Active learning framework with iterative clustering for bioimage classification

Area of Science:

• Bioimage analysis
• Computational biology
• Machine learning in life sciences

Background:

• Rapid advancements in imaging technologies generate vast amounts of research data.
• Manual classification of large bioimage datasets is time-consuming and labor-intensive.
• Need for efficient, semi-automated tools to manage and analyze complex image data.

Purpose of the Study:

• To develop a novel framework, CARTA (Clustering-Aided Rapid Training Agent), for semi-automated bioimage classification.
• To facilitate interactive annotation and feature selection for improved classification accuracy.
• To provide a tool that matches or exceeds human annotator performance in bioimage analysis.

Main Methods:

• Developed CARTA, a framework integrating active learning, genetic algorithms, and self-organizing maps.
• Implemented an interactive annotation method for user-guided feature selection and refinement.
• Validated CARTA on diverse bioimage datasets, including plant and human cell images.

Main Results:

• CARTA achieved high accuracy in classifying subcellular localization, mitotic phases, and apoptosis in cell images.
• Classification performance was comparable to or better than that of human annotators.
• Demonstrated applicability to magnetic resonance imaging of cancer cells and multicolour time-course images.

Conclusions:

• CARTA offers an efficient and accurate solution for semi-automated bioimage classification.
• The framework supports customized feature development, high-throughput phenotyping, and flexible classification schemes.
• CARTA has broad potential applications across various biological imaging research fields.