The implementation of CycleGAN-assisted image translation in deep UV-excited fluorescence microscopy improves the accuracy of lymph node metastasis detection, facilitating intraoperative diagnosis.

新岡 宏彦

doi:10.11239/jsmbe.annual62.125_1

<p>This study addresses the necessity for improved intraoperative diagnostic systems in surgery. The prevalent frozen section procedure is hindered by poor quality and time consumption, leading to exploration of alternatives. Microscopy with ultraviolet surface excitation (MUSE) stands out as a rapid and cost-effective imaging technique. However, labeling MUSE images of unfixed specimens poses challenges for pathologists, hindering supervised learning for AI. To overcome this, a deep-learning pipeline for lymph node metastasis detection is proposed. CycleGAN translate MUSE images of unfixed lymph nodes to formalin‐fixed paraffin‐embedded (FFPE) sample, and diagnostic prediction is performed using deep convolutional neural network (CNN) trained on previous FFPE samples stored in hospital. The pipeline achieves an 84.6% average accuracy, an 18.3% improvement over the CNN-only model. The CycleGAN-driven modality translation can be applied to various intraoperative diagnostic imaging systems, addressing the difficulty in labeling new modality images.</p>

The implementation of CycleGAN-assisted image translation in deep UV-excited fluorescence microscopy improves the accuracy of lymph node metastasis detection, facilitating intraoperative diagnosis.

説明

収録刊行物

詳細情報詳細情報について

書き出し

問題の指摘

The implementation of CycleGAN-assisted image translation in deep UV-excited fluorescence microscopy improves the accuracy of lymph node metastasis detection, facilitating intraoperative diagnosis.

説明

収録刊行物

詳細情報 詳細情報について

書き出し

問題の指摘

詳細情報詳細情報について