[論文レビュー] Multi-AD: Cross-Domain Unsupervised Anomaly Detection for Medical and Industrial Applications

Wahyu Rahmaniar, Kenji Suzuki|arXiv (Cornell University)|Feb 5, 2026

Anomaly Detection Techniques and Applications被引用数 0

ひとこと要約

tldr: Multi-AD は SE アテンション、知識蒸留、そして教師-生徒フレームワーク内の識別器を用いた CNN ベースの教師なし異常検出器で、医療画像および産業画像の異常を検出し、複数データセットで最先端の AUROC を達成します。

ABSTRACT

Traditional deep learning models often lack annotated data, especially in cross-domain applications such as anomaly detection, which is critical for early disease diagnosis in medicine and defect detection in industry. To address this challenge, we propose Multi-AD, a convolutional neural network (CNN) model for robust unsupervised anomaly detection across medical and industrial images. Our approach employs the squeeze-and-excitation (SE) block to enhance feature extraction via channel-wise attention, enabling the model to focus on the most relevant features and detect subtle anomalies. Knowledge distillation (KD) transfers informative features from the teacher to the student model, enabling effective learning of the differences between normal and anomalous data. Then, the discriminator network further enhances the model's capacity to distinguish between normal and anomalous data. At the inference stage, by integrating multi-scale features, the student model can detect anomalies of varying sizes. The teacher-student (T-S) architecture ensures consistent representation of high-dimensional features while adapting them to enhance anomaly detection. Multi-AD was evaluated on several medical datasets, including brain MRI, liver CT, and retina OCT, as well as industrial datasets, such as MVTec AD, demonstrating strong generalization across multiple domains. Experimental results demonstrated that our approach consistently outperformed state-of-the-art models, achieving the best average AUROC for both image-level (81.4% for medical and 99.6% for industrial) and pixel-level (97.0% for medical and 98.4% for industrial) tasks, making it effective for real-world applications.

研究の動機と目的

• Address the lack of annotated data in cross-domain anomaly detection for medical and industrial images.
• Develop a robust unsupervised CNN model (Multi-AD) that leverages channel-wise attention, knowledge distillation, and a discriminator.
• Enable robust anomaly detection across varying anomaly sizes via multi-scale feature integration.
• Demonstrate strong generalization across diverse medical and industrial datasets.
• Achieve best average image-level and pixel-level AUROC on evaluated datasets.

提案手法

• Incorporate squeeze-and-excitation (SE) blocks to enhance feature extraction via channel-wise attention.
• Apply knowledge distillation to transfer informative features from a teacher to a student model.
• Use a discriminator network to strengthen separation between normal and anomalous data.
• Integrate multi-scale features at inference to detect anomalies of different sizes.
• Adopt a teacher–student (T–S) architecture to maintain high-dimensional representation while adapting for anomaly detection.

実験結果

リサーチクエスチョン

• RQ1 Can Multi-AD achieve robust unsupervised anomaly detection across medical and industrial domains?
• RQ2 What is the contribution of SE attention, knowledge distillation, and the discriminator to performance?
• RQ3 How does multi-scale feature integration affect detection of anomalies with varying sizes?
• RQ4 How does the method perform on brain MRI, liver CT, retina OCT, and MVTec AD in image-level and pixel-level AUROC?

主な発見

• Achieves best average image-level AUROC: 81.4% for medical and 99.6% for industrial data.
• Achieves best average pixel-level AUROC: 97.0% for medical and 98.4% for industrial data.
• Demonstrates strong generalization across brain MRI, liver CT, retina OCT, and MVTec AD datasets.
• Outperforms state-of-the-art anomaly detection models in the reported experiments.