Cardiac magnetic resonance (CMR) imaging is usually used in the investigation of advanced cardiac disease, evaluation of the myocardium and assessment of cardiac function. Until recently, evaluation of images has been limited to what the eye can see, but the complex interactions between tissues at the image level is a treasure-trove of information that can only be fully utilized with computational methods. Radiomics, namely the computational analysis of medical images is recently used as a surrogate for the determination of complex image features.
In CMR, there is a growing interest in patient stratification and biomarker identification as well as identifying the features that characterize the spatial distribution and heterogeneity of a myocardial scar. The dominant method for obtaining quantitative descriptors of spatial heterogeneities is mainly based on texture analysis. These techniques encompass a large number of mathematical descriptors that can be used to evaluate the variation in intensity between voxels in a magnetic resonance imaging (MRI) slice as well as in adjacent slices, in order to retrieve measures of intra-scar heterogeneity. Though powerful, this approach does not scale well to high-dimensionality data, making the algorithms sensitive to a necessary initial feature pre-selection step. Despite several decades of research, predictive biomarkers are scarce, and are more effective at identifying non-responders than patients who may benefit from treatment.
In our laboratory, we aim to provide novel insights into radiomic and therapy responsiveness by developing new prediction methods based on robust radiomics and explainable artificial intelligence. We also use deep learning (DL) approach that uses multiple layers to progressively extract higher-level features from the raw input and predict the outcome.
Clinical Translation
|
|
|
|
|
|
Technological Innovation
|
|
|
|
Sequence Developement
|
|
|
|
|
|
