Harvard Catalyst Profiles

Contact, publication, and social network information about Harvard faculty and fellows.

Login and Edit functionaility are currrently unavailable.

Improved Interoperability and Dissemination of Software for Simultaneous Truth an


Brain segmentation is now a key tool in clinical and translational research, and is applied to the evaluation of brain structure in health, in normal development, aging and in pathologies. Quantitative evaluation of brain structure by segmentation has been accepted as a surrogate marker in drug and treatment trials of neurological disorders. Many segmentation algorithms exist, and utilize different assumptions and models, which lead to differing performance in a range of applications and patient groups. There is a need for neuroimaging informatics software that enables rapid and effective evaluation of the quality of segmentations and segmentation algorithms that create them. There is a relative lack of experience in the quantitative neuroimaging community with a range of techniques in pediatric imaging as compared to adult imaging. Consequently, we seek to make available a neuroimaging tool that enables a broad range of users to compare their segmentation results and segmentation methods in terms of both precision (reproducibility) and accuracy (ground truth) against others. This proposal will continue to extend and develop our existing tool for the evaluation of image segmentations based on the STAPLE algorithm. The algorithm for Simultaneous Truth and Performance Level Estimation (STAPLE) described in (Warfield, Zou, and Wells 2004) has become widely cited, recently being identified as a Fast Breaking Paper by Thomson Essential Science Indicators as one of the top 1% most cited papers in the field. The algorithm is a statistical estimation procedure, initially developed for the validation of image segmentation, and also finding application in optimal combination of segmentation schemes and in label fusion methods for segmentation by registration. The algorithm currently allows only ''hard''segmentations, and does not support ''soft''segmentations with probabilistic labels. Therefore, the specific aims of this project are to: 1. Extend the algorithm to enable validation of image segmentations which have probabilistic labels, and 2. Continue the maintenance and dissemination of our documentation, tutorials and open source implementation of the STAPLE program. These specific aims will enable us to distribute the software through NITRC, maintain and enhance the software, its documentation and tutorials, and extend the algorithm and software to an entirely new class of segmentations. This project will significantly improve the ease of use of the algorithm in the neuroimaging community, enable new users to learn the concepts that are critical to validation of image segmentation, and the practical procedures for how to use the software appropriately, and to grow the user community. The extension of the algorithm to probabilistic labels will allow entirely new classes of segmentation algorithm (those which compute a probabilistic or ''soft''segmentation) to be assessed and compared. PUBLIC HEALTH RELEVANCE: This proposal will benefit public health by developing and disseminating key advances in functionality and usability of the Simultaneous Truth and Performance Level Estimation algorithm. This will enable validation of the segmentation of images acquired in neuroinformatics studies, which will improve the capacity of the research community to use, interpret and apply quantitative neuroimage assessments.


Funded by the NIH National Center for Advancing Translational Sciences through its Clinical and Translational Science Awards Program, grant number UL1TR002541.