Unsupervised Detection of Rare Events in Liquid Biopsy Assays

This dataset contains immunofluorescence microscopy images of blood samples used to develop a deep generative model for unsupervised rare cell detection. The samples consist of peripheral blood from normal donors spiked with cancer-related cells.Images are provided as .jpg files at a resolution of 1362×1004 pixels. Each microscopy image is divided into 2,304 tiles, with filenames following the format Tile{tile_number_id}.jpg. The tile IDs are organized by fluorescence channel as follows:Tiles 1–2304: DAPI channelTiles 2305–4608: PanCK channelTiles 4609–6912: CD45/31 channelTiles 9217–11520: Vimentin channelCorresponding tiles across channels share the same spatial location in the original image. For example, tiles 1, 2305, 4609, and 9217 represent the DAPI, PanCK, CD45/31, and Vimentin channels, respectively, for the same field of view.