ML in Cardiovascular Imaging
Semi-supervised segmentation of multiple vendor scans
Cardiac image segmentation is an important first step for many approaches to quantitative analysis for cardiac diagnostic assessment. This process requires partitioning the image into a number of clinically meaningful regions such as the left ventricle, right ventricle, or myocardium. Acquiring this information allows clinicians to understand important features such as the ejection fraction and the volume that the heart is managing at different times. Those features are later used to determine if there is any possible pathology or whether an intervention is required.
In this work, we targeted the challenge of segmenting volumes of patients scanned by machines of different vendors by using synthetic labelling of unannotated time-frames with image registration, proposing a semi-supervised method.
(images source: https://doi.org/10.5281/zenodo.3715890)
Interpretability of a Deep Learning Model in the Application of Cardiac MRI Segmentation with an ACDC Challenge Dataset
Cardiac Magnetic Resonance (CMR) is the most effective tool for the assessment and diagnosis of aheart condition, which malfunction is the world’s leading cause of death. Software tools leveragingArtificial Intelligence already enhance radiologists and cardiologists in heart condition assessment buttheir lack of transparency is a problem. This project investigates if it is possible to discover conceptsrepresentative for different cardiac conditions from the deep network trained to segment crdiacstructures: Left Ventricle (LV), Right Ventricle (RV) and Myocardium (MYO), using explainabilitymethods that enhances classification system by providing the score-based values of qualitativeconcepts, along with the key performance metrics. With introduction of a need of explanations inGDPR explainability of AI systems is necessary. This study applies Discovering and Testing withConcept Activation Vectors (D-TCAV), an interpretaibilty method to extract underlying featuresimportant for cardiac disease diagnosis from MRI data. The method provides a quantitative notionof concept importance for disease classified. In previous studies, the base method is applied to theclassification of cardiac disease and provides clinically meaningful explanations for the predictions ofa black-box deep learning classifier. This study applies a method extending TCAV with a Discoveringphase (D-TCAV) to cardiac MRI analysis. The advantage of the D-TCAV method over the basemethod is that it is user-independent. The contribution of this study is a novel application of theexplainability method D-TCAV for cardiac MRI anlysis. D-TCAV provides a shorter pre-processingtime for clinicians than the base method.