Microscope Image-Based Global Mapping and Task-Level Sequencing for Cell Micromanipulation
Year: 2026
Type: conference-paper
Abstract: This paper presents a perception-to-planning front end for microscope-guided cell micromanipulation. The framework combines stage-prior-guided image stitching for dish-scale scene construction, lightweight U-Net segmentation for mitotic target localization, and locality-aware waypoint sequencing for pre-manipulation trajectory generation within a common front-end workflow. It addresses the limited field of view and operator dependence of manual microscope operation by converting fragmented local observations into a stitched global scene and an executable target list. Results indicate that the proposed scene-construction strategy preserves seam-adjacent targets more reliably than stage-only or feature-only baselines, the task-specific detector localizes mitotic cells with useful accuracy under limited training data, and the locality-aware route reduces unnecessary long-range motion while remaining suitable for non-contact replay validation on the robotic platform. These findings support the feasibility of integrating scene construction, target-centric localization, and task sequencing as an upstream automation layer for scalable microscope-guided bio-micromanipulation. The present validation is stagewise, with pipeline-level non-contact replay used to verify interface consistency and execution readiness before contact.
Language: English
Subfield: Biophysics
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7203837223