Constantin Pape

@cppape.bsky.social

1664 Followers 398 Following

Group leader @ Uni Göttingen for Computational Cell Analytics and @ MPINAT for Machine Intelligence in the Life Sciences. AI for biology and medicine with a focus on microscopy. https://user.informatik.uni-goettingen.de/~pape41/

Results for object segmentation based on foundation model (FM) features with a random forest, compared with supervised learning baselines and classical features. Experiments over five different datasets covering cell classification in spatial proteomics (CRC, HBM), high-content macroscopic imaging for animal phenotyping (Planri), nucleus classification in histopathology (PanNuke), and cell line classification in phase-contrast microscopy (LIVECell). The extra plot repots runtimes for model training and inference.
Evaluation of different methods for pixel classification (top) and object classification (bottom) on the LIVECell dataset. Dark green bars show the F1-Score, which measures the classification / segmentation quality (higher is better), light green bars show the runtimes. Five different settings are compared for each task: Ilastik features + random forest (RF), microSAM embeddings + RF, SAM2 embeddings + RF, and uSAM, SAM2 + attentive probing (DeAP and ObAP). microSAM feature perform best for RF based methods, attentive probing outperforms RF based approaches, but at a much higher runtime.
A cochlea imaged in light-sheet microscopy (right) with staining for spiral ganglion neurons (red) and inner hair cells (blue). You will develop AI-based methods to analyze these structures, for example via segmentation of the individual cells (right) that will support gene therapy development for hearing loss and a better overall understanding of the anatomy of hearing.
Rendering of a full cochlea (left) with three stains (PV, VGlut3, CTBP2) shown in three different colors (red, blue, cyan). The whole cochlea is a spiral shaped structure, with spiral ganglion neurons (SGNs) in the inner helix, labeled by PV and inner hair cells (IHCs) in the outer helix, labeled by Vglur3. The figure also shows zoom ins. The right hand side shows segmentation results for SGNs, IHCs (represented by colored masks) and synapse detections (represented by colored dots).
SynapseNet is a deep learning based software tool that automates the segmentation of vesicles, mitochondria, synaptic compartments, and the active zone. This is visualized in three panels. On the left, a section of an electron tomogram with a synaptic compartment densely filled with vesicles, which appear as round structures with dark boundary and light body in the image, is shwon. The top right panel shows the segmentation result of vesicles, visualized by masks with an individual color per vesicle and outlines for the segmented compartment (red) and active zone (blue). The bottom right panels shows a 3D rendering of the segmentation with vesicles shown as yellow spheres.
Synaptic vesicles segmented with SynapseNet.
Interactive cell segmentation with micro_sam.
Architecture of the segment anything model with additional segmentation decoder.
Improvements in LM (top) and EM (bottom) of our micro-sam model (finetuned) compared to the default SAM model.
Nucleus instance segmentation result averaged over several histopathology datasets for PathoSAM and several other methods.
Nucleus instance segmentation on a whole slide image (WSI) from https://openslide.org/ . The gif starts with a zoomed out version of the WSI and then zooms in. Nuclei are segmented, represented by colored overlays. In the end the nucleus segmentation is toggled off to see the nuclei in the image.
Electron micrograph of a mossy fiber synapse with many synaptic vesicles.