Medical Imaging Analytics, Multimedia Department
Dr. Pavel Kisilev graduated from the Technion, Israel Institute of Technology in 2002 with a Ph.D. degree in Electrical Engineering. He is a lead scientist in the Medical Analytics group in the Multimedia Department at the IBM Research Center in Haifa. Prior to IBM, he was a Research Associate in the EE department at the Technion, and in 2003-2011 was a Senior Research Scientist in HP Labs, Israel. Dr. Kisilev's research interests include computer vision, statistical learning and medical imaging. Dr. Kisilev is an author of over 45 filed and granted patents, of a book chapter, and of over 40 papers in top journals and conferences in various fields of Computer Science.
Towards Intelligent Doctor Assistant
One of the biggest doctor complaints to computer-aided detection and diagnosis systems, is the lack of ineligibility and explanation of its decision process. In a sense, this is the semantic gap that needs to be closed between a 'machine' and a doctor.
In this talk, we will present Machine Learning approaches that allow us to bridge this gap. We start with the feature extraction that combines radiologists' knowledge with discriminative power of automatically constructed features using Deep Neural Nets (DNN). We then present a novel, discriminative method for automatic radiological report generation which is medically sound and based on the standard radiological terminology. We formalize this problem as learning to map a set of diverse image measurements to a set of semantic descriptor values from radiology lexicon. We use a structured learning framework to model individual semantic descriptors and their relationships. The parameters of the model are efficiently learned using the Structured Support Vector Machine (SSVM). The output report for a new image is generated in the form of a set of radiological lexicon descriptors.
Our system can actually 'explain' to a doctor why a particular diagnosis is made, using the standard radiological language.