Abstract
Rapid differentiation between intracerebral hemorrhage (ICH) and ischemic stroke is important in hyper-acute care, yet non-contrast CT (NCCT), the diagnostic gold standard, is unavailable during prehospital transport. Recent trials demonstrate that ultra-early interventions (blood pressure lowering and hemostatic therapy) improve ICH outcomes when initiated within minutes, creating an urgent need for prehospital stroke differentiation. We propose a cross-modal framework that uses paired NCCT–ultrasound data during training to improve ultrasound-only ICH screening at deployment. We collected 100 transtemporal transcranial B-mode ultrasound examinations from acute stroke patients, including 91 with paired NCCT, using a standardized 5-second sweep protocol. Our model combines an ultrasound pathway built on a pretrained ultrasound foundation model (USFM) with attention-based multiple-instance learning (MIL), a 3D ResNet NCCT encoder pretrained on the RSNA ICH dataset, and cross-modal objectives: contrastive feature alignment and CT-guided knowledge distillation. At inference, only ultrasound is required; multi-swipe MAX-pool aggregation yields a single patient-level prediction. Evaluated with patient-disjoint 5-fold cross-validation, the framework achieves AUROC 0.92$\pm$0.07, sensitivity 0.90$\pm$0.13, and specificity 0.86$\pm$0.14 at the Youden-optimal operating point, outperforming CT-only (AUROC 0.86$\pm$0.07) and ultrasound-only (AUROC 0.85$\pm$0.14) baselines. Ablation studies show that the two cross-modal losses provide complementary gains, with distillation acting as both a knowledge transfer mechanism and an implicit regularizer. Subgroup analyses show no significant performance differences across operators of varying expertise ($p{=}0.59$) or acoustic window quality ($p{=}0.92$), supporting the robustness and deployment feasibility of ultrasound-based ICH screening in the prehospital setting.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SWITCH_016.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=fdXpXszFFy
BibTex
@InProceedings{GonJes_Contrastive_MICCAISAT2026,
author = { Gonzalez Riveros, Jesus David AND Simonetti, Renato AND Canals, Pere AND Ribo, Marc AND Rodriguez-Luna, David},
title = { { Contrastive Alignment and CT-Guided Distillation for Ultrasound-Only Intracerebral Hemorrhage Screening } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}
