Huaying Hao , Yitian Zhao , Shaoyi Leng , Yuanyuan Gu , Yuhui Ma , Feiming Wang , Qi Dai , Jianjun Zheng , Yue Liu , Jingfeng Zhang
{"title":"Local salient location-aware anomaly mask synthesis for pulmonary disease anomaly detection and lesion localization in CT images","authors":"Huaying Hao , Yitian Zhao , Shaoyi Leng , Yuanyuan Gu , Yuhui Ma , Feiming Wang , Qi Dai , Jianjun Zheng , Yue Liu , Jingfeng Zhang","doi":"10.1016/j.media.2025.103523","DOIUrl":null,"url":null,"abstract":"<div><div>Automated pulmonary anomaly detection using computed tomography (CT) examinations is important for the early warning of pulmonary diseases and can support clinical diagnosis and decision-making. Most training of existing pulmonary disease detection and lesion segmentation models requires expert annotations, which is time-consuming and labour-intensive, and struggles to generalize across atypical diseases. In contrast, unsupervised anomaly detection alleviates the demand for dataset annotation and is more generalizable than supervised methods in detecting rare pathologies. However, due to the large distribution differences of CT scans in a volume and the high similarity between lesion and normal tissues, existing anomaly detection methods struggle to accurately localize small lesions, leading to a low anomaly detection rate. To alleviate these challenges, we propose a local salient location-aware anomaly mask generation and reconstruction framework for pulmonary disease anomaly detection and lesion localization. The framework consists of four components: (1) a Vector Quantized Variational AutoEncoder (VQVAE)-based reconstruction network that generates a codebook storing high-dimensional features; (2) a unsupervised feature statistics based anomaly feature synthesizer to synthesize features that match the realistic anomaly distribution by filtering salient features and interacting with the codebook; (3) a transformer-based feature classification network that identifies synthetic anomaly features; (4) a residual neighbourhood aggregation feature classification loss that mitigates network overfitting by penalizing the classification loss of recoverable corrupted features. Our approach is based on two intuitions. First, generating synthetic anomalies in feature space is more effective due to the fact that lesions have different morphologies in image space and may not have much in common. Secondly, regions with salient features or high reconstruction errors in CT images tend to be similar to lesions and are more prone to synthesize abnormal features. The performance of the proposed method is validated on one public dataset with COVID-19 and one in-house dataset containing 63,610 CT images with five lung diseases. Experimental results show that compared to feature-based, synthesis-based and reconstruction-based methods, the proposed method is adaptable to CT images with four pneumonia types (COVID-19, bacteria, fungal, and mycoplasma) and one non-pneumonia (cancer) diseases and achieves state-of-the-art performance in image-level anomaly detection and lesion localization.</div></div>","PeriodicalId":18328,"journal":{"name":"Medical image analysis","volume":"102 ","pages":"Article 103523"},"PeriodicalIF":10.7000,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Medical image analysis","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1361841525000714","RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0
Abstract
Automated pulmonary anomaly detection using computed tomography (CT) examinations is important for the early warning of pulmonary diseases and can support clinical diagnosis and decision-making. Most training of existing pulmonary disease detection and lesion segmentation models requires expert annotations, which is time-consuming and labour-intensive, and struggles to generalize across atypical diseases. In contrast, unsupervised anomaly detection alleviates the demand for dataset annotation and is more generalizable than supervised methods in detecting rare pathologies. However, due to the large distribution differences of CT scans in a volume and the high similarity between lesion and normal tissues, existing anomaly detection methods struggle to accurately localize small lesions, leading to a low anomaly detection rate. To alleviate these challenges, we propose a local salient location-aware anomaly mask generation and reconstruction framework for pulmonary disease anomaly detection and lesion localization. The framework consists of four components: (1) a Vector Quantized Variational AutoEncoder (VQVAE)-based reconstruction network that generates a codebook storing high-dimensional features; (2) a unsupervised feature statistics based anomaly feature synthesizer to synthesize features that match the realistic anomaly distribution by filtering salient features and interacting with the codebook; (3) a transformer-based feature classification network that identifies synthetic anomaly features; (4) a residual neighbourhood aggregation feature classification loss that mitigates network overfitting by penalizing the classification loss of recoverable corrupted features. Our approach is based on two intuitions. First, generating synthetic anomalies in feature space is more effective due to the fact that lesions have different morphologies in image space and may not have much in common. Secondly, regions with salient features or high reconstruction errors in CT images tend to be similar to lesions and are more prone to synthesize abnormal features. The performance of the proposed method is validated on one public dataset with COVID-19 and one in-house dataset containing 63,610 CT images with five lung diseases. Experimental results show that compared to feature-based, synthesis-based and reconstruction-based methods, the proposed method is adaptable to CT images with four pneumonia types (COVID-19, bacteria, fungal, and mycoplasma) and one non-pneumonia (cancer) diseases and achieves state-of-the-art performance in image-level anomaly detection and lesion localization.
期刊介绍:
Medical Image Analysis serves as a platform for sharing new research findings in the realm of medical and biological image analysis, with a focus on applications of computer vision, virtual reality, and robotics to biomedical imaging challenges. The journal prioritizes the publication of high-quality, original papers contributing to the fundamental science of processing, analyzing, and utilizing medical and biological images. It welcomes approaches utilizing biomedical image datasets across all spatial scales, from molecular/cellular imaging to tissue/organ imaging.