Deep Learning-driven Automatic Nuclei Segmentation of Label-free Live Cell Chromatin-sensitive Partial Wave...

Area of Science:

• Cell biology
• Biophysics
• Medical imaging

Background:

• Chromatin-sensitive Partial Wave Spectroscopic (csPWS) microscopy provides nanoscale mass density information for studying cell nuclei and their role in carcinogenesis.
• Accurate nuclei segmentation is crucial for csPWS analysis but manual methods are unreliable and time-consuming.

Purpose of the Study:

• To develop an automated, accurate deep learning method for segmenting cell nuclei in label-free live cell csPWS microscopy images.
• To improve the reliability and efficiency of csPWS data analysis for chromatin and carcinogenesis research.

Main Methods:

• Developed csPWS-seg, a U-Net model with an attention mechanism for nuclei segmentation.
• Utilized three distinct csPWS feature images based on structural and biological differences.
• Evaluated model performance using Intersection over Union (IoU) and Dice Similarity Coefficient (DSC) metrics.
• Compared performance against baseline U-Net and SE-U-Net models, and analyzed four loss functions (binary cross-entropy, focal, dice, Jaccard).

Main Results:

• csPWS-seg achieved a median IoU of 0.80 and DSC of 0.88 on HCT116 cell images.
• The model significantly outperformed baseline U-Net and SE-U-Net.
• csPWS-seg with focal loss demonstrated the best segmentation performance among the tested loss functions.

Conclusions:

• csPWS-seg provides accurate and automated nuclei segmentation for csPWS microscopy.
• The method streamlines data analysis, enhances reliability for chromatin studies, and supports diagnostics and understanding of carcinogenesis.