Pixel Embedding Method for Tubular Neurite Segmentation
This work addresses the problem of automatic neuron segmentation for neuroimaging researchers, offering an incremental improvement with a new evaluation metric.
The paper tackles the challenge of segmenting intricate neuronal branches in neuroimaging data by proposing a deep network for pixel-level embeddings and an end-to-end pipeline for neuron structure reconstruction, achieving a significant reduction in error rate compared to classical methods.
Automatic segmentation of neuronal topology is critical for handling large scale neuroimaging data, as it can greatly accelerate neuron annotation and analysis. However, the intricate morphology of neuronal branches and the occlusions among fibers pose significant challenges for deep learning based segmentation. To address these issues, we propose an improved framework: First, we introduce a deep network that outputs pixel level embedding vectors and design a corresponding loss function, enabling the learned features to effectively distinguish different neuronal connections within occluded regions. Second, building on this model, we develop an end to end pipeline that directly maps raw neuronal images to SWC formatted neuron structure trees. Finally, recognizing that existing evaluation metrics fail to fully capture segmentation accuracy, we propose a novel topological assessment metric to more appropriately quantify the quality of neuron segmentation and reconstruction. Experiments on our fMOST imaging dataset demonstrate that, compared to several classical methods, our approach significantly reduces the error rate in neuronal topology reconstruction.