Reproduce “FDP: A Frequency-Decomposition Preprocessing Pipeline for Unsupervised Anomaly Detection in Brain MRI” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hao Li, Zhenfeng Zhuang, Jingyu Lin, Yu Liu, Yifei Chen, Qiong Peng, Lequan Yu, Liansheng Wang

The FDP (Frequency-Decomposition Preprocessing) pipeline improves unsupervised anomaly detection in brain MRI by systematically analyzing and leveraging frequency-domain characteristics. It achieved a 17.63% increase in DICE score and a 20.92% improvement in AUPRC when integrated with a Latent Diffusion Model on the BraTS20 dataset, demonstrating enhanced accuracy and broad compatibility with existing UAD methods.

Key insights

  • Pathological anomalies in brain MRI primarily manifest in distinct low-frequency components, contrary to common image anomalies.
  • Low-frequency signals in healthy brain scans exhibit remarkably consistent representations, providing a reliable basis for reconstruction.
  • Pathological alterations disrupt the coherence and increase dispersion of these otherwise consistent healthy low-frequency patterns.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.