Clustering cell nuclei on microgrooves for disease diagnosis using deep learning.

Area of Science:

• Cell Biology
• Biophysics
• Machine Learning

Background:

• Abnormal nuclear mechanical properties are linked to diseases like laminopathies and cancer.
• Microgroove substrates mimic basement membrane topography, inducing 3D nuclear deformations in cells.
• Nuclear deformations differ between healthy cells and those from laminopathy patients.

Purpose of the Study:

• To evaluate Variational Autoencoder (VAE) and Gaussian Mixture Model (GMM) for clustering nuclear morphologies.
• To assess the impact of image processing on clustering performance.
• To correlate nuclear deformation clusters with wildtype and laminopathy-associated mutations.

Main Methods:

• Culturing wildtype and laminopathy-associated mutation myoblasts on microgroove substrates.
• Applying a Variational Autoencoder (VAE) combined with a Gaussian Mixture Model (GMM) to nuclear image patches.
• Analyzing clustering performance based on image processing parameters.

Main Results:

• VAE and GMM successfully clustered nuclei based on morphology and deformation.
• Clusters corresponded to either wildtype myoblasts or myoblasts with LMNA mutations.
• The approach demonstrated accurate classification of nuclear deformations.

Conclusions:

• Deep learning combined with microgroove substrates enables automatic classification of nuclear deformations.
• This technique offers a promising approach for rapid diagnosis of diseases involving nuclear deformation abnormalities.
• The study highlights the potential of AI in identifying cellular phenotypes related to genetic disorders.