#4 · ML / AI methods

Deep learning automates microscopy and histology image analysis with less labelled data

The dominant thread is deep learning applied to biomedical images, with attention to generalization and reduced reliance on manual annotation. Several items target robustness under distribution shift: CS-ARM-BN uses negative control samples to stabilize batch-normalization statistics across sources, and PanGuide3D conditions pancreatic tumor CT segmentation on a probabilistic pancreas map to hold up across cohorts. Annotation burden is addressed through weak supervision, with cell-level spatial transcriptomics generating image-label pairs for histopathology nuclei, and CHRep predicting spatial gene expression from H&E histology. FiloAnalyzer contributes an open-source GUI for batch filopodia segmentation. Other items are less uniform: the rat OCT benchmark found a CNN Xception baseline outperforming the transferred RETFound model, and the RE-CONFIRM work operates on dynamic functional connectivity rather than images. The thread is methodological rather than biological.