A framework for the objective assessment of registration accuracy.

Validation and accuracy assessment are the main bottlenecks preventing the adoption of image processing algorithms in the clinical practice. In the classical approach, a posteriori analysis is performed through objective metrics. In this work, a different approach based on Petri nets is proposed. The basic idea consists in predicting the accuracy of a given pipeline based on the identification and characterization of the sources of inaccuracy. The concept is demonstrated on a case study: intrasubject rigid and affine registration of magnetic resonance images. Both synthetic and real data are considered. While synthetic data allow the benchmarking of the performance with respect to the ground truth, real data enable to assess the robustness of the methodology in real contexts as well as to determine the suitability of the use of synthetic data in the training phase. Results revealed a higher correlation and a lower dispersion among the metrics for simulated data, while the opposite trend was observed for pathologic ones. Results show that the proposed model not only provides a good prediction performance but also leads to the optimization of the end-to-end chain in terms of accuracy and robustness, setting the ground for its generalization to different and more complex scenarios.

Validation through accuracy prediction in neuroimage registration.

Validation and accuracy assessment are the main bottlenecks preventing the adoption of many medical image processing algorithms in the clinical practice. In the classical approach, a-posteriori analysis is performed based on some predefined objective metrics. The main limitation of this methodology is in the fact that it does not provide a mean to estimate what the performance would be a-priori, and thus to shape the processing workflow in the most suitable way. In this paper, we propose a different approach based on Petri Nets. The basic idea consists in predicting the accuracy that will result from a given processing on a given type of data based on the identification and characterization of the sources of inaccuracy intervening along the whole chain. Here we propose a proof of concept in the specific case of image registration. A Petri Net is constructed after the detection of the possible sources of inaccuracy and the evaluation of their respective impact on the estimation of the deformation field. A training set of five different synthetic volumes is used. Afterward, validation is performed on a different set of five synthetic volumes by comparing the estimated inaccuracy with the posterior measurements according to a set of predefined metrics. Two real cases are also considered. Results show that the proposed model provides a good prediction performance. An extended set of clinical data will allow the complete characterization of the system for the considered task.