Multi-AD: cross-domain unsupervised anomaly detection for medical and industrial applications

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, an unsupervised convolutional neural network (CNN) model for robust anomaly detection across medical and industrial domain images. Our approach utilizes the squeeze-and-excitation (SE) block to enhance feature extraction by applying channel-wise attention, enabling the model to focus on the most relevant features and detect subtle anomalies. Additionally, 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 gains the ability to detect anomalies of varying sizes. Teacher-student (T-S) architecture ensures consistency in representing high-dimensional features while adapting these features to improve 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 accuracy for anomaly localization at both the image level (81.4 % for medical and 99.6 % for industrial) and pixel level (97.0 % for medical and 98.4 % for industrial), making it effective for real-world applications.