IT SPY 2019: The third best Masters thesis is from FIT

The winners of the 10th round of the IT SPY competition, which selects the most interesting and most innovative IT Master's theses from the Czech Republic and Slovakia, have already been announced. According to the expert jury, the third best work is the project of Dušan Drevický from FIT. It focuses on uncertainty in machine learning models and helps computers identify X-ray images where the computers are not certain with their evaluation.

The Master's thesis of Dušan Drevický responds to the growing role of neural networks and machine learning mechanisms, which often have to make their own decisions and adopt conclusions they continue to use. The problem is that they can sometimes incorrectly regard their erroneous statements as correct. For this reason, Drevický created and tested an algorithm which can determine the degree of certainty of its conclusions when examining dental X-ray images. This significantly accelerates the planning of surgeries and eliminates any potential risk of error following from computer analysis of the images.

"The greater expansion of machine learning models in medicine is still hindered by the lack of information on whether the conclusions adopted by an algorithm are correct. Errors in medicine are expensive and serious. In many cases, machine learning models are starting to be indispensable - they can often make better and more accurate conclusions than physicians. For physicians, machine learning models are real time-savers and enable them to focus more on activities that require expert decisions," explains Dušan Drevický. His work addresses the above-mentioned problems and it has been proven that his solution significantly improves the efficiency of physicians and their confidence in the use of artificial intelligence in medicine. "This is especially important in the case of dentists, as there is not enough of them in this country," adds Dušan Drevický. The thesis was created in co-operation with TESCAN 3DIM and will be incorporated in the medical diagnostic software developed by the company.

"The role of automatic learning and computer assistance is growing every year. It is becoming increasingly clear that learning mechanisms must also be balanced out with correction mechanisms. The aim of the thesis was to implement these in an environment with no room for errors - i.e. in medicine - and it produced very convincing results," said this year's academic guarantor of the competition, Prof. Mária Bieliková from FIIT STU in Bratislava. The poster with the conclusions of the thesis can be downloaded HERE and the entire the thesis can be found HERE.

IT SPY is an official competition of Czech and Slovak universities in the field of computer science and information technologies striving to identify the best theses. This year, the expert jury evaluated almost 1600 Master's theses from Czech and Slovak faculties.