CAESNet: Convolutional AutoEncoder based Semi-supervised Network for improving multiclass classification of endomicroscopic images.

This article presents a novel method of semisupervised learning using convolutional autoencoders for optical endomicroscopic images. Optical endomicroscopy (OE) is a newly emerged biomedical imaging modality that can support real-time clinical decisions for the grade of dysplasia. To enable real-time decision making, computer-aided diagnosis (CAD) is essential for its high speed and objectivity. However, traditional supervised CAD requires a large amount of training data. Compared with the limited number of labeled images, we can collect a larger number of unlabeled images. To utilize these unlabeled images, we have developed a Convolutional AutoEncoder based Semi-supervised Network (CAESNet) for improving the classification performance. We applied our method to an OE dataset collected from patients undergoing endoscope-based confocal laser endomicroscopy procedures for Barrett's esophagus at Emory Hospital, which consists of 429 labeled images and 2826 unlabeled images. Our CAESNet consists of an encoder with 5 convolutional layers, a decoder with 5 transposed convolutional layers, and a classification network with 2 fully connected layers and a softmax layer. In the unsupervised stage, we first update the encoder and decoder with both labeled and unlabeled images to learn an efficient feature representation. In the supervised stage, we further update the encoder and the classification network with only labeled images for multiclass classification of the OE images. Our proposed semisupervised method CAESNet achieves the best average performance for multiclass classification of OE images, which surpasses the performance of supervised methods including standard convolutional networks and convolutional autoencoder network. Our semisupervised CAESNet can efficiently utilize the unlabeled OE images, which improves the diagnosis and decision making for patients with Barrett's esophagus.

Tong L ，Wu H ，Wang MD 《-》

Computer-aided detection of early Barrett's neoplasia using volumetric laser endomicroscopy.

Volumetric laser endomicroscopy (VLE) is an advanced imaging system that provides a near-microscopic resolution scan of the esophageal wall layers up to 3-mm deep. VLE has the potential to improve detection of early neoplasia in Barrett's esophagus (BE). However, interpretation of VLE images is complex because of the large amount of data that need to be interpreted in real time. The aim of this study was to investigate the feasibility of a computer algorithm to identify early BE neoplasia on ex vivo VLE images. We used 60 VLE images from a database of high-quality ex vivo VLE-histology correlations, obtained from BE patients ± neoplasia (30 nondysplastic BE [NDBE] and 30 high-grade dysplasia/early adenocarcinoma images). VLE features from a recently developed clinical VLE prediction score for BE neoplasia served as input for the algorithm: (1) higher VLE surface than subsurface signal and (2) lack of layering. With this input, novel clinically inspired algorithm features were developed, based on signal intensity statistics and grayscale correlations. For comparison, generic image analysis methods were examined for their performance to detect neoplasia. For classification of the images in the NDBE or neoplastic group, several machine learning methods were evaluated. Leave-1-out cross-validation was used for algorithm validation. Three novel clinically inspired algorithm features were developed. The feature "layering and signal decay statistics" showed the optimal performance compared with the other clinically features ("layering" and "signal intensity distribution") and generic image analyses methods, with an area under the receiver operating characteristic curve (AUC) of .95. Corresponding sensitivity and specificity were 90% and 93%, respectively. In addition, the algorithm showed a better performance than the clinical VLE prediction score (AUC .81). This is the first study in which a computer algorithm for BE neoplasia was developed based on VLE images with direct histologic correlates. The algorithm showed good performance to detect BE neoplasia in ex vivo VLE images compared with the performance of a recently developed clinical VLE prediction score. This study suggests that an automatic detection algorithm has the potential to assist endoscopists in detecting early neoplasia on VLE. Future studies on in vivo VLE scans are needed to further validate the algorithm.

Swager AF ，van der Sommen F ，Klomp SR ，Zinger S ，Meijer SL ，Schoon EJ ，Bergman JJGHM ，de With PH ，Curvers WL ... -  《-》

Improving Multi-class Classification for Endomicroscopic Images by Semi-supervised Learning.

Wu H ，Tong L ，Wang MD 《-》

Real-time diagnosis of Barrett's esophagus: a prospective, multicenter study comparing confocal laser endomicroscopy with conventional histology for the identification of intestinal metaplasia in new users.

Richardson C ，Colavita P ，Dunst C ，Bagnato J ，Billing P ，Birkenhagen K ，Buckley F ，Buitrago W ，Burnette J ，Leggett P ，McCollister H ，Stewart K ，Wang T ，Zfass A ，Severson P ... -  《-》

Surveillance of Barrett's esophagus: why biopsy if you can endomicroscopy.

Berzosa M ，Wallace MB 《-》