Image-guided neuro-surgical techniques, which utilize both pre- and intra-operative imaging data, are increasingly employed to facilitate surgical treatment of brain tumors. In the absence of a gold standard establishing the precise margin of the imaged tumor, exploiting multi-modality data is challenging. In addition, there exist high variabilities between and within segmentation performers. The goal of this proposal is to develop a computer-assisted informatics tool, namely neurosurgical decision aid (NDA), for improving tumor resection in image-guided neurosurgery. NDA a statistical algorithm-based tool designed to assist surgeons in making a pre-operative resection plan by (1) combining several human experts' decisions on target resection regions based on their independent segmentation results, and by (2) combining information derived from different imaging modalities or sequences. New expectation-maximization algorithms for estimating voxel-wise gold standards will be created. Both statistical simulations and studies using established digital phantoms with known gold standards will be conducted to test the performance of NDA. Immediately post-operatively, NDA then compares the tumor removal rate against the rate predicted in the pre-operative target resection plan. Finally, tumor recurrence rates will be compared in a two-sample clinical study of resected low-grade brain tumors including astrocytomas and oligodendrogliomas, with and without the assistance of NDA. The main goal of the proposed surgical planning and evaluation methodology is to achieve improved localization of lesions, precise definition of tumor margins, and better understanding of tumor relationship with functionally essential gray and white matter structures. The methodological development may also be applicable to other surgical applications.
Showing the most recent 10 out of 42 publications
