Dr. Brani Vidakovic completed his undergraduate work in Mathematics and Statistics at Belgrade University and obtained a Ph.D. in Statistics from Purdue University in West Lafayette, Indiana. After spending 8 years as an Assistant and Associate Professor at Duke University, he arrived at Georgia Tech – first in the ISyE department and eventually to the Wallace H. Coulter Department of Biomedical Engineering at Georgia Tech and Emory University. In the Coulter Department, Vidakovic is the creator of the BME 2400 biostatistics course and is currently conducting research regarding the use of Bayesian statistical methodology in the study of biomedical data.
The Bayesian method, or Bayesian inference, is a relatively new alternative to the traditional “frequentist” inference which is utilized in most statistical tasks. The general Bayesian method can be described as a way of incorporating expert opinion, prior knowledge and extra experimental information into a statistical solution. Through the use of Bayes’ theorem (developed by non-conformist priest Thomas Bayes in the 18th century), “the synergy of prior information and experimental observations leads to a more precise inference.” The advantage provided by the Bayesian method depends on the amount of information available and varies on a case-by-case basis.
In a last few decades, biomedical research has reached a point where the equipment used provides more samples of information than can be individually interpreted. These “high-frequency and massive” data sets cannot be efficiently analyzed using the traditional statistical toolbox, but rather by computational analysis; the results are then inferred to a summary. This kind of analysis involves taking huge data sets and using any other knowledge about the data to reduce the size of the data set to something of meaning to the researcher, a method known as data reduction.
Another area that Vidakovic’s group studies is wavelets and their biomedical applications. Similar to Fourier transformations, wavelets map time-domain signals to time/scale domains. Several advantages arise from the inherent properties of wavelets, such as their ability to parsimoniously describe smooth or discontinuous signals as well as to explore scale-contributions to a signal. During the course of his research, Vidakovic has encountered some surprising results – “It turns out that health is manifested by irregularity of a subject’s bioresponses”. For example, the regularity of diameter changes in pupils can be used to infer ocular health or disease – if the pattern is regular, the data points towards macular degeneration, inferior eye health or damage. This turns out to be a general pattern in biological signals (EEG, mass spectra, gait, etc) and is becoming a very useful clinical diagnostic tool.
Vidakovic has created the BME-specific statistics course, BME 2400, with a primary focus on biomedical and biological applications. Vidakovic has ensured that an appropriate amount of theoretical statistical methodology is taught alongside true-to-the-world biomedical applications. Additionally, Vidakovic is facilitating young biomedical engineers in BME 1300 this semester.
Data mining, reduction and analysis is gradually growing as a need in most areas of research. The experimental equipment can now gather huge data sets at an ever-growing rate, and new approaches need to be developed to make sense of these large data sets. Vidakovic’s methodologies are becoming more fundamentally important with each day. Giving insight into the motivation behind his research, Vidakovic explains, “Thirty years ago statisticians strived for more data, so that Gaussianity could be used. Nowadays, we are drowning in oceans of data, and new methodologies need to be developed.’’