Abstract
Automated pre-scan positioning can improve radiology workflow efficiency by reducing manual interaction during patient setup and scan planning. While external body surfaces can be acquired non-invasively using RGB-D sensors, organ-specific scan planning requires information about internal anatomy that is not directly observable. We formulate internal anatomy estimation from body-surface point clouds as a surface-conditioned implicit reconstruction problem. Our model combines a hierarchical Point Transformer encoder with a modulated SIREN decoder. The encoder extracts a patient-specific latent representation from the body surface, while the decoder predicts anatomical class logits at arbitrary 3D query coordinates, enabling dense volumetric reconstruction without a fixed output grid. We train on 4,780 subjects and evaluate on an independent test set of 1,454 subjects from the German National Cohort (NAKO), reconstructing 20 internal anatomical structures from MRI-derived body-surface point clouds simulating depth observations. Compared with the point-set-based DeformingPointTransformer (DPT) baseline, our method achieves a mean ASSD of (2.95\,\mathrm{mm}) versus the reported DPT mean CD of (4.78\,\mathrm{mm}). For HD95, reported for both methods, the aggregate error decreases from (12.06\,\mathrm{mm}) to (9.26\,\mathrm{mm}), with lower values for 17 of 20 structures. A qualitative zero-shot feasibility experiment on ten unlabeled real depth observations further shows that the trained model can process sensor-derived body surfaces without fine-tuning. These results support surface-conditioned implicit representations for dense internal anatomy estimation from external body observations. Code:
\url{https://github.com/multimodallearning/implicit-anatomy}
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/Off_Grid_029.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=3eaLLv63DL
BibTex
@InProceedings{KrnDen_SurfaceConditioned_MICCAISAT2026,
author = { Krnjaca, Denis AND Heinrich, Mattias P.},
title = { { Surface-Conditioned Implicit Reconstruction of Internal Anatomy for Automated Patient Positioning } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17277},
month = {pending},
page = {pending}
}