Background We developed a computer-assisted diagnosis model to evaluate the feasibility of automated classification of intrapapillary capillary loops (IPCLs) to improve the detection of esophageal squamous cell carcinoma (ESCC). Methods We recruited patients who underwent magnifying endoscopy with narrow-band imaging for evaluation of a suspicious esophageal condition. Case images were evaluated to establish a gold standard IPCL classification according to the endoscopic diagnosis and histological findings. A double-labeling fully convolutional network (FCN) was developed for image segmentation. Diagnostic performance of the model was compared with that of endoscopists grouped according to years of experience (senior > 15 years; mid level 10 – 15 years; junior 5 – 10 years). Results Of the 1383 lesions in the study, the mean accuracies of IPCL classification were 92.0 %, 82.0 %, and 73.3 %, for the senior, mid level, and junior groups, respectively. The mean diagnostic accuracy of the model was 89.2 % and 93.0 % at the lesion and pixel levels, respectively. The interobserver agreement between the model and the gold standard was substantial (kappa value, 0.719). The accuracy of the model for inflammatory lesions (92.5 %) was superior to that of the mid level (88.1 %) and junior (86.3 %) groups (P < 0.001). For malignant lesions, the accuracy of the model (B1, 87.6 %; B2, 93.9 %) was significantly higher than that of the mid level (B1, 79.1 %; B2, 90.0 %) and junior (B1, 69.2 %; B2, 79.3 %) groups (P < 0.001). Conclusions Double-labeling FCN automated IPCL recognition was feasible and could facilitate early detection of ESCC.

This paper focuses on the problem of feature extraction and the classification of microvascular morphological types to aid esophageal cancer detection. We present a patch-based system with a hybrid SVM model with data augmentation for intraepithelial papillary capillary loop recognition. A greedy patch-generating algorithm and a specialized CNN named NBI-Net are designed to extract hierarchical features from patches. We investigate a series of data augmentation techniques to progressively improve the prediction invariance of image scaling and rotation. For classifier boosting, SVM is used as an alternative to softmax to enhance generalization ability. The effectiveness of CNN feature representation ability is discussed for a set of widely used CNN models, including AlexNet, VGG-16, and GoogLeNet. Experiments are conducted on the NBI-ME dataset. The recognition rate is up to 92.74% on the patch level with data augmentation and classifier boosting. The results show that the combined CNN-SVM model beats models of traditional features with SVM as well as the original CNN with softmax. The synthesis results indicate that our system is able to assist clinical diagnosis to a certain extent.

No abstract