Dr. Charlotte Debus

Modern high precision radiotherapy (HPR) holds the promise to escalate the dose in the tumor and improve local control while sparing normal tissue. However, it poses several challenges, from accurate diagnosis and characterization of the tumor (especially in rare entities) over finding the right therapy option for the patient to adequate target delineation in radiotherapy treatment planning, that leads to optimal dose coverage of the tumor in combination with sparing healthy tissue as much as possible. In this context, modern medical imaging has become a vital source of information as it can map morphology and physiology over large volumes at up to submillimeter scales. Advanced, integrative image analysis combines multiple parameters from different modalities of large data cohorts in a complex manner and extracts surrogate parameters for the description of tissue physiology.

Radiomics is a new uprising research field that aims to characterize and distinguish tumors and to predict patient outcome based on image derived markers. In the process, large amounts of quantitative features are extracted from medical imaging data, which characterize the tumor morphology, physiology and spatial heterogeneity. These features are extracted from large patient cohorts and analyzed with machine learning methods and classification algorithms. The goal is to find image-derived prognosticators and biomarkers that can aid in personalizing treatment for the patient.

Analyzing the time course of longitudinal image data from follow-up allows for discrimination of different treatment fates, e.g. loco-regional failure due to recurrences, treatment induced side-effects such as inflammation or pseudo-progression, as well as development of distant metastasis and their contribution to the cancer specific cause of death.

In our group on “Radiomics in Radiation Oncology” we use means of radiomics and integrative multimodal image analysis approaches for improvement of diagnostics with respect to both tumor classifications in rare tumor entities, as well as tumor delineation for improved treatment. This includes the development of new prognosticators and predictors from dose (de)escalation trials and adaptive treatment and analysis of patient cohorts from clinical trials.

A vital task is the development of new methods regarding evaluation algorithms and software. Software tools are developed in collaboration with department of medical image computing. We aim to implement tools for all relevant image processing and analysis tasks within the MITK framework.

Our Projects:

Up to date, radiomics analysis is mostly conducted using standard morphological imaging techniques like CT and T1/T2 weighted MRI. Recent technological advances regarding faster image acquisition protocols, both in MRI and other modalities, have made dynamic image acquisition over administration of a traceable substance (PET, Perfusion MRI) possible. In the context of HP-RT, these more complex functional and metabolic imaging modalities are of special interest, as they might shed light on physiological aspects of the tumor substructure, heterogeneity and extend as well as physiological transport processes. However, quantitative analysis of such 4D data sets is non-trivial and requires advanced physical modelling and image processing.

Our goal is the integration of such physiological imaging into radiomics analysis, with special regards to dynamic acquisition methods (perfusion MRI, dynamic PET). In this context, we develop relevant software tools for data analysis as well as new models and biomarkers from these imaging techniques. Another aspect is the integration of Monte Carlo simulations of dose distributions using different biological models for consideration of uncertainties and subsequent tumor recurrences.

The PREDICT project is a newly started innovative training network (ITN) within a Marie Sklodowska-Curie action, that aims to analyze large patient cohorts of image data acquired over course of radiotherapy. Within this project three main studies are conducted:

1. In head and neck tumors, patients with HPV infection driven tumors show better response to therapy compared to patients with HPV negative tumors. Therefore, one study aims at the identification of HPV status from radiomics signatures.
2. In highly malignant brain tumors (high grade glioma) diffuse tumor infiltration pattern in surrounding normal brain tissue constitutes a therapeutic challenge for target volume definition. We use integrative multimodal image analysis for mapping of intratumoral heterogeneity and detection of tumor borders in order to improve target volume definition.
3. Formation of fibrotic tissue through irradiation is a major side effect in radiotherapy treatment of lung carcinomas. Early detection of induction of radiation-induced lung fibrosis can help in adapting therapy and preventing side effects. In preclinical studies, we analyze longitudinal lung CT data over the course of irradiation in order to derive image based radiomics signature correlating with the occurrence of fibrosis.