Key result
Unsupervised anomaly detection algorithm accurately classifies ~94% of healthy and interstitial lung disease cases.
An unsupervised anomaly detection model using a U-Net-based autoencoder and temporal noise distinction algorithm successfully differentiates between persistent lung disease anomalies and transient environmental noise in wearable device recordings.
Answers are grounded in this paper and cite their sources.
Should not yet change clinical lung sound assessment; leaves open unsupervised anomaly detection for ILD screening in larger prospective studies.
No abstract yet. The journal hasn't published one we can show.
No takes yet. Share an insight, caveat, or question.
Hirasawa et al. (2026) studied Interstitial lung disease (ILD) (n=36). Unsupervised anomaly detection algorithm (U-Net-based autoencoder with MSE-SSIM loss and ADR metric) was evaluated on Binary classification accuracy based on Anomaly Detection Rate (ADR) threshold of 50%. The proposed unsupervised anomaly detection algorithm correctly classified 29 of 30 healthy cases as normal and 5 of 6 interstitial lung disease cases as abnormal.
Synapse has enriched 10 related papers. Consider them for comparative context: