{"title":"An explainable unsupervised learning approach for anomaly detection on corneal <i>in vivo</i> confocal microscopy images.","authors":"Ningning Tang, Qi Chen, Yunyu Meng, Daizai Lei, Li Jiang, Yikun Qin, Xiaojia Huang, Fen Tang, Shanshan Huang, Qianqian Lan, Qi Chen, Lijie Huang, Rushi Lan, Xipeng Pan, Huadeng Wang, Fan Xu, Wenjing He","doi":"10.3389/fbioe.2025.1576513","DOIUrl":null,"url":null,"abstract":"<p><strong>Background: </strong><i>In vivo</i> confocal microscopy (IVCM) is a crucial imaging modality for assessing corneal diseases, yet distinguishing pathological features from normal variations remains challenging due to the complex multi-layered corneal structure. Existing anomaly detection methods often struggle to generalize across diverse disease manifestations. To address these limitations, we propose a Transformer-based unsupervised anomaly detection method for IVCM images, capable of identifying corneal abnormalities without prior knowledge of specific disease features.</p><p><strong>Methods: </strong>Our method consists of three submodules: an EfficientNet network, a Multi-Scale Feature Fusion Network, and a Transformer Network. A total of 7,063 IVCM images (95 eyes) were included for analysis. The model was trained exclusively on normal IVCM images to capture and differentiate structural variations across four distinct corneal layers: epithelium, sub-basal nerve plexus, stroma, and endothelium. During inference, anomaly scores were computed to distinguish pathological from normal images. The model's performance was evaluated on both internal and external datasets, and comparative analyses were conducted against existing anomaly detection methods, including generative adversarial networks (AnoGAN), generate to detect anomaly model (G2D), and discriminatively trained reconstruction anomaly embedding model (DRAEM). Additionally, explainable anomaly maps were generated to enhance the interpretability of model decisions.</p><p><strong>Results: </strong>The proposed method achieved an the areas under the receiver operating characteristic curve of 0.933 on internal validation and 0.917 on an external test dataset, outperforming AnoGAN, G2D, and DRAEM in both accuracy and generalizability. The model effectively distinguished normal and pathological images, demonstrating statistically significant differences in anomaly scores (p < 0.001). Furthermore, visualization results indicated that the detected anomalous regions corresponded to morphological deviations, highlighting potential imaging biomarkers for corneal diseases.</p><p><strong>Conclusion: </strong>This study presents an efficient and interpretable unsupervised anomaly detection model for IVCM images, effectively identifying corneal abnormalities without requiring labeled pathological samples. The proposed method enhances screening efficiency, reduces annotation costs, and holds great potential for scalable intelligent diagnosis of corneal diseases.</p>","PeriodicalId":12444,"journal":{"name":"Frontiers in Bioengineering and Biotechnology","volume":"13 ","pages":"1576513"},"PeriodicalIF":4.3000,"publicationDate":"2025-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12179219/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Frontiers in Bioengineering and Biotechnology","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.3389/fbioe.2025.1576513","RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2025/1/1 0:00:00","PubModel":"eCollection","JCR":"Q1","JCRName":"BIOTECHNOLOGY & APPLIED MICROBIOLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
Background: In vivo confocal microscopy (IVCM) is a crucial imaging modality for assessing corneal diseases, yet distinguishing pathological features from normal variations remains challenging due to the complex multi-layered corneal structure. Existing anomaly detection methods often struggle to generalize across diverse disease manifestations. To address these limitations, we propose a Transformer-based unsupervised anomaly detection method for IVCM images, capable of identifying corneal abnormalities without prior knowledge of specific disease features.
Methods: Our method consists of three submodules: an EfficientNet network, a Multi-Scale Feature Fusion Network, and a Transformer Network. A total of 7,063 IVCM images (95 eyes) were included for analysis. The model was trained exclusively on normal IVCM images to capture and differentiate structural variations across four distinct corneal layers: epithelium, sub-basal nerve plexus, stroma, and endothelium. During inference, anomaly scores were computed to distinguish pathological from normal images. The model's performance was evaluated on both internal and external datasets, and comparative analyses were conducted against existing anomaly detection methods, including generative adversarial networks (AnoGAN), generate to detect anomaly model (G2D), and discriminatively trained reconstruction anomaly embedding model (DRAEM). Additionally, explainable anomaly maps were generated to enhance the interpretability of model decisions.
Results: The proposed method achieved an the areas under the receiver operating characteristic curve of 0.933 on internal validation and 0.917 on an external test dataset, outperforming AnoGAN, G2D, and DRAEM in both accuracy and generalizability. The model effectively distinguished normal and pathological images, demonstrating statistically significant differences in anomaly scores (p < 0.001). Furthermore, visualization results indicated that the detected anomalous regions corresponded to morphological deviations, highlighting potential imaging biomarkers for corneal diseases.
Conclusion: This study presents an efficient and interpretable unsupervised anomaly detection model for IVCM images, effectively identifying corneal abnormalities without requiring labeled pathological samples. The proposed method enhances screening efficiency, reduces annotation costs, and holds great potential for scalable intelligent diagnosis of corneal diseases.
期刊介绍:
The translation of new discoveries in medicine to clinical routine has never been easy. During the second half of the last century, thanks to the progress in chemistry, biochemistry and pharmacology, we have seen the development and the application of a large number of drugs and devices aimed at the treatment of symptoms, blocking unwanted pathways and, in the case of infectious diseases, fighting the micro-organisms responsible. However, we are facing, today, a dramatic change in the therapeutic approach to pathologies and diseases. Indeed, the challenge of the present and the next decade is to fully restore the physiological status of the diseased organism and to completely regenerate tissue and organs when they are so seriously affected that treatments cannot be limited to the repression of symptoms or to the repair of damage. This is being made possible thanks to the major developments made in basic cell and molecular biology, including stem cell science, growth factor delivery, gene isolation and transfection, the advances in bioengineering and nanotechnology, including development of new biomaterials, biofabrication technologies and use of bioreactors, and the big improvements in diagnostic tools and imaging of cells, tissues and organs.
In today`s world, an enhancement of communication between multidisciplinary experts, together with the promotion of joint projects and close collaborations among scientists, engineers, industry people, regulatory agencies and physicians are absolute requirements for the success of any attempt to develop and clinically apply a new biological therapy or an innovative device involving the collective use of biomaterials, cells and/or bioactive molecules. “Frontiers in Bioengineering and Biotechnology” aspires to be a forum for all people involved in the process by bridging the gap too often existing between a discovery in the basic sciences and its clinical application.