Skip to main content

A Clinical Decision Support system based on Quantitative multimodal brain MRI for personalized treatment in neurological and psychiatric disorders

Periodic Reporting for period 3 - CDS-QUAMRI (A Clinical Decision Support system based on Quantitative multimodal brain MRI for personalized treatment in neurological and psychiatric disorders)

Reporting period: 2018-09-01 to 2020-02-29

A large number of neurological disorders lack objective criteria for primary diagnosis, early differential diagnosis with regard to subtypes in treatment response and disease progression or effective therapy monitoring. Correct diagnoses at early stages and correct patient stratification are hence often severely delayed and patients are left without any appropriate treatment for years resulting in a tremendous negative socio-economic impact. In contrast to many clinical disciplines, neither predictive physiological biomarkers nor imaging-based surrogate markers have been established yet. However, scientific studies based on advanced magnetic resonance imaging (MRI) methods indicate that patients with neurological and psychiatric disorders show specific subtle changes in multiple MRI readouts that are only detectable by quantitative approaches.

Hence a clinical decision support system (CDS) enabling personalized diagnostics and treatment for neurological and psychiatric disorders is envisioned that is based on multimodal quantitative (QUA) magnetic resonance imaging (MRI) and multi-parametric
classification and shall be demonstrated for major depressive disorder (treatment response prediction) and multiple sclerosis (disease progression type prediction).
One objective of the CDS QUAMRI project is to fully integrate advanced analysis algorithms of multi-modal structural, functional and metabolic MRI data into a single software framework to make them accessible for non-expert clinical users in order to allow for large scale clinical trials and more widespread use in neuroscience as a basis for the future use in clinical decision making. The quantification and feature extraction algorithms for quantitative magnetic resonance imaging (qMRI), diffusion weighted imaging (DWI), task-based functional MRI (task-fMRI), resting-state functional MRI (rs-fMRI) and perfusion imaging (arterial spin labelling (ASL)) shall be enhanced to access the full information content of the data at high reliability as required for future use in diagnostics, stratification and monitoring of patients.

At the end of the third funding period the software development is almost finalized. The implementation of a data base system that serves as software framework for MRI data and analysis workflow management and visualization has been fully established. The adoptation and integration of MRI data analysis modules for functional imaging, perfusion imaging, quantitative structural imaging and basic analysis options for metabolic MRI has been completed. The implementation and integration of a machine learning module has been accomplished as well. An impressive number of novel MRI data analysis methods that give access to so far inaccessible tissue parameters, enhance the data quality along with the precision and accuracy of quantitative data analysis for myeline imaging, axon density imaging, functional imaging, metabolic imaging and perfusion imaging have been developed and are largely integrated into the respective modules.

The second objective of the CDS QUAMRI project is to develop a prototype Clinical Decision support system using classifiers that discriminate different patient subgroups (treatment response, disease progression) using machine-learning based classification methods and multi-modal MRI data from patients with major depressive disorder as well as multiple sclerosis.

At the end of the third funding period retrospective classification of patient subgroups has been achieved. To that multi-modal MRI data from patients with major depressive disorder, multiple sclerosis and healthy volunteers were pooled and respective features were extracted as input into the classifier development. Different machine-learning based classifiers have been developed and tested with respect to their specificity and sensitivity of retrospective classification of patients versus healthy volunteers and patient sub-groups. It was also investigated whether features with low versus high spatial resolution show better discriminating power and whether the use of multiple features extracted from complementary MRI readouts (structural, functional, metabolic) enhance the specificity and sensitivity of the classification. The best combination of classifiers and selection of features has been identified for the different classification tasks and initial prospective classification trials of single patient data sets have been performed. While distinction of patients from healthy volunteers and different disease progression subgroups in multiple sclerosis is robustly possible, the prediction of the development of multiple sclerosis from its earliest symptoms (clinical isolated syndrome) turned out to be challenging. In patients with major depressive disorder promising results with respect to the prediction of treatment response have been obtained based on structural and functional MRI data, but the sensitivity and specificity has to be further enhanced to serve as instrument in clinical decision making for single patients. To that a prospective clinical trial in patients with major depressive disorder is ongoing to collect a large and consistent data sample combining the 5 most predictive MRI modalities for different treatment options. It will be finalized end of 2020. The study investigates changes in the brain state at different time points before and after treatment as input for the development of further improved classifiers for treatment response prediction.
We obtained a software solution that offers full integration of a reconstruction, processing and quantification pipeline for 10 MRI modalities into a common data handling, visualization, feature extraction and classification framework: anatomical MRI (aMRI), quantitative relaxation rate measurements (relax), myeline water imaging (MI), task-based functional MRI (task-fMRI), resting-state functional MRI (rs-fMRI), diffusion weighted MRI (DWI), perfusion imaging (ASL), magnetic resonance spectroscopic imaging (MRSI) and non-proton imaging (xMRI). In addition, major advances with respect to novel quantification and feature extraction algorithms for these MRI modalities have been implemented and will be available to the public via distribution of open source analysis modules. This effort enables a more widespread use of state-of-the-art multi-modal quantitative MRI in neuroscience and large clinical trials. This in turn is the basis for future application of advanced MRI in clinical decision making with regard to a large number of neurological and psychiatric disorders.

The envisioned classification based Clinical Decision Support System will be a step changer in diagnostic MRI. It will make use of disease specific MRI based surrogate markers to enable (1) unambiguous early diagnostics of a significant number of neurologic and psychiatric disorders that so far lack objective diagnostics criteria, (2) the distinction of different patient subgroups responding in a different manner to a specific treatment (for instance major depression or chronic pain) or having different disease progression perspectives (for instance early schizophrenia or multiple sclerosis) and related monitoring and (3) to use this information on a case by case basis in clinical decision making in order to facilitate personalized treatment without the need for repeated unnecessary trials with inefficient treatment forms for the first time.