Model-based generation of large databases of cardiac images: synthesis of pathological cine MR sequences from real healthy cases

Abstract

Collecting large databases of annotated medical images is crucial for the validation and testing of feature extraction, statistical analysis and machine learning algorithms. Recent advances in cardiac electromechanical modeling and image synthesis provided a framework to generate synthetic images based on realistic mesh simulations. Nonetheless, their potential to augment an existing database with large amounts of synthetic cases requires further investigation. We build upon these works and propose a revised scheme for synthesizing pathological cardiac sequences from real healthy sequences. Our new pipeline notably involves a much easier registration problem to reduce potential artifacts, and takes advantage of mesh correspondences to generate new data from a given case without additional registration. The output sequences are thoroughly examined in terms of quality and usability on a given application: the assessment of myocardial viability, via the generation of 465 synthetic cine MR sequences.

Existing System

Fast Convolution based Methodology, Myocardial Tracking and Deformation Algorithms, Bestel-Clement-Sorine Electromechanical Model.

Proposed System

We presented a new approach for the generation of realistic 3D cine MR cardiac sequences, which makes substantial improvements to the current state-of-the-art. Our contributions are both on technical and applicative aspects. First, our revised processing only involves the registration of rather similar image sequences. This easier registration problem reduces the amount of potential artifacts. In contrast, registration along the sequence involves larger transformations and potential artifacts that could propagate along the sequence. Besides, this strategy biases the validation of the synthesis methodology, if evaluated via tracking along the sequence by a similar algorithm. Then, our new pipeline takes advantage of mesh correspondences to generate variants of a given simulation. No additional registration is involved and correspondences are exact. This opens the door to the generation of large databases of synthetic cases at small computational cost, demonstrated on a given application: the assessment of myocardial viability. Finally, we propose a thorough qualitative and quantitative evaluation of the quality and usability of the generated images, in the context of the generation of large databases. We notably investigate whether the designed database is balanced below and above the diagnosis limit of a given algorithm, within the limits of detectable disease and image realism. Our pipeline, demonstrated on cine MR sequences, is highly suitable for the generation of large databases at reduced computational cost, and paves the ground for bridging the gap between synthetic mesh simulations and real imaging data.

Architecture

BLOCK DIAGRAM