MICRON: Experimental Data and Resources

This repository provides dataset resources and train/test split information used in the MICRON experiments across three imaging mass cytometry (IMC) datasets: brain tumors, breast cancer, and diabetes. Brain Tumor Dataset (Karimi et al.) The brain tumor IMC dataset profiles the tumor microenvironment from 139 patients with primary glioblastoma (GBM) and 46 patients with metastatic brain tumors (BrM). For each sample, we considered 18 common channels (17 protein markers) shared across samples. The samples were obtained from different tissue compartments, including the tumor core, invasive margin, and adjacent non-malignant tissue. Because certain patients contributed multiple spatial regions, the dataset includes 270 GBM samples and 123 BrM samples. The classification task distinguishes primary glioblastoma from metastatic brain tumors. Breast Cancer Dataset (Jackson et al.) The breast cancer IMC dataset profiles primary breast tumor tissues using 52 imaging channels, including 40 protein markers. The dataset comprises 376 tumor samples, including 274 samples from patients who were alive at last follow-up and 102 samples from patients who had died. The classification task distinguishes samples according to binary survival status (alive or deceased). Diabetes Dataset (Damond et al.) The diabetes IMC dataset profiles pancreatic tissue samples from 67 donors, including 33 non-diabetic donors and 34 donors with type 1 diabetes (T1D). The classification task distinguishes non-diabetic from T1D samples. Data Availability and Reproducibility The train/test split files used in the MICRON experiments are provided here to facilitate reproducibility. The original breast cancer and diabetes imaging data and associated annotations are available through the dataset resources provided by their respective publications. For the breast cancer and diabetes experiments, users can organize the downloaded data following the same directory structure and file organization demonstrated for the brain tumor dataset in the MICRON GitHub repository. The same MICRON analysis pipeline can then be applied to the corresponding datasets.