Multi-View Elliptic Fourier Analysis of the Sternum for Automated Postmortem-to-Antemortem Computed Tomography Candidate Ranking.

Accurate postmortem-to-antemortem computed tomography (CT) candidate ranking can support personal identification in forensic radiology, but scalable methods remain needed. The sternum is included in routine chest and postmortem CT, but its thin, flat morphology may not be fully captured by conventional global three-dimensional (3D) shape features. We evaluated whether a multi-view contour representation based on elliptic Fourier analysis (EFA) could improve sternum-based CT candidate ranking over conventional 3D radiomic shape features. Paired antemortem and postmortem CT scans from an institutional dataset were analyzed against a reference database augmented with public chest CT scans as distractors. Sternum masks were automatically segmented and underwent Mahalanobis distance-based quality control. Area-normalized elliptic Fourier descriptors were extracted from coronal, sagittal, and axial projections of pose-normalized masks. For internal evaluation, leave-one-person-out cross-fitting separated selection of the view combination and harmonic order from held-out evaluation. The analysis included 58 postmortem queries and 985 antemortem references. Using cross-fitted configuration selection, EFA achieved a rank-1 identification rate of 86.2%, compared with 50.0% for radiomics, with a paired difference of 36.2 percentage points (95% confidence interval, 22.4-50.0; Holm-adjusted P < 0.001). Median true-match ranks were 1.0 for EFA and 1.5 for radiomics; mean ranks were 31.9 and 15.7, reflecting a few marked EFA tail failures. Per-scan processing was dominated by automated segmentation (median, 27.1 s), whereas candidate ranking against 985 references required only 14.9 ms per query. Multi-view EFA significantly improved sternum-based candidate ranking over radiomic shape features, providing a contour-based framework for postmortem-to-antemortem CT comparison in forensic radiology.