LINKS: Learning-based multi-source IntegratioN frameworK for Segmentation of infant brain images

摘要: Segmentation of infant brain MR images is challenging due to insufficient image quality, severe partial volume effect, and ongoing maturation myelination processes. In the first year life, contrast between white gray matters undergoes dramatic changes. particular, inverted around 6-8months age, matter tissues are isointense in both T1- T2-weighted thus exhibit extremely low tissue contrast, which poses significant challenges for automated segmentation. Most previous studies used multi-atlas label fusion strategy, has limitation equally treating different available modalities often computationally expensive. To cope with these limitations, this paper, we propose a novel learning-based multi-source integration framework segmentation images. Specifically, employ random forest technique effectively integrate features from together Here, include initially only multi-modality (T1, T2 FA) later also iteratively estimated refined probability maps matter, cerebrospinal fluid. Experimental results on 119 infants show that proposed method achieves better performance than other state-of-the-art methods. Further validation was performed MICCAI grand challenge ranked top among all competing Moreover, alleviate possible anatomical errors, our can be combined an anatomically-constrained labeling approach further improving accuracy.