An explainable deep learning pipeline reported near-readout performance for identifying metastasising cells in fluorescence microscopy images. The system segments single cells and classifies them as normal or metastasising, with a DINOv3 backbone reaching an F1-weighted score of 98.35%. The key addition for clinical translation is the “explainable” workflow: rather than treating results as black-box classification, the pipeline aims to provide structured reasoning support for where the model is detecting signals in images. This aligns with growing interest in evidence-linked AI for pathology and translational microscopy. For early-stage validation, the critical gap is external generalization across institutions, staining protocols, and tissue preparation variance. If validated, the approach could reduce manual analysis time in research settings and support future biomarker studies tied to metastatic biology.