Myoblast Mutation Classification via Microgroove-Induced Nuclear Deformations

Xingjian Zhang, Claire Leclech, Bettina Roellinger, Catherine Coirault, Elsa Angelini, Abdul I. Barakat

Keywords: microgrooves, cell mutation, nuclear deformation, image classification

Abstract: Microgroove substrates induce 3D nuclear deformations in various adherent cell types. In this study, we explore the capacity of a CNN classifier to identify myoblast mutations through subtle differences in nuclear deformations on 2D fluorescence microscopy images. A large set of experimental images from immunostained nuclei screened on microgroove platforms is exploited. Leveraging ResNet-50 in a weakly-supervised setting, we present preliminary results to accurately classify healthy myoblasts from laminopathy-associated mutations. We achieved F1 scores of 0.99 and 0.94 at whole-image and patch levels evaluations. These results demonstrate the potential for microgroove screening as a functional diagnostic device of diseases characterized by aberrant nuclear deformations.

Submission Number: 68