FDP: A Frequency-Decomposition Preprocessing Pipeline for Unsupervised Anomaly Detection in Brain MRI

TL;DR

This work tackles unsupervised anomaly detection in brain MRI under limited annotated data by revealing that pathological lesions predominantly reside in low-frequency content and healthy low-frequency representations are consistent. It introduces FDP, a frequency-decomposition preprocessing pipeline with a Frequency Reconstruction Module and High-Frequency Supplement that reconstructs low-frequency content from a learnable prior while preserving high-frequency anatomy, enabling downstream models to suppress pathology without sacrificing diagnostic detail. FDP is designed to be plug-in and architecture-agnostic, yielding substantial improvements across multiple baselines (e.g., a 17.63% gain in DICE with LDM on BraTS20) and demonstrating robust cross-dataset generalization. The approach leverages 2D-DFT-based frequency separation, an attentive prior context bank, and L1 supervision to produce healthy-frequency reconstructions that improve anomaly localization while maintaining clinical fidelity.

Abstract

Due to the diversity of brain anatomy and the scarcity of annotated data, supervised anomaly detection for brain MRI remains challenging, driving the development of unsupervised anomaly detection (UAD) approaches. Current UAD methods typically utilize artificially generated noise perturbations on healthy MRIs to train generative models for normal anatomy reconstruction, enabling anomaly detection via residual maps. However, such simulated anomalies lack the biophysical fidelity and morphological complexity characteristic of true clinical lesions. To advance UAD in brain MRI, we conduct the first systematic frequency-domain analysis of pathological signatures, revealing two key properties: (1) anomalies exhibit unique frequency patterns distinguishable from normal anatomy, and (2) low-frequency signals maintain consistent representations across healthy scans. These insights motivate our Frequency-Decomposition Preprocessing (FDP) framework, the first UAD method to leverage frequency-domain reconstruction for simultaneous pathology suppression and anatomical preservation. FDP can integrate seamlessly with existing anomaly simulation techniques, consistently enhancing detection performance across diverse architectures while maintaining diagnostic fidelity. Experimental results demonstrate that FDP consistently improves anomaly detection performance when integrated with existing methods. Notably, FDP achieves a 17.63% increase in DICE score with LDM while maintaining robust improvements across multiple baselines. The code is available at https://github.com/ls1rius/MRI_FDP.

Figures (8)

• Figure 1: The effect of high-pass filter threshold on anomaly content in brain MRIs. The DICE coefficient between high-pass-filtered images and ground truth shows anomaly information primarily reside in low-frequency components.
• Figure 2: Training and inference pipeline of the proposed FDP method. Left: Training Phase. High-frequency signals are used for both frequency-domain reconstruction and as auxiliary input (HFSup) to enhance structural details. Right: Inference Phase. The input MRI with lesions is processed by FDP for frequency reconstruction, then fed into the generative model.
• Figure 3: The horizontal and vertical axes represent the high-pass filtering threshold $m$ and the normalized real and imaginary parts of the amplitude, respectively.
• Figure 4: Visual comparison of results with other methods. Origin means the input MRIs, and Seg denotes the ground truth.
• Figure 5: Low-Frequency Signals Quantitative Analysis.