Multivariate mixture model for myocardium segmentation combining multi-source images (1612.08820v1)

Abstract: This paper proposes a method for simultaneous segmentation of multi-source images, using the multivariate mixture model (MvMM) and maximum of log-likelihood (LL) framework. The segmentation is a procedure of texture classification, and the MvMM is used to model the joint intensity distribution of the images. Specifically, the method is applied to the myocardial segmentation combining the complementary texture information from multi-sequence (MS) cardiac magnetic resonance (CMR) images. Furthermore, there exist inter-image mis-registration and intra-image misalignment of slices in the MS CMR images. Hence, the MvMM is formulated with transformations, which are embedded into the LL framework and optimized simultaneously with the segmentation parameters. The proposed method is able to correct the inter- and intra-image misalignment by registering each slice of the MS CMR to a virtual common space, as well as to delineate the indistinguishable boundaries of myocardium consisting of pathologies. Results have shown statistically significant improvement in the segmentation performance of the proposed method with respect to the conventional approaches which can solely segment each image separately. The proposed method has also demonstrated better robustness in the incongruent data, where some images may not fully cover the region of interest and the full coverage can only be reconstructed combining the images from multiple sources.

• The paper introduces a multivariate mixture model that fuses LGE, T2-weighted, and bSSFP images for enhanced myocardial segmentation.
• It employs a maximum log-likelihood framework with affine and free-form deformation registration to correct inter- and intra-image misalignment.
• Numerical results show significant improvements in Dice scores, demonstrating increased segmentation accuracy and robustness for MI analysis.

Multivariate Mixture Model for Myocardial Segmentation Combining Multi-Source Images

The paper presents a method designed to achieve superior myocardial segmentation by leveraging multi-source cardiac MRI data, specifically combining three sequences: late gadolinium enhancement (LGE), T2-weighted, and balanced-steady state free precession (bSSFP) cine sequences. This approach addresses significant limitations in traditional myocardial segmentation methods, particularly issues related to indistinguishable myocardial boundaries in LGE images, which are critical in visualizing myocardial infarction (MI).

Methodological Innovation

The proposed technique utilizes a multivariate mixture model (MvMM) within a maximum log-likelihood (LL) framework. The MvMM models the joint intensity distributions of multi-source images, allowing for simultaneous segmentation. This is crucial in overcoming misalignment challenges across different imaging sequences, achieved by integrating transformations within the LL framework. These transformations correct both inter- and intra-image misalignments using affine and free-form deformation (FFD) registration techniques, yielding improved delineation of pathological myocardial boundaries.

Numerical Results and Method Performance

Quantitatively, the proposed method demonstrates statistically significant improvements in segmentation accuracy over conventional approaches that treat each image separately. For LGE MRI, the method reports Dice scores of 0.866 for endocardium and 0.896 for epicardium, with a myocardium Dice score of 0.717. Notably, the introduction of shift correction resulted in further enhancements to these metrics, emphasizing the importance of addressing intra-image motion in clinical imaging scenarios.

Discussion

The segmentation results from multi-source images exhibit enhanced robustness against incomplete coverage scenarios, where certain images do not fully cover the region of interest. This reflects the model's ability to reconcile diverse information from complementary image sources to construct a cohesive segmentation. The approach's adaptability to hetero-coverage data makes it particularly valuable in clinical environments, where imaging modality differences can complicate analysis.

Implications and Future Directions

From a practical standpoint, this research provides a framework for translating comprehensive, multi-source MRI data into actionable diagnostic insights. Theoretically, the method enriches the landscape of medical image segmentation by demonstrating the feasibility and benefits of an integrated multi-source approach. The implications extend beyond cardiac imaging, suggesting potential applications in other domains where multi-modality data could enhance diagnostic accuracy.

Further exploration could include investigating non-rigid inter-sequence alignment methods and extending the framework to additional imaging modalities such as CT or PET scans. Such endeavors could further validate and expand the applicability of the presented framework, promoting broader clinical adoption and integration into routine diagnostic workflows.

In conclusion, the proposed multivariate mixture model contributes valuable advancements to automated medical image segmentation, offering a statistically robust and practically viable solution to a longstanding clinical challenge in myocardial analysis.