Data Science &
Deep Learning

QANet is a neural framework for post-hoc quality assessment of cell instance segmentation in microscopy images. Instead of producing masks, it receives an image and a segmentation from any method, then predicts a quantitative quality score without requiring ground-truth annotations at inference time.

The model uses the RibCage architecture for multi-scale comparison between image content and segmentation structure, making it sensitive to both global shape consistency and fine boundary errors. It is trained with synthetically perturbed segmentations and evaluated on 2D and 3D Cell Segmentation Benchmark datasets across multiple segmentation methods.