ABSTRACT
Background and Aims: Recent studies have shown that artificial intelligence-based computer-aided detection systems possess great potential in reducing the heterogeneous performance of doctors during endoscopy. However, most existing studies are based on high-quality static images available in open-source databases with relatively small data volumes, and, hence, are not applicable for routine clinical practice. This research aims to integrate multiple deep learning algorithms and develop a system (DeFrame) that can be used to accurately detect intestinal polyps in real time during clinical endoscopy. Methods: A total of 681 colonoscopy videos were collected for retrospective analysis at Xiangya Hospital of Central South University from June 2019 to June 2020. To train the machine learning (ML)-based system, 6,833 images were extracted from 48 collected videos, and 1,544 images were collected from public datasets. The DeFrame system was further validated with two datasets, consisting of 24,486 images extracted from 176 collected videos and 12,283 images extracted from 259 collected videos. The remaining 198 collected full-length videos were used for the final test of the system. The measurement metrics were sensitivity and specificity in validation dataset 1, precision, recall and F1 score in validation dataset 2, and the overall performance when tested in the complete video perspective. Results: A sensitivity and specificity of 79.54 and 95.83%, respectively, was obtained for the DeFrame system for detecting intestinal polyps. The recall and precision of the system for polyp detection were determined to be 95.43 and 92.12%, respectively. When tested using full colonoscopy videos, the system achieved a recall of 100% and precision of 80.80%. Conclusion: We have developed a fast, accurate, and reliable DeFrame system for detecting polyps, which, to some extent, is feasible for use in routine clinical practice.
ABSTRACT
Colorectal cancer (CRC) is a common and lethal disease. Globally, CRC is the third most commonly diagnosed cancer in males and the second in females. The most effective way to prevent CRC is through using colonoscopy to identify and remove precancerous growths at an early stage. The detection and removal of colorectal polyps have been found to be associated with a reduction in mortality from colorectal cancer. However, the false negative rate of polyp detection during colonoscopy is often high even for experienced physicians. With recent advances in deep learning based object detection techniques, automated polyp detection shows great potential in helping physicians reduce false positive rate during colonoscopy. In this paper, we propose a novel anchor-free instance segmentation framework that can localize polyps and produce the corresponding instance level masks without using predefined anchor boxes. Our framework consists of two branches: (a) an object detection branch that performs classification and localization, (b) a mask generation branch that produces instance level masks. Instead of predicting a two-dimensional mask directly, we encode it into a compact representation vector, which allows us to incorporate instance segmentation with one-stage bounding-box detectors in a simple yet effective way. Moreover, our proposed encoding method can be trained jointly with object detector. Our experiment results show that our framework achieves a precision of 99.36% and a recall of 96.44% on public datasets, outperforming existing anchor-free instance segmentation methods by at least 2.8% in mIoU on our private dataset.
