Docbot (United States)

Introduction: Video capsule endoscopy (VCE) is a standard noninvasive technology for diagnostic imaging of the small bowel. However, the reading process is time consuming, and an important lesion present in a single frame can easily be missed. The development of computer-assisted algorithms for VCE may aid in improving diagnostic yield and reading efficiency. We therefore developed convolutional neural networks (CNNs) for identifying anatomic landmarks and abnormalities in VCE.1177_B Figure 2. ROC Curves of CNNs for Identification of Landmarks and Abnormalities in VCE ImagesMethods: Utilizing a database of 439 de-identified capsule videos (PillCam SB3) from December 2016 to April 2018, we extracted a total of 57,191 frames. All image frames were annotated by anatomic location (1,412 esophagus, 17,555 stomach, 36,315 small bowel and 1,909 colon), and by the presence (22,627 frames) or absence (34,564 frames) of abnormalities: AVMs, erythema, varices, bulges, masses, ulcerations, erosions, blood, red villi, diverticula, polyps, and xanthomas. We developed 2 CNNs using Tensorflow with Mobilenet V2 to detect normal vs abnormal (binary) and location (multi-class). The models were pre-trained on ImageNet and then fine-tuned on capsule images (80% for training 20% for validation), using augmentation methods to prevent overfitting and improve generalizability. We used an Adam optimizer to output a probability between 0 and 1 for binary classification and softmax activation to output probabilities of multi-class. Results: Extremely high accuracies (97% for abnormalities and 98% for location) were achieved for both CNNs (Table 1). Receiver Operating Characteristic (ROC) Curves are shown in Figure 1. Reproducibility was perfect (Kappa=1.0). These CNNs can each process 500 frames per second on inference and our system can process a typical 8-12 hour capsule study in 10-15 minutes.1177_A Figure 1. Performance of CNNs for Identification of Landmarks and Abnormalities in VCE ImagesConclusion: Our CNNs have a high sensitivity and specificity for detecting anatomic location and abnormalities, and operate at a speed comparable to the time required to download from the recorder. This new technology has the potential to significantly reduce reading times and improve read accuracy. Further studies are needed to validate the algorithm's performance as a real-time aid for capsule reads, and its applicability with different capsule manufacturers.

Introduction: For every percentage increase in adenoma detection rate (ADR), risk of interval colorectal cancer is reduced by 3%. Adenoma prevalence is estimated to be 50%. Ideally, ADR should reflect adenoma prevalence, yet ADR varies between 5 and 55% among colonoscopist. We hypothesize that computer-assisted polyp identification could help bring low ADRs closer to true prevalence and reduce interval colorectal cancer incidence. Methods: Utilizing our colonoscopy quality database (Qulaoscopy, Docbot, Inc,), we applied the VGG-16 model as starting point (see [1]), which is a convolutional neural network (CNN) with 16 convolutional layers that was trained on natural images as part of the ImageNet challenge. We then “fine-tuned” this CNN on a set of 9000 screening colonoscopy images originating from all locations in the colorectum (and half of which contain polyps) to classify the presence or absence of polyps in images. The RGB images are scaled to a size of 224x224 pixels for this purpose and the CNN is trained using ten-fold crossvalidation, i.e. we repeat the training and testing ten times with different parts of the data for training/testing the model. Results: On the task of classifying polyp versus non-polyp containing images we obtain an accuracy of 96% and AUC (area under the ROC curve) of 0.99, which is an almost perfect score (the maximum attainable AUC is 1). At a processing rate of 170 images per second, the algorithm was easily applied to live video. Conclusion: Our deep learning methology achieved extremely high accuracy for polyp detection in images and is proven applicable to live video. We predict that application of this technology during colonoscopy will assist all colonoscopist to achieve ADRs that approach true adenoma prevalence.Figure: AUC (area under the ROC curve) for Polyp Detection.