eCurves: A Temporal Shape Encoding.

Area of Science:

• Medical imaging analysis
• Computational anatomy
• Biomedical engineering

Background:

• Analyzing dynamic anatomical changes from 3D+time medical scans is crucial for understanding disease progression and treatment outcomes.
• Existing methods for temporal shape analysis often rely on complex models or point-to-point correspondences, limiting their applicability.

Purpose of the Study:

• To present a novel framework for temporal shape analysis that captures the shape and changes of anatomical structures in 3D+time medical scans.
• To develop a method that is robust, easy to implement, and does not require point-to-point correspondence or prior models.

Main Methods:

• Encoding anatomical shapes using spectral signatures (eigenvalues and eigenfunctions of the Laplace operator) at each time point.
• Tracking eigenmodes across time based on eigenfunction similarity to capture morphing shapes.
• Applying the framework to cardiac datasets and using spectral signatures for classification tasks.

Main Results:

• The proposed spectral signature encoding effectively captures shape dynamics over time.
• Classifiers trained on this encoding significantly outperformed deformation-based methods in discriminating healthy individuals from Tetralogy of Fallot (TOF) patients.
• The right ventricle, significantly impacted by TOF, showed superior discrimination with the new encoding.

Conclusions:

• The temporal shape analysis framework provides a simple yet powerful approach for analyzing anatomical changes in medical imaging.
• The method demonstrates robustness by only assuming pose invariance within a time series.
• Its ease of implementation and minimal parameter dependency make it a practical tool for clinical research and applications.