#18 · ML / AI methods

Deep-learning tools automate quantification of cell and organelle morphology in imaging

The common thread is automated, quantitative extraction of structure from microscopy images, delivered as reusable software and reference data rather than single-use analyses. MurineCyto-Det provides 333 annotated 1024x1024 murine BALF tiles with 14,551 cell instances, segmentation masks, boxes, and baseline benchmarks for leukocyte segmentation and detection. VesiclePy applies a deep-learning pipeline to large volume electron microscopy, covering vesicle segmentation, classification, proofreading, and 3D spatial analysis. BudFinder uses a masked autoencoder vision transformer, pretrained on about 250K unlabeled yeast image crops, to detect budding events and quantify replicative lifespan. The distinctive element is BudFinder's use of unlabeled data for pretraining, while the other two items pair analysis with curated annotation or 3D spatial reconstruction. Together they target different modalities but share an emphasis on packaged computational resources for quantitative image analysis.