Abstract

MIDAS (Medical Image Dataset Annotation Service) is a custom-tailored tool for creating and managing datasets either for deep learning, as well as machine learning or any form of statistical research. The aim of the project is to provide one-fit-all platform for creating medical image datasets that could easily blend in hospital's workflow. In our work, we focus on the importance of medical data anonimization, discussing the flaws of current solutions and how they can be made more secure. We also explore the concept of automation in medical data aggregation, discussing in detail the DICOM format and how it can be used to automatically extract statistical information from existing medical cases. In the solution, presented by us, data uploaded to our system is anonymised, so that images in datasets will contain no personal information. Apart from sheer data aggregation, annotators are able to select the area of interest (eg. tumor) and add additional parameters, such as tumor size. After data collection, MIDAS is able to calculate statistics and draw graphs from annotated images, as well as DICOM metadata and additional data provided by doctors, such as mean BMI, average tumor size or percentage of malicious tumors per age group.

Citations

Authors (7)