The problem of segmenting 2D and 3D image-based filaments is crucial in a wide range of applications, including neuronal reconstruction and tracing in microscopic images, blood vessel tracing in fundus images, human vasculature segmentation in 2D digital subtraction angiography and 3D magnetic resonance angiography images. Although there are plenty of filament segmentation methods, it remains challenging to precisely segment 2D and 3D image-based filaments. This is evidenced by the recent BigNeuron initiative that calls for innovations in addressing the demands from neuronal science community where a significant number of neuronal images have been routinely produced in wet labs,
while there still lack sufficiently accurate tools to automatically segment the neurite structures.
To address this challenge, we propose in this paper a dedicated pipeline by learning structured and contextual features from data. The structured features aim to integrate local spatial label patterns into the feature space, thus endowing the follow-up tree classifiers capability to grouping training examples with similar structure into the same leaf node when splitting the feature space, and further yielding contextual features to capture more of the global contextual information.
Figure 1: Pipeline of our approach, which consists of two main components: The first component is for learning structured features, the second component is for learning context distance features and training boosted tree classifiers. Details are provided in Our approach.
- Lin Gu, Xiaowei Zhang, He Zhao, Huiqi Li, and Li Cheng. Segment 2D and 3D Filaments by Learning Structured and Contextual Features. In IEEE Transactions on Medical Imaging (TMI), 2016. [pdf] [Supplementary] [Source Codes]