Volume analysis using multimodal surface similarity.

Area of Science:

• Medical Imaging
• Computer Vision
• Data Fusion

Background:

• Combining volume data from multiple imaging modalities is crucial but challenging due to differing artifacts and delineated structures.
• Leveraging joint information across modalities can extend classification capabilities, but artifacts complicate data exploitation.

Purpose of the Study:

• To present a robust framework for multimodal data fusion using an information-theoretic measure of isosurface similarity.
• To overcome challenges posed by differing artifacts and structures in multimodal datasets.
• To enable improved feature selection and classification based on inter-modality isosurface relationships.

Main Methods:

• Developed a framework utilizing an information-theoretic measure of isosurface similarity between different imaging modalities.
• Created a similarity space to provide an overview of modality differences.
• Expressed multimodal classification based on similarities and dissimilarities of individual modality isosurfaces.

Main Results:

• The proposed similarity space offers a concise overview of inter-modality differences.
• Feature selection is improved by utilizing the derived similarity space.
• Multimodal classification is effectively reformulated in terms of isosurface similarities.

Conclusions:

• The framework enables robust feature extraction and classification in multimodal imaging.
• Demonstrated efficacy in applications such as dual energy computed tomography for industrial manufacturing parts.
• Offers a novel approach to handling artifacts and structural variations in multimodal data fusion.