Maintaining mitochondrial integrity is therefore crucial, as dysfunctions can lead to severe pathologies such as myopathies, neurodegenerative diseases or metabolic conditions like diabetes. Better understanding mitochondrial function and dysfunction is key to addressing these challenges. This understanding partly relies on the observation of mitochondria and the analysis of their morphology under different conditions.

Electron microscopy (EM) is the gold-standard technique for visualizing mitochondria, as it provides high-resolution imaging of cellular ultrastructure. However, segmentation and morphological analysis remain challenging due to the lack of contrast and color information in EM images. Existing pipelines to overcome these limitations are often time-consuming or too complex for users without experience in advanced deep-learning models.

To tackle this challenge, a research team from the Restore Institute in Toulouse, led by Mathieu Vigneau and Jean-Philippe Pradère, has developed EMito-Metrix, a computational tool designed for the automatic segmentation and analysis of mitochondria from 2D EM images.

The team created six species-specific models and one generalist model by training a segmentation algorithm with their own annotated EM images. Their results demonstrate that the tool enables highly specific detection of mitochondria according to their species of origin. With its user-friendly interface, EMito-Metrix allows users to easily visualize and analyze 26 mitochondrial metrics, presented through automatically generated graphs. In addition, EMito-Metrix includes a machine learning module that provides predictive analytical capabilities to assess how experimental factors, such as genetic mutations or drug treatments, may affect mitochondrial morphology and ultrastructure.

To validate their tool, the researchers analyzed mitochondria across the entire tree of life. More than 35 000 mitochondria were processed, with over 800 objects per species. The results obtained with EMito-Metrix are compelling, enabling precise segmentation of mitochondria by species and efficient quantitative analysis of their metrics.

The AI algorithm is capable of accurately detecting mitochondria (in colour) from tissues from different species imaged in ME (left). For each segmented mitochondrion, the tool extracts 26 morphology and ultrastructure metrics that can be displayed using graphs. The radar plot (centre) illustrates striking differences in metrics between vertebrates and invertebrates. Based on these metrics, the neural network is able to predict the class to which each mitochondrion belongs with 94% accuracy (right).

In conclusion, EMito-Metrix supports mitochondrial research by simplifying morphological analysis, saving researchers valuable time and reducing the risk of bias.

Access to EMito-Metrix solution here: https://www.emitometrix.org/