MitoVis: A Unified Visual Analytics System for End-to-End Neuronal Mitochondria Analysis

Area of Science:

• Neuroscience
• Cell Biology
• Bioinformatics

Background:

• Mitochondrial morphology impacts neuronal function and is linked to neurodegenerative diseases.
• Current mitochondria analysis relies on manual methods or deep learning with limitations in accuracy and adaptability.
• Automated analysis faces challenges due to model variability and the need for human proofreading.

Purpose of the Study:

• To introduce MitoVis, a novel system for end-to-end visualization and interactive analysis of neuronal mitochondria morphology.
• To develop an active learning framework for accurate fine-tuning of deep learning models for mitochondria analysis.
• To provide visual guides for efficient interactive error correction in mitochondria analysis.

Main Methods:

• Developed MitoVis, an end-to-end visualization system for neuronal mitochondria.
• Implemented a novel active learning framework using contrastive learning for model fine-tuning.
• Integrated interactive visual guides for efficient error identification and correction.

Main Results:

• MitoVis enables accurate fine-tuning of neural network models through its active learning framework.
• Interactive visual guides facilitate rapid identification and correction of analysis errors.
• Case studies demonstrated up to 13.3x faster total analysis time compared to conventional manual workflows.

Conclusions:

• MitoVis offers a powerful and efficient solution for neuronal mitochondria morphology analysis.
• The system addresses limitations of existing manual and automated methods.
• MitoVis significantly accelerates research, aiding the study of neuronal function and neurodegenerative diseases.