Wenhua Su, Dachao Zheng, Jiacheng Zhou, Qiushu Chen, Liwen Chen, Yuwei Yang, Yiyan Fei, Haijun Yao, Jiong Ma, Lan Mi
{"title":"利用荧光寿命检测和机器学习对多灶性皮肤肿瘤边缘进行快速、精确的评估","authors":"Wenhua Su, Dachao Zheng, Jiacheng Zhou, Qiushu Chen, Liwen Chen, Yuwei Yang, Yiyan Fei, Haijun Yao, Jiong Ma, Lan Mi","doi":"10.1063/5.0224181","DOIUrl":null,"url":null,"abstract":"The precise determination of surgical margins is essential for the management of multifocal cutaneous cancers, including extramammary Paget’s disease. This study introduces a novel strategy for precise margin identification in such tumors, employing multichannel autofluorescence lifetime decay (MALD), fluorescence lifetime imaging microscopy (FLIM), and machine learning, including confidence learning algorithms. Using FLIM, 51 unstained frozen sections were analyzed, of which 13 (25%) sections, containing 5003 FLIM patches, were used for training the residual network model (ResNet–FLIM). The remaining 38 (75%) sections, including 16 918 patches, were retained for external validation. Application of confidence learning with deep learning reduced the reliance on extensive pathologist annotation. Refined labels obtained by ResNet–FLIM were then incorporated into a support vector machine (SVM) model, which utilized fiber-optic-based MALD data. Both models exhibited substantial agreement with the pathological assessments. Of the 35 MALD-measured tissue segments, six (17%) segments were selected as the training dataset, including 900 decay profiles. The remaining 29 segments (83%), including 2406 decay profiles, were reserved for external validation. The ResNet–FLIM model achieved 100% sensitivity and specificity. The SVM–MALD model demonstrated 94% sensitivity and 83% specificity. Notably, fiber-optic-MALD allows assessing 12 sites per patient and delivering predictions within 10 min. Variations in the necessary safe margin length were observed among patients, highlighting the necessity for patient-specific approaches to determine surgical margins. This innovative approach holds potential for wide clinical application, providing a rapid and accurate margin evaluation method that significantly reduces a pathologist’s workload and improves patient outcomes through personalized medicine.","PeriodicalId":8198,"journal":{"name":"APL Photonics","volume":"72 1","pages":""},"PeriodicalIF":5.4000,"publicationDate":"2024-09-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Rapid and precise multifocal cutaneous tumor margin assessment using fluorescence lifetime detection and machine learning\",\"authors\":\"Wenhua Su, Dachao Zheng, Jiacheng Zhou, Qiushu Chen, Liwen Chen, Yuwei Yang, Yiyan Fei, Haijun Yao, Jiong Ma, Lan Mi\",\"doi\":\"10.1063/5.0224181\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"The precise determination of surgical margins is essential for the management of multifocal cutaneous cancers, including extramammary Paget’s disease. This study introduces a novel strategy for precise margin identification in such tumors, employing multichannel autofluorescence lifetime decay (MALD), fluorescence lifetime imaging microscopy (FLIM), and machine learning, including confidence learning algorithms. Using FLIM, 51 unstained frozen sections were analyzed, of which 13 (25%) sections, containing 5003 FLIM patches, were used for training the residual network model (ResNet–FLIM). The remaining 38 (75%) sections, including 16 918 patches, were retained for external validation. Application of confidence learning with deep learning reduced the reliance on extensive pathologist annotation. Refined labels obtained by ResNet–FLIM were then incorporated into a support vector machine (SVM) model, which utilized fiber-optic-based MALD data. Both models exhibited substantial agreement with the pathological assessments. Of the 35 MALD-measured tissue segments, six (17%) segments were selected as the training dataset, including 900 decay profiles. The remaining 29 segments (83%), including 2406 decay profiles, were reserved for external validation. The ResNet–FLIM model achieved 100% sensitivity and specificity. The SVM–MALD model demonstrated 94% sensitivity and 83% specificity. Notably, fiber-optic-MALD allows assessing 12 sites per patient and delivering predictions within 10 min. Variations in the necessary safe margin length were observed among patients, highlighting the necessity for patient-specific approaches to determine surgical margins. This innovative approach holds potential for wide clinical application, providing a rapid and accurate margin evaluation method that significantly reduces a pathologist’s workload and improves patient outcomes through personalized medicine.\",\"PeriodicalId\":8198,\"journal\":{\"name\":\"APL Photonics\",\"volume\":\"72 1\",\"pages\":\"\"},\"PeriodicalIF\":5.4000,\"publicationDate\":\"2024-09-05\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"APL Photonics\",\"FirstCategoryId\":\"101\",\"ListUrlMain\":\"https://doi.org/10.1063/5.0224181\",\"RegionNum\":1,\"RegionCategory\":\"物理与天体物理\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"OPTICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"APL Photonics","FirstCategoryId":"101","ListUrlMain":"https://doi.org/10.1063/5.0224181","RegionNum":1,"RegionCategory":"物理与天体物理","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"OPTICS","Score":null,"Total":0}
Rapid and precise multifocal cutaneous tumor margin assessment using fluorescence lifetime detection and machine learning
The precise determination of surgical margins is essential for the management of multifocal cutaneous cancers, including extramammary Paget’s disease. This study introduces a novel strategy for precise margin identification in such tumors, employing multichannel autofluorescence lifetime decay (MALD), fluorescence lifetime imaging microscopy (FLIM), and machine learning, including confidence learning algorithms. Using FLIM, 51 unstained frozen sections were analyzed, of which 13 (25%) sections, containing 5003 FLIM patches, were used for training the residual network model (ResNet–FLIM). The remaining 38 (75%) sections, including 16 918 patches, were retained for external validation. Application of confidence learning with deep learning reduced the reliance on extensive pathologist annotation. Refined labels obtained by ResNet–FLIM were then incorporated into a support vector machine (SVM) model, which utilized fiber-optic-based MALD data. Both models exhibited substantial agreement with the pathological assessments. Of the 35 MALD-measured tissue segments, six (17%) segments were selected as the training dataset, including 900 decay profiles. The remaining 29 segments (83%), including 2406 decay profiles, were reserved for external validation. The ResNet–FLIM model achieved 100% sensitivity and specificity. The SVM–MALD model demonstrated 94% sensitivity and 83% specificity. Notably, fiber-optic-MALD allows assessing 12 sites per patient and delivering predictions within 10 min. Variations in the necessary safe margin length were observed among patients, highlighting the necessity for patient-specific approaches to determine surgical margins. This innovative approach holds potential for wide clinical application, providing a rapid and accurate margin evaluation method that significantly reduces a pathologist’s workload and improves patient outcomes through personalized medicine.
APL PhotonicsPhysics and Astronomy-Atomic and Molecular Physics, and Optics
CiteScore
10.30
自引率
3.60%
发文量
107
审稿时长
19 weeks
期刊介绍:
APL Photonics is the new dedicated home for open access multidisciplinary research from and for the photonics community. The journal publishes fundamental and applied results that significantly advance the knowledge in photonics across physics, chemistry, biology and materials science.