Abstract
The fusion index (FI) is widely used to assess myoblast differentiation into myotubes, but manual quantification is tedious and biased. Existing automated methods face limitations in nucleus segmentation and classification and often misidentify myoblast nuclei located above or below myotubes, leading to FI overestimation. To address these issues, we developed MyoFuse, an AI-based workflow enabling fully automated and unbiased FI quantification. MyoFuse combines nucleus segmentation using Cellpose with a classification network trained with Svetlana on fluorescence images of mouse C2C12 and human primary myotubes. Performance was evaluated by comparing MyoFuse-derived FI values with expert annotations and by assessing classification accuracy at the single-nucleus level. MyoFuse achieved strong accuracy scores in mouse C2C12 and human primary myotubes (0.954 and 0.911 respectively), and showed strong correlations with expert-derived FI values (r = 0.991 and r = 0.937 respectively; p < 0.0001). Unlike current approaches that overestimate FI, MyoFuse reliably segments nuclei even within dense clusters and distinguishes myotube from myoblast nuclei based solely on cytoplasmic staining. By processing large images, it further reduces selection bias and accounts for heterogeneity in myotube density. MyoFuse thus provides a robust, high-throughput, and accurate method for FI quantification in skeletal muscle cell culture. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1038/s41598-026-40047-y.