Performance of Convolutional Neural Networks for Polyp Localization on Public Colonoscopy Image Datasets

Artificial intelligence in colonoscopy: from detection to diagnosis

This study reviews the recent progress of artificial intelligence for colonoscopy from detection to diagnosis. The source of data was 27 original studies in PubMed. The search terms were "colonoscopy" (title) and "deep learning" (abstract). The eligibility criteria were: (1) the dependent variable of gastrointestinal disease; (2) the interventions of deep learning for classification, detection and/or segmentation for colonoscopy; (3) the outcomes of accuracy, sensitivity, specificity, area under the curve (AUC), precision, F1, intersection of union (IOU), Dice and/or inference frames per second (FPS); (3) the publication year of 2021 or later; (4) the publication language of English. Based on the results of this study, different deep learning methods would be appropriate for different tasks for colonoscopy, e.g., Efficientnet with neural architecture search (AUC 99.8%) in the case of classification, You Only Look Once with the instance tracking head (F1 96.3%) in the case of detection, and Unet with dense-dilation-residual blocks (Dice 97.3%) in the case of segmentation. Their performance measures reported varied within 74.0-95.0% for accuracy, 60.0-93.0% for sensitivity, 60.0-100.0% for specificity, 71.0-99.8% for the AUC, 70.1-93.3% for precision, 81.0-96.3% for F1, 57.2-89.5% for the IOU, 75.1-97.3% for Dice and 66-182 for FPS. In conclusion, artificial intelligence provides an effective, non-invasive decision support system for colonoscopy from detection to diagnosis.

Optical diagnosis in still images of colorectal polyps: comparison between expert endoscopists and PolyDeep, a Computer-Aided Diagnosis system

Background

PolyDeep is a computer-aided detection and classification (CADe/x) system trained to detect and classify polyps. During colonoscopy, CADe/x systems help endoscopists to predict the histology of colonic lesions.

Objective

To compare the diagnostic performance of PolyDeep and expert endoscopists for the optical diagnosis of colorectal polyps on still images.

Methods

PolyDeep Image Classification (PIC) is an in vitro diagnostic test study. The PIC database contains NBI images of 491 colorectal polyps with histological diagnosis. We evaluated the diagnostic performance of PolyDeep and four expert endoscopists for neoplasia (adenoma, sessile serrated lesion, traditional serrated adenoma) and adenoma characterization and compared them with the McNemar test. Receiver operating characteristic curves were constructed to assess the overall discriminatory ability, comparing the area under the curve of endoscopists and PolyDeep with the chi- square homogeneity areas test.

Results

The diagnostic performance of the endoscopists and PolyDeep in the characterization of neoplasia is similar in terms of sensitivity (PolyDeep: 89.05%; E1: 91.23%, p=0.5; E2: 96.11%, p<0.001; E3: 86.65%, p=0.3; E4: 91.26% p=0.3) and specificity (PolyDeep: 35.53%; E1: 33.80%, p=0.8; E2: 34.72%, p=1; E3: 39.24%, p=0.8; E4: 46.84%, p=0.2). The overall discriminative ability also showed no statistically significant differences (PolyDeep: 0.623; E1: 0.625, p=0.8; E2: 0.654, p=0.2; E3: 0.629, p=0.9; E4: 0.690, p=0.09). In the optical diagnosis of adenomatous polyps, we found that PolyDeep had a significantly higher sensitivity and a significantly lower specificity. The overall discriminative ability of adenomatous lesions by expert endoscopists is significantly higher than PolyDeep (PolyDeep: 0.582; E1: 0.685, p < 0.001; E2: 0.677, p < 0.0001; E3: 0.658, p < 0.01; E4: 0.694, p < 0.0001).

Conclusion

PolyDeep and endoscopists have similar diagnostic performance in the optical diagnosis of neoplastic lesions. However, endoscopists have a better global discriminatory ability than PolyDeep in the optical diagnosis of adenomatous polyps.

Self-supervised representation learning using feature pyramid siamese networks for colorectal polyp detection

