Daphne Koller : 2012 Plenary Session

 

Tuesday, April 3, 2012
Location: Fisher Conference Center, Arrillaga Alumni Center

"Data-Driven Medicine"
9:45am - 10:15am

Abstract:

Clinical markers for prognosis or diagnosis are generally carefully designed or selected by a human expert, based on prior knowledge on what might be of clinical relevance. But what about other factors that might not be as apparent to a human, and yet might be of clinical significance? In this talk, I will describe two experiments that utilized an unbiased, data-driven exploration of clinical data sets, using machine learning methods to identify patterns in the data that were of prognostic relevance.


In the first part, I will describe work on discovery and prediction from continuous physiologic (e.g., heart rate, respiratory rate) time-series data. These methods were applied to data from premature infants in the neonatal ICU (NICU), leading to the discovery of novel clinical biomarkers, and to the construction of the Physiscore, a new risk prediction score that combines patterns from continuous physiological signals to predict infants at risk for developing major complications in the NICU. Physiscore successfully predicts downstream development of morbidity in premature infants using only 3 hours of routinely collected non-invasive data. Physiscore significantly outperformed all previously developed neonatal monitoring scores developed, including those that require invasive tests.


The second part will describe the Computational Pathologist (C-Path) System, which measures a rich quantitative feature set from microscopic images from tumor biopsies. We applied the C-Path system to microscopic images from two independent cohorts of breast cancer patients, and we used C-Path measurements to construct a prognostic model. C-Path predictions were strongly associated with survival on both patient cohorts, and provided independent information of other clinical, pathological, and molecular factors. C-Path identified several novel features that are significantly associated with survival, including, strikingly, features of the non-cancer tissue surrounding the tissue.


Bio:

Daphne Koller is the Rajeev Motwani Professor in the Computer Science Department at Stanford University. Her main research interest is in developing and using machine learning and probabilistic methods to model and analyze complex domains. Her current research projects include models in computational biology, computational medicine, and in extracting semantic meaning from sensor data of the physical world. Daphne Koller is the author of over 180 refereed publications. She has received 9 best paper or best student paper awards,and has given keynote talks at over 10 different major conferences. She was awarded the Arthur Samuel Thesis Award in 1994, the Sloan Foundation Faculty Fellowship in 1996, the ONR Young Investigator Award in 1998, the Presidential Early Career Award for Scientists and Engineers (PECASE) in 1999, the IJCAI Computers and Thought Award in 2001, the MacArthur Foundation Fellowship in 2004, the ACM/Infosys award in 2008, and was inducted into the National Academy of Engineering in 2011. She also has a long-standing interest in education, and her contributions in that area were recognized via the Cox Medal for excellence in fostering undergraduate research at Stanford in 2003, and by being named a Bass University Fellow in Undergraduate Education.