Radiation-Preserving Selective Imaging for Pediatric Hip Dysplasia: A Cross-Modal Ultrasound-Xray Policy with Limited Labels

Duncan Stothers, Ben Stothers, Emily Schaeffer, Kishore Mulpuri
Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare, PMLR 317:239-246, 2026.

Abstract

We study an ultrasound-first, radiation-preserving policy for developmental dysplasia of the hip (DDH) that requests an X-ray (XR) only when needed. We (i) pretrain modality-specific encoders (ResNet-18) with SimSiam on a large unlabelled registry (37,186 ultrasound; 19,546 radiographs), (ii) freeze the backbones and fit small, measurement-faithful heads on DDH-relevant landmarks and measurements, (iii) calibrate a one-sided conformal deferral rule on ultrasound predictions that provides finite-sample marginal coverage guarantees under exchangeability, using a held-out calibration set. Ultrasound heads predict Graf $\alpha$/$\beta$ and femoral head coverage; X-ray heads predict acetabular index (AI), center-edge (CE) angle and IHDI grade. On our held-out labeled evaluation set, ultrasound measurement error is modest (e.g., $\alpha$ MAE $\approx$ 9.7$\circ$, coverage MAE $\approx$ 14.0 percentage points), while radiographic probes achieve AI and CE MAEs of $\approx$ 7.6$\circ$ and $\approx$ 8.9$\circ$, respectively. The calibrated US-only policy is explored across rule families (alpha-only; alpha OR coverage; alpha AND coverage), conformal miscoverage levels ($\delta$$\alpha$,$\delta$cov), and per-utility trade-offs using decision-curve analysis. Conservative settings yield high coverage (e.g., $\tilde$0.90 for $\alpha$) with near-zero US-only rates; permissive settings (e.g., alpha OR coverage at larger deltas) achieve non-zero US-only throughput with expected coverage trade-offs. The result is a simple, reproducible pipeline that turns limited labels into interpretable measurements and tunable selective imaging curves suitable for clinical handoff and future external validation.

Cite this Paper

BibTeX
@InProceedings{pmlr-v317-stothers26a, title = {Radiation-Preserving Selective Imaging for Pediatric Hip Dysplasia: A Cross-Modal Ultrasound-Xray Policy with Limited Labels}, author = {Stothers, Duncan and Stothers, Ben and Schaeffer, Emily and Mulpuri, Kishore}, booktitle = {Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare}, pages = {239--246}, year = {2026}, editor = {Wu, Junde and Pan, Jiazhen and Zhu, Jiayuan and Luo, Luyang and Li, Yitong and Xu, Min and Jin, Yueming and Rueckert, Daniel}, volume = {317}, series = {Proceedings of Machine Learning Research}, month = {20--21 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v317/main/assets/stothers26a/stothers26a.pdf}, url = {https://proceedings.mlr.press/v317/stothers26a.html}, abstract = {We study an ultrasound-first, radiation-preserving policy for developmental dysplasia of the hip (DDH) that requests an X-ray (XR) only when needed. We (i) pretrain modality-specific encoders (ResNet-18) with SimSiam on a large unlabelled registry (37,186 ultrasound; 19,546 radiographs), (ii) freeze the backbones and fit small, measurement-faithful heads on DDH-relevant landmarks and measurements, (iii) calibrate a one-sided conformal deferral rule on ultrasound predictions that provides finite-sample marginal coverage guarantees under exchangeability, using a held-out calibration set. Ultrasound heads predict Graf $\alpha$/$\beta$ and femoral head coverage; X-ray heads predict acetabular index (AI), center-edge (CE) angle and IHDI grade. On our held-out labeled evaluation set, ultrasound measurement error is modest (e.g., $\alpha$ MAE $\approx$ 9.7$\circ$, coverage MAE $\approx$ 14.0 percentage points), while radiographic probes achieve AI and CE MAEs of $\approx$ 7.6$\circ$ and $\approx$ 8.9$\circ$, respectively. The calibrated US-only policy is explored across rule families (alpha-only; alpha OR coverage; alpha AND coverage), conformal miscoverage levels ($\delta$$\alpha$,$\delta$cov), and per-utility trade-offs using decision-curve analysis. Conservative settings yield high coverage (e.g., $\tilde$0.90 for $\alpha$) with near-zero US-only rates; permissive settings (e.g., alpha OR coverage at larger deltas) achieve non-zero US-only throughput with expected coverage trade-offs. The result is a simple, reproducible pipeline that turns limited labels into interpretable measurements and tunable selective imaging curves suitable for clinical handoff and future external validation.} }
Endnote
%0 Conference Paper %T Radiation-Preserving Selective Imaging for Pediatric Hip Dysplasia: A Cross-Modal Ultrasound-Xray Policy with Limited Labels %A Duncan Stothers %A Ben Stothers %A Emily Schaeffer %A Kishore Mulpuri %B Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare %C Proceedings of Machine Learning Research %D 2026 %E Junde Wu %E Jiazhen Pan %E Jiayuan Zhu %E Luyang Luo %E Yitong Li %E Min Xu %E Yueming Jin %E Daniel Rueckert %F pmlr-v317-stothers26a %I PMLR %P 239--246 %U https://proceedings.mlr.press/v317/stothers26a.html %V 317 %X We study an ultrasound-first, radiation-preserving policy for developmental dysplasia of the hip (DDH) that requests an X-ray (XR) only when needed. We (i) pretrain modality-specific encoders (ResNet-18) with SimSiam on a large unlabelled registry (37,186 ultrasound; 19,546 radiographs), (ii) freeze the backbones and fit small, measurement-faithful heads on DDH-relevant landmarks and measurements, (iii) calibrate a one-sided conformal deferral rule on ultrasound predictions that provides finite-sample marginal coverage guarantees under exchangeability, using a held-out calibration set. Ultrasound heads predict Graf $\alpha$/$\beta$ and femoral head coverage; X-ray heads predict acetabular index (AI), center-edge (CE) angle and IHDI grade. On our held-out labeled evaluation set, ultrasound measurement error is modest (e.g., $\alpha$ MAE $\approx$ 9.7$\circ$, coverage MAE $\approx$ 14.0 percentage points), while radiographic probes achieve AI and CE MAEs of $\approx$ 7.6$\circ$ and $\approx$ 8.9$\circ$, respectively. The calibrated US-only policy is explored across rule families (alpha-only; alpha OR coverage; alpha AND coverage), conformal miscoverage levels ($\delta$$\alpha$,$\delta$cov), and per-utility trade-offs using decision-curve analysis. Conservative settings yield high coverage (e.g., $\tilde$0.90 for $\alpha$) with near-zero US-only rates; permissive settings (e.g., alpha OR coverage at larger deltas) achieve non-zero US-only throughput with expected coverage trade-offs. The result is a simple, reproducible pipeline that turns limited labels into interpretable measurements and tunable selective imaging curves suitable for clinical handoff and future external validation.
APA
Stothers, D., Stothers, B., Schaeffer, E. & Mulpuri, K.. (2026). Radiation-Preserving Selective Imaging for Pediatric Hip Dysplasia: A Cross-Modal Ultrasound-Xray Policy with Limited Labels. Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare, in Proceedings of Machine Learning Research 317:239-246 Available from https://proceedings.mlr.press/v317/stothers26a.html.

Related Material