Bioimage QC Agent

Bioimage QC Agent is an interactive microscopy image segmentation and quality-control platform designed to improve the reliability and reproducibility of bioimage analysis workflows. The system integrates classical watershed segmentation, Cellpose deep-learning models, and the Segment Anything Model (SAM / micro-SAM) with an agentic quality-control framework that automatically detects segmentation failures and recommends corrective actions.

The platform follows an observe → decide → act → evaluate loop. Segmentation outputs are analyzed using image-quality metrics, object statistics, and benchmarking measures to identify over-segmentation, under-segmentation, low contrast, blur, and other common failure modes. When issues are detected, the QC agent recommends parameter adjustments and can automatically rerun segmentation to improve results.

A complementary Vision Transformer (ViT)-based quality-control classifier provides a learned alternative to rule-based QC assessment, enabling automated image-quality prediction without requiring manually annotated QC datasets.

Bioimage QC Agent supports:

• Watershed-based nuclei segmentation
• Cellpose deep-learning segmentation
• Segment Anything Model (SAM) and micro-SAM segmentation
• Automated segmentation benchmarking (Dice, IoU, object counts)
• Rule-based quality-control assessment
• ViT-based QC classification
• Agentic parameter correction and rerun optimization
• Interactive visualization and reporting through a Streamlit dashboard

The platform was developed using the BBBC038 / 2018 Data Science Bowl nuclei segmentation benchmark dataset and demonstrates practical agentic AI for biomedical image analysis.

The web dashboard allows users to:

• Upload fluorescence or brightfield microscopy images
• Interactively tune segmentation parameters
• Compare raw images, masks, overlays, and object statistics
• Inspect QC metrics and segmentation performance
• View agent recommendations and automatic corrections
• Download masks, overlays, metrics, and QC reports

Open Tool