MagChains bright-field microscopy images and deep learning code for quantitative detection of Salmonella Typhimurium

Description

This dataset contains 800 original bright-field microscopy images and the associated deep learning code developed for MagChains-based quantitative detection of Salmonella Typhimurium. The images capture magnetic bead chain morphologies used for image-based estimation of bacterial concentration. The image collection comprises 80 independent assays across eight concentration groups, including a bacteria-free control, with 10 fields of view acquired per assay. Images from the same assay constitute a multi-field-of-view sample and should be kept together when partitioning the data into training, validation, and test sets to prevent information leakage. The accompanying code implements a multi-field-of-view attention-based regression framework using a shared ResNet18 image encoder and multiple-instance learning to aggregate morphological features across fields of view and generate an assay-level concentration prediction. These resources support examination of magnetic bead chain morphology and reuse of the image-based analysis approach for quantitative bacterial detection.

Institutions

• China Agricultural University

  Beijing, Beijing

Categories

Licence