Related Experiment Video
Updated: Jul 14, 2026
14:13
Atomic Force Microscopy of Red-Light Photoreceptors Using PeakForce Quantitative Nanomechanical Property Mapping
Published on: October 24, 2014
AFMap-UNet enables accurate nuclear segmentation of atomic force microscopy images with minimal training data
Arthur Henrique Rocha1, Cleyton Alexandre Biffe2, Ed Carlos Santos E Silva1
1Laboratório de Genética e Cardiologia Molecular, Instituto do Coração (InCor), Faculdade de Medicina da Universidade de São Paulo, São Paulo, Brazil.
Scientific Reports
|July 12, 2026
Summary
AFM-based nuclear mechanics quantification is enhanced by AFMap-UNet, a novel deep learning model. This tool integrates Atomic Force Microscopy (AFM) and optical data for precise cell nucleus segmentation, even with limited data.
Area of Science:
- Biophysics
- Cell Biology
- Artificial Intelligence
Background:
- Nuclear mechanical properties are crucial for cellular functions like transcription and signaling.
- Atomic Force Microscopy (AFM) offers high-resolution imaging of cell topography and mechanics.
Purpose of the Study:
- To develop a high-performance deep learning model for precise nuclear segmentation using AFM and optical microscopy data.
- To enable spatially resolved quantification of nuclear mechanics for advanced cell analysis.
Main Methods:
- Integration of AFM topography maps with enhanced optical microscopy channels.
- Application of a two-stage, region-guided U-Net deep learning architecture (AFMap-UNet).
- Validation of model performance on small, data-constrained training sets.
Main Results:
- AFMap-UNet achieved high precision (99% average precision) and segmentation accuracy (96% median Dice coefficient).
- The model demonstrated robust performance even with minimal training data (15 images).
- This represents the first deep learning model for high-performance, spatially resolved nuclear mechanics quantification.
Conclusions:
- AFMap-UNet significantly advances AFM-based cell analysis by enabling accurate nuclear segmentation and mechanical property quantification.
- The model's scalability makes it suitable for data-limited AFM studies.
- This technology opens new avenues for research in cell mechanics, disease modeling, and drug discovery.
