What question did this study set out to answer?

This research aims to assess the effectiveness of geometric feature extraction in classifying tissue types using diffusion tensor imaging.

March 26, 2026Open Access

Geometric Feature Extraction for Diffusion Tensor Classification: Benchmark Results on Stanford HARDI

Key Points

This research aims to assess the effectiveness of geometric feature extraction in classifying tissue types using diffusion tensor imaging.
Applied geometric feature extraction to diffusion tensor imaging data
Operated on 3x3 symmetric positive-definite diffusion tensors
Conducted evaluations on Stanford HARDI dataset and PolSAR validation
Achieved 76.44% accuracy with geometric features compared to 69.22% with raw tensors
Improved accuracy by 7.2 percentage points
Curvature anisotropy metric effectively identifies fiber crossings that fractional anisotropy fails to detect
Cross-domain validation achieved an accuracy of 96.73% on PolSAR

Abstract

OmniSciences Technical Report TR-2026-002 We present benchmark results for proprietary geometric feature extraction applied to diffusion tensor imaging (DTI) tissue classification. The method operates on 3×3 symmetric positive-definite diffusion tensors on the Riemannian symmetric space GL+(3)/SO(3). Key results on Stanford HARDI (4-class): 76.44% accuracy with geometric features vs 69.22% with raw tensors (+7.2pp) Curvature Anisotropy (CA) metric detects fiber crossings where FA fails Cross-domain validation: same framework achieves 96.73% on PolSAR Patent-pending (U.S. Provisional Application No. 64/011,831).

Geometric Feature Extraction for Diffusion Tensor Classification: Benchmark Results on Stanford HARDI

Key Points

Abstract

Cite This Study