The use of imaging in radiotherapy is far-reaching and rapidly expanding. However, image quality and interpretation vary greatly depending on experience and access to imaging technologies. Computer-based image analysis is also expanding to aid interpretation and utilization of the vast amount of new imaging data.
The list of potential useful imaging features and measures is growing rapidly, increasing hope for true individualized diagnostics and prognostics. These imaging biomarkers are gaining popularity both due to their non-invasiveness and due to technological advancements allowing a glance into ever more subtle biological processes at a higher resolution.
The aim of this WP2 is to aid the use of these new possibilities. This will be approached in two ways:
- Adding to the common framework radiotherapy centers need to stand on in order to successfully conduct studies based on medical imaging
- Investigate specific potential valuable information that can be derived from medical images
WP2.1 National Imaging QA
In many multicenter-imaging studies, little initial imaging QA is performed rendering the conclusions drawn from the acquired data weak.
Establish a framework that supports the use of a nationwide imaging protocol in multicenter studies.
WP2.1 will focus on a national MRI protocol. We will facilitate phantom measurement of quantitative MR metrics, derived parameters from diffusion MRI (e.g. ADC), and hemodynamic parameters (DCE/DSC). The geometrical validity of the scans will be assessed.
Phantoms will be scanned in clinical setup using a set of non-clinical benchmark- and clinical sequences. To ensure consistent quantification between participating centers in multicenter studies, imaging of calibration phantoms with the proposed sequences should be performed.
Improvement of local imaging capabilities. Improved quality of nationwide imaging protocol facilitating faster translation of promising imaging biomarkers and technologies into the clinic. WP2.1 is a precursor for WP2.2.
Eliminating a lack of consistency in image acquisition. The work will be linked directly to the multicenter imaging protocols within the DCCC.
WP2.2 Reproducibility of image features
Several studies have shown poor reproducibility of extracted spatial information and functional parameters from different images. For instance, different results have been obtained when calculating an image parameter on two scans separated only by a short period of time.
To provide a method for evaluation of the robustness of image features
i) Set up a standard pipeline for evaluating the reproducibility of a selection of image features on a set of images.
ii) Apply pipeline to evaluate the reproducibility of quantitative parameters on both anatomical (CT, MR) and functional (e.g. PET, MRDWI) images using both phantom and patient scans.
Compile a list of planned or existing patient data sets available from the participating centers (e.g. repeated MR scans from patients with brain metastases are available from WP2.4). Phantom data will be acquired in WP2.1.
For the first version of the data pipeline, the focus will be on 1) setting up image loading, 2) calculation of a standard set of parameters, and 3) comparison of the calculated parameters between different scans. Collaboration with WP12.
WP2.2 will lead to a better understanding of which features should be included in imaging biomarker studies.
This work – in conjunction with the image acquisition QA work described in WP2.1 – will lead to more robust conclusions in studies evaluating the use of imaging biomarkers.
WP2.3 AI for lung substructures
WP2.3 addresses two clinical needs: 1) To understand the dependence of severe acute toxicity on the exposure of substructures of the lung, 2) automated image annotation and dose registration in order to gain sufficient data at an acceptable cost to an advantage that understanding. WP2.3 will link to the QA tasks (WP2.1) and may expand to other disease sites or inform local studies of functional sparing strategies in lung cancer in the associated DCCC centers.
i) Artificial intelligence (AI) based vessel and airway segmentation on CT scans. Test dependence on CT image acquisition protocol, cf. WP2.1
ii) Investigate the possibility for AI base airway and vessel segmentation on CBCT across centers.
iii) Demonstrate method to correlate dose with short term vessel/airway changes
Dose plans, planning CT, CBCT, and first follow-up CT scans will be extracted from participating centers (confirmed RH and Herlev, open for all). Data will be exported to computer scientists for annotation and dose registration. Dose-response modeling based on annotated features will be attempted in collaboration with WP14.
Automated substructure annotation in the lung. Correlate radiation dose exposure to airway and vessel damage.
WP2.3 will aid our understanding of the dependence of severe acute toxicity on the exposure of substructures of the lung.
WP2.4 The added prognostic value of MRI biomarkers
In quantitative imaging (qMRI) ADC represents a strong candidate for an imaging biomarker. WP2.4 investigates if ADC is adding truly independent information to treatment response prediction.
To provide a framework for how to test if a candidate imaging biomarker adds to the potentially available prognostic capacity of baseline clinical data. Specifically, WP2.4 will show if ADC is adding prognostic information for patients suffering from brain metastases
In collaboration with WP14 a predictive model will be established based on patients treated for brain metastases in the involved centers (currently 3 centers). Baseline clinical data and follow-up data will be collected.
The clinical data will be used to generate a multi-endpoint model combining three-cause specific cox models: 1. Tumor relapse in-field, 2. tumor relapse in the brain, 3. out of the field, tumor relapse in the body. The model will be tested on existing data from patients with brain metastases. The specificity and sensitivity of the model will be compared to ADC data.
Provide a framework for how to test if a candidate imaging biomarker adds to the potentially available prognostic capacity of baseline clinical data. Specifically, if ADC is adding information to the overall prognosis of patients suffering from brain metastases.
WP2.4 will together with WP2.2 provide information on the impact and reliability of imaging biomarkers.
Professor, medical physicsOdense University Hospital
Claus F. Behrens
Henrik Dahl Nissen
Medicinsk fysiker, PhDSygehus Lillebælt, Vejle Sygehus
MR fysiker, LektorOdense University Hospital
Ph.D.Aarhus University Hospital
Ivan R. Vogelius
Abraham George Smith
PhD studentRigshospitalet, Copenhagen
Anne Louise Højmark Bisgaard
PhD studentOdense University Hospital