Biological data annotation via a human-augmenting AI-based labeling system.

Area of Science:

• Computational Biology
• Artificial Intelligence in Medicine
• Digital Pathology

Background:

• Deep learning and AI are increasingly vital in biology, driven by large datasets.
• Training AI models requires extensive, high-quality labeled data, which is challenging to obtain for microscopy images.

Purpose of the Study:

• To develop an AI system that assists in real-time data annotation for biological images.
• To reduce the manual workload and enhance the quality of annotations for AI model training.

Main Methods:

• Introduction of the Human-Augmenting Labeling System (HALS), a human-in-the-loop AI.
• HALS utilizes a multi-part AI system with three deep learning models.
• Real-time learning from human input with minimal initial examples.

Main Results:

• Demonstrated a significant manual work reduction of 90.60% in cell type annotation tasks.
• Achieved an average data quality improvement of 4.34% across various use-cases.
• Experiments conducted with seven expert pathologists validated the system's effectiveness.

Conclusions:

• HALS effectively reduces annotator workload and improves annotation quality in biological image analysis.
• The human-in-the-loop approach accelerates AI model development in fields like digital pathology.
• This system offers a scalable solution for generating high-quality labeled datasets in life sciences.