Tasks to Perform
- To establish and to provide a harmonized acquisition paradigm to be used in the different available study concepts and to guide new submissions
- To scientifically compare different postprocessing software approaches and to evaluate their consistence and reproducibility with the available study data
- To establish a roadmap or standard algorithm on how to acquire and to process
- Planning and testing of a centralized perfusion data evaluation to assist local and centralized blinded reads
- Data pooling of the different studies to allow for larger scale data analysis
- To establish and organize a perfusion expert board including also perfusion assessment in other organs – e.g. breast and / or prostate
- To evaluate new trends in the acquisition and post-processing of T1 dynamic data
Harmonization of the Acquisition Paradigms
As there are several IIS studies on the use of Gadovist® as contrast media for perfusion and T1dynamics currently planned and initiated, there is a need for a centralized organization and guidance of these studies.
To achieve reproducible and comparable results and to allow the pooling of the results from different studies, the following prerequisites in respect of the data acquisition have to be fulfilled:
A harmonization of the used perfusion sequences – type of sequence, temporal resolution, number of repetitions etc. - to be used for data acquisition and to fit into the planned post-processing algorithms
A standardization of the injection strategies- amount of contrast media, use of power injection, injection speed, saline flush – specific for the use of Gadovist®
The results from the expert panel discussions will enter into the standardized protocol
Standardization of the Data-Processing
To achieve an algorithm to post-process the DCE and DSC MRP data it is first necessary to establish and continuously expand a “Perfusion Platform/Database” including source data of different study populations in a homogeneous data quality allowing for secondary analysis/post-processing
The data will be used to normalize for variations and to establish a standardized data analysis including post-processing of raw data, pre- and post-processing. For this study task we will use a research software tool that allows for a free adaptation of the parameters and to calculate the data with multiple algorithms
Lately this data standardization will help to develop reproducible and comparative parameters for disease assessment.
The diseases to evaluate are defined by the individual study proposals listed and described in detail in the protocol files
- Brain tumor perfusion study (E. Larsson – Sweden)
- Cerebrovascular Disease/Stroke ( A. Dörfler – Germany)
- Brain tumor follow-up studies (M. Law – USA, TB Nguyen – Canada)
- Dementia/MCI (A. Dörfler - Germany)
Statistical Analysis
The statistical analysis plan is based on a Modeling and Validation approach.
The selection of the best perfusion protocol will be done using a modeling step to identify the best perfusion protocol across all patients, indications and scanner types (manufacturer) and to validate the finding with a set of patient data that was not used for modeling. The criteria for selecting the best protocol will be defined within the modeling step due to lack of basic data.
Step 1: Modeling
The data of the four Gadovist studies will be pooled to provide one global homogeneous acquired dataset. In a first step, 33% of the data (about 70 patients) will be selected randomly from the global dataset to be used as modeling dataset. This modeling dataset will be split into 10 equally sized sub-datasets for cross-validation. Nine sub-datasets will be used to identify the best perfusion protocol using criteria that will be selected during modeling. The tenth sub-dataset will be used to calculate the performance of the selected perfusion protocol. This will be repeated for all 10 combinations. The most frequent perfusion protocol selected will be used in the validation step. The modeling dataset is found to be not sufficient in size, when the cross-validation does not allow choosing the best perfusion protocol, e.g. due to instable models. The modeling dataset will then be increased to 50% of the global dataset (about 100 patients) and the procedure repeated. If this is still not sufficient, the modeling dataset will be increased to 67% of the global dataset (about 140 patients).
Step 2: Validation
The validation will be done using data of all patients who were not part of the modeling. This step will be performed once only. The final chosen perfusion protocol will be applied to the validation dataset to assess its performance. The validation will be done for the whole validation dataset and as subgroup analyses by indication and by scanner type (manufacturer).
Additional Perfusion Data
As the single studies will start at different time points, patient data from the clinical routine as well as from other studies using the standard protocol will be included up to 10% of the proposed total number of patients.
These data will undergo a quality control and later after the project a full analysis if they fit into the concept.