Ask more Info
8-2013-736

Multi-Modal Deep Neural Networks for Multi-Sequence MRI Guided Procedures

Image guided surgery is becoming the standard for high accuracy, low risk procedures. In this context, MRI is often the imaging modality of choice since it provides unrivalled contrast between soft tissues and consequently best detectability of the procedure target. MRI is conceptually a multi-sequence technique in which multiple scans providing complementary information are used jointly to optimize target lesion detection as well as safe navigation to the target.
A joint research of Nahum Kiryati, TAU and Arnaldo Mayer, Sheba Medical Center.
Two major applications are the detection of target lesions for prostate biopsy and the mapping of critical brain structures in neurosurgery. The first application has the purpose of selecting the right sampling sites in a prostate biopsy (Fig. A), avoiding painful sampling at many unsure sites or even missing the correct one. The second application maps critical white matter (WM) areas of the brain such as motor, language, and vision related tracts (Fig. B) to avoid their damage and related morbidity during the resection of a brain tumor. Both applications are heavily dependent on manual analysis of multiple complex MRI sequences requiring extensive radiological expertise in prostate imaging or neuroanatomy. In brain surgery scenario many hours may be required to perform the necessary mapping, an amount of time which is often unavailable for urgent operations. For prostate, while biopsies are usually performed by urologists, however the required knowledge to accurately select biopsy targets on multi-sequence MRI is often available to few body radiologists, highly specialized in prostate imaging.
In this context, we have developed at the CILAB (Sheba Medical Center) in collaboration with Tel Aviv University, a unique AI technology for the 3-D analysis of multiple MRI sequences. The technology relies on proprietary (patent pending) deep neural networks that automatically detect target lesion or  map complex anatomical structures in multi-sequence MRI for image guided procedures. The first targeted application is for MRI guided brain surgery and prostate biopsy.
The brain application has reached clinical prototype maturity, being used in Sheba as a support tool for real pre-surgical planning. The validation was published in top tier journal IEEE transactions in medical imaging and is the object of a grand challenge with over 100 research group participating around the world.