Efficient Pairwise Neuroimage Analysis Using the Soft Jaccard Index and 3D Keypoint Sets.

Area of Science:

• Medical image analysis
• Computer vision
• Bioinformatics

Background:

• Large-scale medical image indexing requires robust similarity measures.
• Existing methods often struggle with the inherent uncertainty in keypoint data.
• Predicting familial relationships from medical imaging is an emerging challenge.

Purpose of the Study:

• To introduce a novel pairwise distance measure for medical image keypoint sets.
• To generalize the Jaccard index using soft set equivalence (SSE) and an adaptive kernel framework.
• To enable accurate large-scale medical image indexing and family relationship prediction.

Main Methods:

• Developed a generalized Jaccard index incorporating soft set equivalence (SSE).
• Introduced an adaptive kernel framework to model uncertainty in keypoint appearance and geometry.
• Proposed a new kernel to quantify keypoint geometry variability (location and scale).
• Estimated the distance measure for O(N^2) image pairs in [Formula: see text] operations using keypoint indexing.

Main Results:

• Achieved near 100% accuracy in identifying monozygotic twins from T1-weighted MRI brain volumes.
• Successfully predicted family relationships, outperforming standard hard set equivalence (HSE) and appearance kernels.
• Demonstrated a practical application by automatically pairing three subjects with uncertain genotyping to their families.
• Predicted group categories, achieving an Area Under the Curve (AUC) of 0.97 for sex prediction.

Conclusions:

• The proposed pairwise distance measure effectively handles keypoint uncertainty for medical image indexing.
• The method enables accurate prediction of familial relationships and group categories from brain imaging data.
• This work represents a significant advancement in image-based family identification and medical image analysis.