A Flexible Tool for Automated Cortical Sulci Extraction Songfeng Zheng and Zhuowen Tu Songfeng Zheng (1), Zhuowen Tu (2), Alan Yuille (1), Ivo Dinov (2), Paul Thompson (2), Arthur Toga (2) Address: (1) Statistics Department, UCLA; (2) LONI, UCLA

Abstract:

Cortical sulci are important structures in the human brain, therefore, automatically extracting them is desirable in the medical image community. We developed a flexible software tool for automatically extracting cortical sulci. It is a learning based tool, integrating the generative information with the discriminative information. The learning algorithm is called a Probability Boosting Tree. The software was developed with C++ and has already been integrated into the SHIVA plugin. It works on MRI volumes as well as on extracted surfaces. Most sulci detection methods in the literature detect all the sulci regions, and then specify which part is central sulcus, which part is precentral and so on; but in our system, for different sulcus, we only need to use one different model because there is no parameter to tune in the system. We have tested our system on several major cortical sulci and the results are impressive. The experiments were conducted on raw volume data, extracted surface data and brains suffering from Williams Syndrome. We also provided quantitative measures of the error. The algorithm works fast: it needs about 8 hours for training on an ordinary PC and only 1 minute for testing on a typical volume of size 181X271X181 to detect one major cortical sulcus. We know the training process is offline, so it does not matter in application as long as testing is efficient.