Abschlussarbeiten

Synthetic Data Generation in the Medical Domain using Generative Adversarial Networks

Synthetic data generation has become important in the medical field for two reasons. First, medical data are not widely available to researchers due to patient privacy protections. Second, in some cases such as rare diseases, only limited data are available, making diagnosis or treatment difficult even for experts. Synthetic medical data generation can address these issues by providing artificial medical data that resembles real data while not associated with real patients.

The goal of this thesis is to generate synthetic tabular data relevant to the medical field using generative adversarial networks (GANs) and to investigate how synthetic data can improve data analysis (machine learning) performance compared to cases where only real data are considered.

Contact person: Dr. Sina Sadeghi <sina.sadeghi@medizin.uni-leipzig.de>

Methods for estimating discriminatory power of data record attributes in privacy-preserving record linkage

Record linkage is the challenge of identifying duplicate data records across data sources. Not every attribute holds the same power in telling two records apart, e.g. gender vs. birth date for personal records. There exist a few algorithms for computing this “discriminatory power” of attributes given two datasets and a set of known true matches. However, in the context of privacy-preserving record linkage (PPRL), this is not possible. Access to data is tightly regulated and there exists no “gold standard” for true matches. The goal of the thesis is to research methods for computing the discriminatory power of record attributes, given a single dataset in the absence of a standard for true matches, and to evaluate them on an in-house PPRL framework.

Contact person: Maximilian Jugl, email: Maximilian.Jugl@medizin.uni-leipzig.de​

NLP approaches for information extraction about patient care evaluations from public blog system

Evaluations and review statements from patients about the patient care process in a hospital are eminently important for the hospital for improving internal processes but also to detect hate speech. Publicly, such evaluations and reviews are available in blog systems.
The goal of this thesis is to use modern methods from Natural Language Processing (NLP) to extract reasonable information from such patient reviews in a way it can be used for further processing such as opinion mining and hate speech detection.

Contact person: Macedo Maia <macedo.sousamaia@medizin.uni-leipzig.de>

Data Quality Assessment Library for structured Patient Data

Assessing the data quality is an important step before the analysis can start. The current state of the data quality of a data set can be obtained in different dimensions including integrity, consistency, completeness, and accuracy. Each of these dimensions contains different domains and these include again indicators acting as measurable items for data quality. The goal of this thesis is to design a python library including most of the already defined data quality indicators for structured data.

Contact person: Dr. Sina Sadeghi <sina.sadeghi@medizin.uni-leipzig.de>