Cluster-based human-in-the-loop strategy for improving machine learning-based circulating tumor cell detection in...

Area of Science:

• Oncology
• Biotechnology
• Computational Biology

Background:

• Liquid biopsy for metastatic cancer faces challenges in distinguishing circulating tumor cells (CTCs) from non-CTCs.
• Current gold standard methods rely on laborious manual image analysis.
• Machine learning (ML) shows promise for automation but requires expert input for uncertain predictions and limited data.

Purpose of the Study:

• To develop an automated, human-in-the-loop ML system for improved CTC detection and differentiation.
• To reduce the time and effort required for expert evaluation in liquid biopsy analysis.
• To enhance classification performance using a targeted sampling strategy with limited labeled data.

Main Methods:

• Combined self-supervised deep learning with a conventional ML classifier.
• Implemented a human-in-the-loop strategy with targeted sampling from high-uncertainty clusters.
• Iteratively refined the ML model by incorporating expert-labeled data.

Main Results:

• Demonstrated feasibility on metastatic breast cancer patient data.
• Achieved improved classification performance compared to traditional methods.
• Significantly reduced expert evaluation time versus the CellSearch system.

Conclusions:

• The proposed human-in-the-loop ML approach effectively improves CTC detection in liquid biopsies.
• Targeted sampling strategies enhance ML model performance and reduce diagnostic workload.
• This method offers a more efficient and accurate alternative to current gold standards for metastatic cancer analysis.