@cppape

Group leader @ Uni Goettingen applying Image Analysis and Deep Learning to large microscopy data in biology. Also @cppape.bsky.social

 Goettingen
user.informatik.uni-goettingen.de/~pape41/
Joined December 2015
  • Tweets1K
  • Following558
  • Followers1.3K
We compare foundation models (SAM and DINO variants) with classical features from ilastik and supervised learning. The FM features outperform classical approaches. Attentive probing is better than RF, but too slow for interactive use. See example results in the fig.
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.

ALT 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.

1
2
116
We use embeddings from foundation models as features, either for a RF or different attentive probing variants. The features are averaged over masks for object classification. See the image for a method overview.
1
2
97
Segmenting cells in microscopy is much easier these days thanks to foundation models. Can we use these models for other tasks, e.g. cell classification? See our latest work! We find big improvements for object and pixel classification compared to classical approaches.
valuation 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.

ALT valuation 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.

1
3
16
541
How can we use foundation models such as (micro)SAM to improve electron microscopy segmentation? Check out our preprint! We found substantial improvements for nucleus, mito, and neurite-segmentation using initialization and semi-supervised learning with foundation models.
1
9
47
3,544
Looking for a PhD position at the intersection of AI, imaging, and gene therapy? Apply for this position in my lab: tinyurl.com/2a2v6tvx Part of sfb1690.uni-goettingen.de/ that studies hearing, vision, and more. Plus, you can create pretty pictures like the one below :).
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.

ALT 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.

2
8
2
40
2,610
We comprehensively evaluate our method and use it to analyze intact mouse and gerbil cochleae, including SGN sub-types, and to validate (opto-)genetic therapies preclinically (see screenshot). CochleaNet is also applicable to lower-resolution data from commercial systems.
1
1
104
Cochleae were cleared, stained and imaged with a high-resolution light-sheet microscope (doi.org/10.1038/s41587-025-0…). After preprocessing,we segment and analyze the data with CochleaNet, using the three dedicated networks trained on newly annotated data.
1
1
31
Preprint alert! CochleaNet, our framework for analyzing cochlear light-sheet data. It consists of three networks to segment spiral ganglion neurons, inner hair cells, and detect synapses. Rendering of a full cochlea below, find the preprint at doi.org/10.1101/2025.11.16.6…
1
4
386
3. The model selection now supports two additional models (Histopathology, Medical Imaging) and uses human readable names, see the screenshot below. For more on these models check out: computational-cell-analytics…
1
1
3
270
2. We added preliminary support for automatic tracking via trackastra, developed by @martweig. See the video for an example result and check out computational-cell-analytics… for details.
1
3
138
Load more