HemExp: Clinically-Guided Latent Diffusion for Modeling Hematoma Expansion

2026-06-13 · Source: Computer Vision and Pattern Recognition · Field: Science & Research — Health & Medical Research, Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

HemExp is a clinically-guided latent diffusion model designed to predict hematoma expansion (HE) after spontaneous intracerebral hemorrhage (ICH), a critical factor in neurosurgical triage. Unlike binary risk predictors, HemExp generates patient-specific follow-up non-contrast CT images and segmentations for intraparenchymal and intraventricular hemorrhage. The model is conditioned on baseline imaging, clinical variables, and an explicit expansion indicator, enabling controllable simulation of clinical scenarios. It employs a hemorrhage-aware multi-head variational autoencoder and models progression as the difference between baseline and follow-up latent representations using a conditional diffusion model. Trained on 450 patients and evaluated on 107 from a held-out institution, HemExp produces spatial HE probability maps by generating multiple synthetic follow-up images to estimate plausible hematoma volume distributions. Perturbing clinical inputs like symptom-onset-to-imaging time or anticoagulant status shifts these predicted distributions, demonstrating robust estimation of outcomes such as hematoma volume and intraventricular involvement.

Key takeaway

For neurosurgical teams making acute triage and treatment decisions for spontaneous intracerebral hemorrhage patients, HemExp offers a more nuanced, uncertainty-aware approach than traditional binary predictors. You can simulate various clinical scenarios by perturbing inputs like symptom-onset-to-imaging time or anticoagulant status, gaining insight into potential hematoma expansion and mass effects. This supports more informed, patient-specific care planning by providing distributions of plausible follow-up hematoma volumes.

Key insights

HemExp uses clinically-guided latent diffusion to generate patient-specific CT images, predicting hematoma expansion and providing uncertainty-aware outcomes.

Principles

• Model progression as latent representation differences.
• Condition generation on clinical variables for control.
• Estimate outcome distributions via multiple synthetic images.

Method

HemExp uses a hemorrhage-aware multi-head VAE and a conditional diffusion model to generate follow-up CT images, modeling progression as the difference between baseline and follow-up latent representations.

In practice

• Simulate ICH progression under varying clinical inputs.
• Generate spatial HE probability maps for patients.
• Estimate distributions of follow-up hematoma volumes.

Topics

• Latent Diffusion Models
• Hematoma Expansion
• Intracerebral Hemorrhage
• Medical Image Generation
• Neurosurgical Decision Support
• Uncertainty Quantification

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• Computational Imaging Priors for Wireless Capsule Endoscopy: Monte Carlo-Guided Hemoglobin Mapping for Rare-Anomaly Detection — A Monte Carlo-inspired prior improves wireless capsule endoscopy classification by better distinguishing hemoglobin from confounding factors.
• Clinically Interpretable Sepsis Early Warning via LLM-Guided Simulation of Temporal Physiological Dynamics — LLM-guided temporal simulation improves sepsis early warning by providing interpretable physiological trajectory predictions.
• Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression — A treatment-conditioned diffusion framework accurately forecasts neurodegenerative disease progression by integrating imaging and pharmacological data, preserving anatomical detail.
• I Gave a Model Eyes and a Genome. Here’s What It Learned About Brain Cancer. — Cross-attention fusion of pathology and genomics significantly improves glioblastoma survival prediction over unimodal approaches.
• Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images — Contrastive learning, especially with expert-guided hard negatives, improves fine-grained pediatric brain tumor classification from WSIs.
• A Patient-Specific Pulmonary Arterial Tree Digital Twin to Extract Pulmonary Embolism Biomarkers — Automated digital twins of pulmonary arteries can rapidly quantify PE severity and clot distribution.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.