Establishing Stochastic Object Models from Noisy Data via Ambient Measurement-Integrated Diffusion

2025-12-16

AI Summary

Existing stochastic object models (SOMs) for medical image quality assessment struggle to simultaneously ensure anatomical fidelity and robustness to measurement noise. Method: We propose an unsupervised SOM construction framework tailored to noisy clinical imaging data (e.g., CT, mammography), introducing a novel “measurement-integrated diffusion” paradigm that explicitly decouples measurement noise from diffusion process noise. Our approach incorporates measurement consistency constraints and an environment-aware loss function, enabling learning of high-fidelity, task-oriented clean SOMs directly from noisy observations—without requiring ground-truth clean data. Contribution/Results: Validated on real clinical datasets, our models achieve significantly improved anatomical structure fidelity and yield more reliable image quality assessments. This work overcomes two key limitations of prior methods: the anatomical rigidity of traditional mathematical models and the reliance of data-driven approaches on clean reference labels. It establishes a new paradigm for unsupervised clinical image modeling.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Unsupervised & Self-Supervised Learning
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Task-based measures of image quality (IQ) are critical for evaluating medical imaging systems, which must account for randomness including anatomical variability. Stochastic object models (SOMs) provide a statistical description of such variability, but conventional mathematical SOMs fail to capture realistic anatomy, while data-driven approaches typically require clean data rarely available in clinical tasks. To address this challenge, we propose AMID, an unsupervised Ambient Measurement-Integrated Diffusion with noise decoupling, which establishes clean SOMs directly from noisy measurements. AMID introduces a measurement-integrated strategy aligning measurement noise with the diffusion trajectory, and explicitly models coupling between measurement and diffusion noise across steps, an ambient loss is thus designed base on it to learn clean SOMs. Experiments on real CT and mammography datasets show that AMID outperforms existing methods in generation fidelity and yields more reliable task-based IQ evaluation, demonstrating its potential for unsupervised medical imaging analysis.

Problem

• Develops a method to create clean stochastic object models from noisy medical imaging data.
• Proposes an unsupervised diffusion technique that decouples measurement noise from anatomical variability.
• Enhances task-based image quality evaluation by generating more realistic anatomical models.

Innovation

• Unsupervised diffusion with noise decoupling
• Measurement-integrated strategy aligning noise trajectory
• Ambient loss for clean stochastic object models