Compression keeps stem cell image models accurate on standard CPUs
Retention-Constrained Post-Training Quantization of Cellpose-SAM for Stem Cell Microscopy
Emerging Technologies
Summary
Stem cell research uses computer models to identify cells in microscope images, but these models can be too large for common lab computers. The authors tested ways to shrink a popular model called Cellpose-SAM without losing accuracy. They found that some compression methods keep the model nearly as good as the full version while using much less storage. Extremely simple compression made the model unreliable, showing that careful testing is needed for each image type to keep results trustworthy.
What this means in practice
- •For laboratory it teams: Shrink cell image segmentation models to run efficiently on standard lab CPUs without losing accuracy across different microscope types.
- •For biomedical device makers: Implement reliable compressed segmentation models on edge hardware for portable stem cell imaging systems.
Authors
Sebastián A. Cruz Romero
Abstract
Induced pluripotent stem cell (iPSC) culture increasingly relies on segmentation foundation models, yet deployment on laboratory CPUs and edge hardware requires compression schemes that are both efficient and auditable. We present a deployment-oriented evaluation of compressed Cellpose-SAM using a pre-specified retention criterion: the 95% cluster-bootstrap interval of mean change from FP32 must remain above a fixed -0.02 margin for every imaging modality. On a stratified 176-field panel spanning BBBC038 nuclei, BBBC039 U2OS fluorescence, and NIST iPSC images across density regimes, weight-only W8A16 preserves instance F1 across all modalities. A sensitivity-guided mixed W4/W8 scheme, using four INT8 exceptions, achieves a 6.76x reduction in weight storage with no observed catastrophic failures (0/176 fields), matching W8A16 at this sample size. In contrast, ternary weight-only quantization achieves 12.08x compression but fails catastrophically on 169/176 fields. These results demonstrate that compression should be evaluated by modality-stratified downstream retention rather than single-number accuracy, and establish a reproducible protocol for auditing compressed foundation models in regulated stem-cell imaging.