Use of U-Net Convolutional Neural Networks for Automated Segmentation of Fecal Material for Objective Evaluation of Bowel Preparation Quality in Colonoscopy.

Background: Adequate bowel cleansing is important for colonoscopy performance evaluation. Current bowel cleansing evaluation scales are subjective, with a wide variation in consistency among physicians and low reported rates of accuracy. We aim to use machine learning to develop a fully automatic segmentation method for the objective evaluation of the adequacy of colon preparation. Methods: Colonoscopy videos were retrieved from a video data cohort and transferred to qualified images, which were randomly divided into training, validation, and verification datasets. The fecal residue was manually segmented. A deep learning model based on the U-Net convolutional network architecture was developed to perform automatic segmentation. The performance of the automatic segmentation was evaluated on the overlap area with the manual segmentation. Results: A total of 10,118 qualified images from 119 videos were obtained. The model averaged 0.3634 s to segmentate one image automatically. The models produced a strong high-overlap area with manual segmentation, with 94.7% ± 0.67% of that area predicted by our AI model, which correlated well with the area measured manually (r = 0.915, p < 0.001). The AI system can be applied in real-time qualitatively and quantitatively. Conclusions: We established a fully automatic segmentation method to rapidly and accurately mark the fecal residue-coated mucosa for the objective evaluation of colon preparation.

A novel machine learning-based algorithm to identify and classify lesions and anatomical landmarks in colonoscopy images.

OBJECTIVES: Computer-aided diagnosis (CAD)-based artificial intelligence (AI) has been shown to be highly accurate for detecting and characterizing colon polyps. However, the application of AI to identify normal colon landmarks and differentiate multiple colon diseases has not yet been established. We aimed to develop a convolutional neural network (CNN)-based algorithm (GUTAID) to recognize different colon lesions and anatomical landmarks. METHODS: Colonoscopic images were obtained to train and validate the AI classifiers. An independent dataset was collected for verification. The architecture of GUTAID contains two major sub-models: the Normal, Polyp, Diverticulum, Cecum and CAncer (NPDCCA) and Narrow-Band Imaging for Adenomatous/Hyperplastic polyps (NBI-AH) models. The development of GUTAID was based on the 16-layer Visual Geometry Group (VGG16) architecture and implemented on Google Cloud Platform. RESULTS: In total, 7838 colonoscopy images were used for developing and validating the AI model. An additional 1273 images were independently applied to verify the GUTAID. The accuracy for GUTAID in detecting various colon lesions/landmarks is 93.3% for polyps, 93.9% for diverticula, 91.7% for cecum, 97.5% for cancer, and 83.5% for adenomatous/hyperplastic polyps. CONCLUSIONS: A CNN-based algorithm (GUTAID) to identify colonic abnormalities and landmarks was successfully established with high accuracy. This GUTAID system can further characterize polyps for optical diagnosis. We demonstrated that AI classification methodology is feasible to identify multiple and different colon diseases.