Uni-Loc: Weakly Supervised 3D CT Pathology Localization via Spatially Grounded Vision-Language Classification
Organizations: BioMedical Engineering and Imaging Institute, Icahn School of Medicine at Mount Sinai, New York, NY, USA
Abstract
Accurate localization of abnormalities in three-dimensional (3D) CT scans is essential for linking automated predictions with clinically relevant pathological findings. We propose Uni-Loc, a vision-language framework that unifies pathology classification and 3D localization using scan-level supervision. Uni-Loc derives both predictions and localization maps from spatial similarities between pathology-conditioned queries and image features. A token-to-voxel upsampling module enables high-resolution localization, while gradient-guided thresholding identifies pathology regions without additional learned parameters. Evaluated on five localization tasks comprising 355 annotated CT scans, Uni-Loc achieves macro-averaged voxel-level AUROC/AUPR of 96.1%/34.5%, outperforming relevance propagation (80.7%/13.9%) and gradient saliency (88.7%/8.1%). Uni-Loc also achieves classification AUROC of 88.39% on CT-RATE and 78.03% on RAD-ChestCT, improving over a matched baseline. These results demonstrate effective 3D pathology localization without requiring pathology-specific voxel-level supervision.
Figures & tables
| Method | GGO (10) | IO (112) | PE (78) | LAD (120) | AS (35) | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| AUROC | AUPR | AUROC | AUPR | AUROC | AUPR | AUROC | AUPR | AUROC | AUPR | |
| Uni-Loc | .989 .01 | .449 .10 | .948 .07 | .521 .21 | .924 .08 | .441 .23 | .979 .02 | .218 .16 | .964 .05 | .098 .09 |
| Uni-CLS + LXT | .885 .10 | .180 .08 | .777 .14 | .263 .14 | .839 .16 | .129 .13 | .830 .19 | .063 .07 | .703 .35 | .058 .10 |
| Uni-CLS + saliency | .881 .05 | .089 .03 | .844 .05 | .152 .08 | .851 .07 | .040 .03 | .956 .02 | .091 .07 | .902 .06 | .032 .04 |
| C-CRM + LXT | .756 .06 | .161 .05 | .780 .12 | .276 .16 | .815 .10 | .033 .02 | .799 .12 | .018 .03 | .584 .19 | .015 .03 |
| C-CRM (cosine) | .870 .11 | .036 .02 | .785 .15 | .178 .19 | .972 .04 | .213 .12 | .863 .13 | .048 .07 | .324 .28 | .001 .00 |
| Threshold | GGO | PE | LAD | AS | IO | Avg. |
|---|---|---|---|---|---|---|
| GradT | .492 | .363 | .208 | .096 | .391 | .310 |
| Fixed ( ) | .289 | .408 | .212 | .109 | .261 | .256 |
| Oracle global ( ) | .529 | .323 | .107 | .071 | .496 | .305 |
| Oracle per-pathology | .529 | .409 | .238 | .117 | .496 | .358 |
| Model | CT-RATE | RAD-ChestCT | ||
|---|---|---|---|---|
| AUPR | AUROC | AUPR | AUROC | |
| Uni-Loc | 65.03 | 88.39 | 55.43 | 78.03 |
| Uni-CLS | 63.77 | 87.54 | 53.53 | 77.39 |
| Chest-OMDL [ 3 ] | – | 80.70 | – | 72.00 |
| COLIPRI-CRM [ 21 ] | 61.28 | 86.12 | 52.55 | 76.57 |
Medical Image Localization
Medical image localization finds anatomical landmarks, lesions, organs, or devices in scans and videos to support diagnosis and image-guided care.
See all 111 papers