset/ ├── Example_DataLoadingAndHandling.ipynb ├── Labels/ │ ├── Labels_M1L.pt │ └── ... └── Processed_Data/ ├── 00mz_names.pt ├── Edi3-Metastase-M1-Lu-Left.pt └── ... The folder "Processed_Data" contains one .pt file per specimen. In each file, the first entry is the preprocessed HER2-stained reference image, followed by the preprocessed MALDI-MSI images ordered by increasing m/z value.

The file "00mz_names.pt" contains the corresponding names of the m/z-values in the same order. The folder "Labels" contains .pt label files for the annotated validation and test specimens. The Jupyter Notebook "Example_DataLoadingAndHandling.ipynb" provides examples and functions to load and visualize the data.

For a detailed explanation see README.txt and the provided code example.