Automatic image annotation for fluorescent cell nuclei segmentation.

Area of Science:

• Life Sciences
• Biotechnology
• Computational Biology

Background:

• Dataset annotation is crucial for training deep learning models, especially in life science microscopy for tasks like instance segmentation.
• Manual or semi-automated annotation of cell nuclei in microscopy images is labor-intensive and costly, despite data augmentation techniques.
• Accurate cell nuclei segmentation is vital for various biological research applications.

Purpose of the Study:

• To develop a fully automated system for annotating custom fluorescent cell nuclei image datasets.
• To significantly reduce the time and cost associated with cell nuclei image annotation.
• To provide high-quality training data for machine learning models used in cell nuclei identification and segmentation.

Main Methods:

• Development of a novel system for fully automating the annotation process of fluorescent cell nuclei images.
• Utilizing the automated system to generate annotated datasets for deep learning applications.
• Comparative analysis of segmentation performance between automatically and manually annotated datasets.

Main Results:

• Achieved a reduction in nuclei labeling time by up to 99.5%.
• Demonstrated that automatically annotated datasets provide segmentation performance equal to manually annotated data.
• Showcased a unified workflow from raw data to segmentation and tracking without pre-trained models.

Conclusions:

• The proposed system effectively automates cell nuclei image annotation, drastically cutting down labeling time and costs.
• Automated annotation produces high-quality training data, achieving segmentation performance comparable to manual annotation.
• This approach enables a streamlined, end-to-end workflow for nuclei segmentation and tracking in microscopy, independent of external models or datasets.