Mikhail Valeryevich Milchenko PhD, Presenter: Nothing to Disclose

Dhanashree Rajderkar MD, Abstract Co-Author: Nothing to Disclose

Abraham Z. Snyder PhD, Abstract Co-Author: Nothing to Disclose

Joshua S. Shimony MD, PhD, Abstract Co-Author: Nothing to Disclose

Tammie Smith Benzinger MD, PhD, Abstract Co-Author: Research Grant, Eli Lilly and Company

Sarah C. Jost MD, Abstract Co-Author: Nothing to Disclose

Daniel Marcus PhD, Abstract Co-Author: Owner, Radiologics, Inc

Combining information from multiple neuroimaging MRI sequences has potential as a diagnostic tool in diseases such as glioblastoma multiforme. Our group is developing an informatics and image processing framework that allows automatic numerical comparison of data across patients. However, the quality of clinical images is not uniform. It depends on scanning equipment, technician training, patient motion and other random factors. We describe the structure of our neuro-informatics framework and discuss underlying technologies, algorithms and challenges.

Archiving of DICOM scans is performed using XNAT, an open source imaging informatics platform. Scans for each DICOM study are classified according to sequence and resolution. The pre-selected primary image is registered to standard atlas space by 12-parameter affine transformation. Additional data types, e.g., T2-weighted structural images, diffusion tensor images, perfusion maps, etc. are registered to the primary image using a multimodal registration algorithm. The registration procedure depends on an automatic set of rules reflecting the quality of each scan for registration purposes. Virtually any imaging modality, e.g. metabolic maps derived from positron emission tomography (PET), are easily integrated into the scheme.

In our experience, the quality of clinically acquired images is highly variable. Use of such data in research critically depends on achieving standardized processing. Our approach is to develop a processing pipeline that adapts to the heterogeneity of the data while maintaining a standard sequence of steps. In practice, this typically involves an iterative process that cycles between development, processing and quality control steps.

Automated quantitative evaluation of clinically derived multi-modality neuroimaging is challenging. The presented neuroinformatics framework enables automated quantitative evaluation of multi-modality neuroimaging studies that derive from clinical data. Current applications under development include improved surgical planning for glioblastoma resections and comparative evaluation of alternative perfusion and diffusion tensor imaging algorithms.

Milchenko, M, Rajderkar, D, Snyder, A, Shimony, J, Benzinger, T, Jost, S, Marcus, D, Streamlined Automatic Registration of Clinical Multi-Modality MR Head Scans for Research Applications.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12029047.html