MorphoLearn: A morphology-driven workflow to decipher 3D electron microscopy segmentation in diatoms
This paper introduces MorphoLearn, a morphology-driven AI framework that enables scalable and accurate 3D segmentation of diverse diatom ultrastructures from FIB-SEM data by optimizing lightweight neural networks, leveraging transfer learning, and employing boundary-aware strategies to overcome challenges posed by high morphological variability and limited annotations.