Unraveling Normal Anatomy via Fluid-Driven Anomaly Randomization

TL;DR

UNA addresses the challenge of analyzing brain images across multiple modalities and pathologies by introducing a modality-agnostic framework for healthy brain anatomy reconstruction. It employs fluid-driven anomaly randomization, based on advection-diffusion dynamics, to synthesize unlimited realistic pathology profiles from few seeds, enabling training on a mix of synthetic and real data without pathology annotations. The model leverages contralateral inputs and intra-subject self-contrastive learning to learn healthy anatomy beyond gold-standard labels, achieving state-of-the-art reconstruction across CT and MR contrasts and enabling unsupervised anomaly detection without fine-tuning. By bridging healthy and diseased anatomy and cross-modality gaps, UNA enables large-scale analysis of uncurated clinical images with pathology and broadens opportunities for general-purpose medical image analysis. $P(\mathbf{x}, t)$ denotes pathology probability evolving under $\frac{\partial P}{\partial t} = - \nabla \cdot (\mathbf{V}(\mathbf{x}) P) + \nabla \cdot (D(\mathbf{x}) \nabla P)$ with $P(\mathbf{x},0)=P_0(\mathbf{x})$ and zero-Neumann boundaries, illustrating the forward anomaly-generation process integral to UNA.$

Abstract

Data-driven machine learning has made significant strides in medical image analysis. However, most existing methods are tailored to specific modalities and assume a particular resolution (often isotropic). This limits their generalizability in clinical settings, where variations in scan appearance arise from differences in sequence parameters, resolution, and orientation. Furthermore, most general-purpose models are designed for healthy subjects and suffer from performance degradation when pathology is present. We introduce UNA (Unraveling Normal Anatomy), the first modality-agnostic learning approach for normal brain anatomy reconstruction that can handle both healthy scans and cases with pathology. We propose a fluid-driven anomaly randomization method that generates an unlimited number of realistic pathology profiles on-the-fly. UNA is trained on a combination of synthetic and real data, and can be applied directly to real images with potential pathology without the need for fine-tuning. We demonstrate UNA's effectiveness in reconstructing healthy brain anatomy and showcase its direct application to anomaly detection, using both simulated and real images from 3D healthy and stroke datasets, including CT and MRI scans. By bridging the gap between healthy and diseased images, UNA enables the use of general-purpose models on diseased images, opening up new opportunities for large-scale analysis of uncurated clinical images in the presence of pathology. Code is available at https://github.com/peirong26/UNA.

Figures (5)

• Figure 1: UNA's framework overview for modality-agnostic learning of healthy anatomy, supported by fluid-driven anomaly randomization.
• Figure 2: Qualitative comparisons on healthy anatomy reconstruction, between UNA, and the state-of-the-art modality-agnostic T1w synthesis method. Testing images are generated from real healthy subjects encoded with randomly simulated pathology profiles. Pathology regions are circled in red.
• Figure 3: Qualitative comparisons on healthy anatomy reconstruction between UNA and state-of-the-art modality-agnostic synthesis models. Testing images are from real stroke datasets (ISLESHernandez2022ISLES and ATLASLiew2017ATLAS), where the stroke lesion annotations are provided, yet the ground truth healthy anatomy is unavailable. The last row shows a failure case of UNA, where it "over-corrects" the diseased anatomy. Pathology regions are circled in red.
• Figure 4: Visualizations of directly applying UNA's healthy anatomy reconstruction for anomaly detection. The estimated anomaly is computed as the absolute difference between diseased T1w MRI scans and UNA's reconstructed healthy anatomy.
• Figure 5: Ablations on UNA's healthy anatomy reconstruction.