Answers are grounded in this paper and cite their sources.
CPSA-SAM2 is an instance-segmentation model tailored to cell analysis in optomechatronic medical imaging, where speckle noise, a limited numerical aperture and real-time control loops constrain conventional networks. The model augments SAM2 with a positional attention module that encodes optical geometry and lens distortion, a semantic attention module that recalibrates channel responses, and low-rank adaptation for embedded deployment. On confocal and cytology sources rendered with physically grounded degradations it reaches a Dice coefficient of 0.9645 and a 95% Hausdorff distance of 4.87 pixels, generalises to real endoscopic and microscopy datasets, and sustains about 25 frames per second at 512×512, supporting closed-loop optical diagnosis.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: