Bacteria Detection in Optical Endomicroscopy Images using Synthetic Images DOI
Mehmet Demirel, Bethany Mills, Erin Gaughan

et al.

Published: July 15, 2024

Pneumonia, a lung-related illness often resulting from bacterial infection, requires quick and accurate identification, especially in intensive care situations. Optical endomicroscopy (OEM) offers solution by providing real-time acquisition of vivo situ optical biopsies, enhancing the speed identification. However, sheer volume images produced OEM for analysis poses significant challenge, potentially delaying critical treatments. Prior approaches to bacteria detection imagery have relied on unsupervised models. These models are hindered either need manual threshold setting or high computational demands, making unfeasible. To address these challenges, supervised learning methods can be considered, as they shown superior performance efficiency various medical applications. heavily depend availability vast quantities accurately labeled data, which is scarce images. this end, paper we introduce novel approach generate synthetic within image sequences, enabling use deep techniques. We developed two simulate movement embedded into real, bacteria-free background assess efficacy employed 3D U-Net training. The results revealed that U-Net, when trained exhibited 3.86% enhancement correlation with real annotations over state-of-the-art

Language: Английский

EmiNet: Annotation Free Moving Bacteria Detection on Optical Endomicroscopy Images DOI Creative Commons
Mehmet Demirel, Bethany Mills, Erin Gaughan

et al.

Published: March 14, 2024

Pneumonia, a respiratory disease often caused by bacterial infection in the distal lung, necessitates prompt and precise diagnosis, particularly critical care settings. Optical endomicroscopy (OEM) facilitates realtime acquisition of vivo situ optical biopsies, thus expediting detection. Nonetheless, visually analysing vast number images generated OEM real time can be challenging, potentially impeding timely intervention. In this regard, to rapidly segment detect bacteria, we propose EmiNet, novel dual-stream network that integrates capabilities Transformer Convolutional Neural Networks (CNN) within an encoder-decoder architecture simultaneously captures local-global appearance motion features. Within introduce multimodal cross-channel attention module integration features with Furthermore, compensate for lack annotated training data, developed synthetic dataset simulating integrating these models onto backgrounds devoid bacteria. The authenticity was confirmed through Visual Turing Test, where medical experts assessed mixture images. results indicate are almost indistinguishable from ones. EmiNet's performance is evaluated on both datasets. Experiments show EmiNet surpasses state-of-the-art segmentation leads 6.8% improvement detection correlation over bacteria algorithms.

Language: Английский

Citations

0
Bacteria Detection in Optical Endomicroscopy Images using Synthetic Images DOI
Mehmet Demirel, Bethany Mills, Erin Gaughan

et al.

Published: July 15, 2024

Pneumonia, a lung-related illness often resulting from bacterial infection, requires quick and accurate identification, especially in intensive care situations. Optical endomicroscopy (OEM) offers solution by providing real-time acquisition of vivo situ optical biopsies, enhancing the speed identification. However, sheer volume images produced OEM for analysis poses significant challenge, potentially delaying critical treatments. Prior approaches to bacteria detection imagery have relied on unsupervised models. These models are hindered either need manual threshold setting or high computational demands, making unfeasible. To address these challenges, supervised learning methods can be considered, as they shown superior performance efficiency various medical applications. heavily depend availability vast quantities accurately labeled data, which is scarce images. this end, paper we introduce novel approach generate synthetic within image sequences, enabling use deep techniques. We developed two simulate movement embedded into real, bacteria-free background assess efficacy employed 3D U-Net training. The results revealed that U-Net, when trained exhibited 3.86% enhancement correlation with real annotations over state-of-the-art

Language: Английский

Citations

0