RP01 - Context modelling and mapping of guidelines and SOPs
PI: Prof. Dr. Britta Böckmann
PhD student: Catharina Beckmann
•Relevant information of a guideline, which fits the context of the patient, but also of the user has to be searched for and compared with internal clinic standards
•Can patient data such as co-morbidities, comedications, ECOG be included to identify relevant passages?
•Is it possible to determine and display the SOP appropriate to the patient and user context?
RP02 - Analysis of unstructured text from publications
PI: Prof. Dr. Peter Horn
PhD student: Henning Schäfer
•There is currently no german terminology on melanoma
•Is a partially automated creation of terminology for malignant melanoma possible?•Which tools and resources can be used to create a German terminology?
•Which standardized terminologies/ontologies should be used to link the terms?
•How can the usefulness/correctness be evaluated?
•Are word embeddings and transformer architectures like BERT (Bidirectional Encoder Representations from Transformers) helpful in a multilingual context?
•Is the approach transferable to other diseases?
•Is the terminology useful at the point of care?
RP03 - Extraction of argumentation structures
PI: Prof. Dr. Torsten Zesch
PhD student: Marie Bexte
•Occurrence of terms not a sufficient criteria for the relevance of a document
--> Extraction of argument structures
•Documents in several languages
•Individual style (shortened, spelling mistakes)
RP04 - Analysis of clinical image data including additional clinical data
PI: PD Dr. Felix Nensa
PhD student: Yasmin Schmitt
•Radiomics: great potential in oncological imaging
•Numerous derived image parameters and often further clinical data predictive machine learning models
•DFG funded Radiomicsproject on melanoma response to immunotherapy at UH Essen (Radiology + Dermatology)
•Unresolved problem: integration into the PoCand especially interpretability by clinicians
RP05 - Analysis of pre-clinical image data including additional clinical data
PI: Prof. Dr. Markus Kukuk
PhD student: Daniel Sauter
”Can deep learning techniques be used to teach a model for predicting clinical endpoints from photographs or histological images of the primary tumor alone or from a combination of the two?”
Preclinical image data source: primary melanomas
•Photos or dermatoscopicimages
•not consistently digitized and stored in the PACS
•5-year survival rate,
•Therapy options and response
RP06 - Analysis of genome data
PI: Prof. Dr. Sven Rahmann
PhD student: Hamdiye Uzuner
Personalized representation of the reliability of genome variantsDiagnosis and therapeutic decisions in oncology are based on gene variants or tumor mutational burden.
Technology:Currently gene panels (few genes, deep sequencing); soon genome sequencing (WGS)
Question:How can we distinguish between
•„non-safe variants“ ?
Doctors at the point of care can prioritiserelevant reliable variants and probably exclude ineffective therapy options.
RP07 - Treatment decision for melanoma patients Identification of similar patients at PoC
PI: Prof. Dr. Andreas Stang
PhD student: Wolfgang Galetzka
In the treatment of melanoma patients, unusual constellationsrepeatedly occur, which make treatment decisions more difficult.
•The aim of this project is to find similar patients ("statistical patient twins") from the previous melanoma patients database of the Department of Dermatology, who have been treated in the past at the Department of Dermatology of Essen University Hospital and for whom the survival time is also available via the clinical tumourregister.
•At the PoC, doctors can then compare how similar patients have been treated in the past and what the outcome was
RP08 - Predictive modeling for patient similarity based on an openEHR model of melanoma
PI: Prof. Dr. Britta Böckmann
PhD student: Jessica Swoboda
•Currently, there are no reliable predictive biomarkers available regarding primary resistance or development of resistance to therapy or the risk of (severe) side effects, especially those caused by immunotherapy
•Can a knowledge-based relevance model, developed on the basis of previous internal patient data, provide indications for this?•How can the model be made more flexible for dynamic questions?
•How can the generated context-, patient-and user-dependent knowledge be integrated and visualized?
RP09 - Context modelling at the point of care
PI: Prof. Dr. Sabine Sachweh
PhD student: Eva Hartmann
•Determination of specific point of care influencing factors
•Formalization of a specific context model based on the factors
•Adaptive user interface concept for efficient knowledge-based decision making
RP10 - Context-sensitive, personalized search at the PoC
PI: Prof. Dr. Norbert Fuhr
PhD student: Sameh Frihat
•Search for patient independent data and knowledge
•Integrated search in multiple information sources
RP11 - Giving information to counteract wrong conclusions - Empirical study on acceptance
PI: Prof. Dr. Nicole Krämer
PhD student: Alisa Küper
•How can an information system counteract the tendency to human error (confirmation bias and availability heuristics)?
•Under what circumstances is such a system accepted?
RP12 - Evaluation and proposal system for current and relevant literature at the PoC
PI: Prof. Dr. Christoph Friedrich
PhD student: Ahmad Idrissi-Yaghir
•Guidelines on melanoma of the skin are regularly updated
•Is a suggestion system for current personalisedliterature recommendations possible?
•What criteria does a rating system for literature recommendations on skin melanoma need?
•How can such a system be evaluated quantitatively and qualitatively?
•Can word embeddingsand transform architectures improve these systems?
•Can the context modelling from the dissertation project RP9 be used for literature recommendations?