The results of data augmentation project were presented in the section “Computer Vision: Computer Vision Tasks or How to Correctly Prepare Images” of the DataFest Siberia conference in Novosibirsk in September 2019.
Data science specialist of Novel Software Systems Timur Fatykhov gave an overview of various data analysis methods, their advantages and disadvantages, as well as a new way to expand access to the synthesis account of MRI images by generative neural networks (GAN). This method allows to generate slices of tomograms with the presence or absence of pathologies. This approach improves the quality of the classification model, which allows more accurate determination of the presence of hydrocephalus (with an average quality metric, roc-auc is available from 0.85 to 0.95).
DataFest Conference is one of a series of international free Data Science conferences for researchers, engineers and developers in the field of data analysis. More than 30 speakers and 500 participants from different countries gathered in Novosibirsk to participate in this event.