Mariam Nassar

Mapping Clinical Trials to Cancer Patient Profiles

In this project, we work on a tool for clinicians to find and rank potentially relevant clinical trials for their cancer patients. This requires a more specialized search for fitting clinical trials by matching cancer patient profiles using methods of information extraction, clinical text processing, clustering and topic extraction, ranking of results and sentence complexity analysis. The goal of this project is to provide a time saving alternative to reduce the number of clinical trials which has to be read by clinicians by filtering and ranking clinical trials eligibility criteria for a given patient profile.

Developing Clinical Decision Support Systems using Machine Learning

The systematic documentation of patient health data leads to an increasing amount of structured data. This is a valuable resource that can be used for personalized diagnosis and therapy. However, analyzing ths data faces challenges like the sparsity and heterogenity of data.
In collaboration with the company Healthcare X.0, we are developing Clinical Decision Support System based on Machine Learning approaches to find patterns in a patient’s data and comparing these with other patients and to use this information to assist the diagnostic and therapeutic decision process in the clinics.

Identifying White Blood Cells using Machine Learning
Identifying the number of different white blood cells (WBC) in human blood is an established clinical routine, where WBC are labeled with fluorescent markers.
I work on a novel approach based on machine learning where WBC are identified label-free, i.e., without any markers.
I am developing an open-source workflow that seamlessly connects the recorded images from the instruments with machine learning. The goal is a presentation of the results relevant for clinicians.
This enables fast, cheap and highly accurate identification of WBC, without destroying the cells and leaves marker channels free to answer other biological questions.

Information retrieval and ranking for computational models
Modelling biological systems is necessary to characterize biological phenomena and to study the behavior of biological systems. As the science is developing day by day, the number of models is rising too. Therefore searching for models is becoming more and more important. Models can be searched by different aspects. For example we can search models by keywords or by certain structures. Models can be searched by their ontology terms too. However, as we are to optimize methods to search for models, combining search results is equally important.
From this perspective, we are interested in how to create a metric to combine results from our current ranked retrieval into a single similarity score.