EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

MitoSkel: AI tool for semantic segmentation and quantification of mitochondria from light microscopy images

Soumaya Zaghbani, Rubaiya Kabir Pranti, Lukas Faber, Ana J. Garcia-Saez

Biomedical Signal Processing and Control · 2025-08 · 106 · 107762

DOI: 10.1016/j.bspc.2025.107762 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Biomedical Signal Processing and Control, 2025-08. Verfasst von Soumaya Zaghbani und 3 weiteren Autor:innen. Die Arbeit wurde bisher 10-mal zitiert (Crossref) und nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: Alexander von Humboldt-Stiftung.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

• MitoSkel, a AI-based tool, provides automated segmentation and quantification of mitochondrial morphology. • Introduced GAU-Net, a U-Net variant enhanced with Gabor filters and a Threshold Attention Mechanism (TAM). • Demonstrated GAU-Net’s segmentation performance across diverse datasets, cell lines, and microscopy techniques. • Comprehensive evaluation highlights the adaptability and generalizability of GAU-Net to diverse medical imaging datasets. Mitochondria are cellular organelles regulating key processes, including metabolism, calcium signaling and cell death. They form an intricate tubular network in cells whose shape is tightly associated with function and that can be readily visualized with fluorescence microscopy. However, accurately quantifying mitochondrial morphology from light microscopy images remains a complex task, which underscores the need for an efficient and user-friendly tool for automatically detecting and analyzing the organelles morphology. Here, we introduce a novel artificial intelligence-based system for automatic mitochondria segmentation and quantification called MitoSkel. MitoSkel utilizes a new architecture of U-Net that we call GAU-Net, which integrates a trainable Gabor filter layer and a Thresholding Attention Mechanism (TAM). We trained GAU-Net on a comprehensive dataset of fluorescence microscopy images to perform pixel-level segmentation of mitochondria under various conditions. Following segmentation, MitoSkel applies skeletonization to extract the morphological features of individual mitochondria, facilitating the quantification of parameters such as length, branching points, and connectivity. Thorough evaluation across diverse cell lines imaged using different microscopy methods demonstrates the proficiency of MitoSkel in handling varying mitochondrial shapes, sizes, and densities. Compared to existing methods, our approach achieved improved segmentation accuracy and efficiency. MitoSkel promises to be a valuable tool for the study of mitochondrial shape and its connection with organelle dynamics, function and related diseases.

Förderung

  • Alexander von Humboldt-Stiftung

Zitieren

APA

Zaghbani, S., Pranti, R. K., Faber, L., & Garcia-Saez, A. J. (2025). MitoSkel: AI tool for semantic segmentation and quantification of mitochondria from light microscopy images. Biomedical Signal Processing and Control, 106, 107762. https://doi.org/10.1016/j.bspc.2025.107762

Vancouver

Zaghbani S, Pranti RK, Faber L, Garcia-Saez AJ. MitoSkel: AI tool for semantic segmentation and quantification of mitochondria from light microscopy images. Biomedical Signal Processing and Control. 2025;106:107762. doi:10.1016/j.bspc.2025.107762

BibTeX
@article{Zaghbani2025,
  author = {Zaghbani, Soumaya and Pranti, Rubaiya Kabir and Faber, Lukas and Garcia-Saez, Ana J.},
  title = {MitoSkel: AI tool for semantic segmentation and quantification of mitochondria from light microscopy images},
  journal = {Biomedical Signal Processing and Control},
  year = {2025},
  volume = {106},
  pages = {107762},
  doi = {10.1016/j.bspc.2025.107762}
}
Kostenlos starten →