Colorectal cancer is a leading cause of cancer-related deaths globally. In recent years, the use of convolutional neural networks in computer-aided diagnosis (CAD) has facilitated simpler detection of early lesions like polyps during real-time colonoscopy. However, the majority of existing techniques require a large training dataset annotated by experienced experts. To alleviate the laborious task of image annotation and utilize the vast amounts of readily available unlabeled colonoscopy data to further improve the polyp detection ability, this study proposed a novel self-supervised representation learning method called feature pyramid siamese networks (FPSiam). First, a feature pyramid encoder module was proposed to effectively extract and fuse both local and global feature representations among colonoscopic images, which is important for dense prediction tasks like polyp detection. Next, a self-supervised visual feature representation containing the general feature of colonoscopic images is learned by the siamese networks. Finally, the feature representation will be transferred to the downstream colorectal polyp detection task. A total of 103 videos (861,400 frames), 100 videos (24,789 frames), and 60 videos (15,397 frames) in the LDPolypVideo dataset are used to pre-train, train, and test the performance of the proposed FPSiam and its counterparts, respectively. The experimental results have illustrated that our FPSiam approach obtains the optimal capability, which is better than that of other state-of-the-art self-supervised learning methods and is also higher than the method based on transfer learning by 2.3 mAP and 3.6 mAP for two typical detectors. In conclusion, FPSiam provides a cost-efficient solution for developing colorectal polyp detection systems, especially in conditions where only a small fraction of the dataset is labeled while the majority remains unlabeled. Besides, it also brings fresh perspectives into other endoscopic image analysis tasks.

Public Imaging Datasets of Gastrointestinal Endoscopy for Artificial Intelligence: a Review

With the advances in endoscopic technologies and artificial intelligence, a large number of endoscopic imaging datasets have been made public to researchers around the world. This study aims to review and introduce these datasets. An extensive literature search was conducted to identify appropriate datasets in PubMed, and other targeted searches were conducted in GitHub, Kaggle, and Simula to identify datasets directly. We provided a brief introduction to each dataset and evaluated the characteristics of the datasets included. Moreover, two national datasets in progress were discussed. A total of 40 datasets of endoscopic images were included, of which 34 were accessible for use. Basic and detailed information on each dataset was reported. Of all the datasets, 16 focus on polyps, and 6 focus on small bowel lesions. Most datasets (n = 16) were constructed by colonoscopy only, followed by normal gastrointestinal endoscopy and capsule endoscopy (n = 9). This review may facilitate the usage of public dataset resources in endoscopic research.

Hybrid Models for Endoscopy Image Analysis for Early Detection of Gastrointestinal Diseases Based on Fused Features

The gastrointestinal system contains the upper and lower gastrointestinal tracts. The main tasks of the gastrointestinal system are to break down food and convert it into essential elements that the body can benefit from and expel waste in the form of feces. If any organ is affected, it does not work well, which affects the body. Many gastrointestinal diseases, such as infections, ulcers, and benign and malignant tumors, threaten human life. Endoscopy techniques are the gold standard for detecting infected parts within the organs of the gastrointestinal tract. Endoscopy techniques produce videos that are converted into thousands of frames that show the disease's characteristics in only some frames. Therefore, this represents a challenge for doctors because it is a tedious task that requires time, effort, and experience. Computer-assisted automated diagnostic techniques help achieve effective diagnosis to help doctors identify the disease and give the patient the appropriate treatment. In this study, many efficient methodologies for analyzing endoscopy images for diagnosing gastrointestinal diseases were developed for the Kvasir dataset. The Kvasir dataset was classified by three pre-trained models: GoogLeNet, MobileNet, and DenseNet121. The images were optimized, and the gradient vector flow (GVF) algorithm was applied to segment the regions of interest (ROIs), isolating them from healthy regions and saving the endoscopy images as Kvasir-ROI. The Kvasir-ROI dataset was classified by the three pre-trained GoogLeNet, MobileNet, and DenseNet121 models. Hybrid methodologies (CNN-FFNN and CNN-XGBoost) were developed based on the GVF algorithm and achieved promising results for diagnosing disease based on endoscopy images of gastroenterology. The last methodology is based on fused CNN models and their classification by FFNN and XGBoost networks. The hybrid methodology based on the fused CNN features, called GoogLeNet-MobileNet-DenseNet121-XGBoost, achieved an AUC of 97.54%, accuracy of 97.25%, sensitivity of 96.86%, precision of 97.25%, and specificity of 99.48%.

Hybrid Techniques for Diagnosing Endoscopy Images for Early Detection of Gastrointestinal Disease Based on Fusion Features

Gastrointestinal (GI) diseases, particularly tumours, are considered one of the most widespread and dangerous diseases and thus need timely health care for early detection to reduce deaths. Endoscopy technology is an effective technique for diagnosing GI diseases, thus producing a video containing thousands of frames. However, it is difficult to analyse all the images by a gastroenterologist, and it takes a long time to keep track of all the frames. Thus, artificial intelligence systems provide solutions to this challenge by analysing thousands of images with high speed and effective accuracy. Hence, systems with different methodologies are developed in this work. The first methodology for diagnosing endoscopy images of GI diseases is by using VGG-16 + SVM and DenseNet-121 + SVM. The second methodology for diagnosing endoscopy images of gastrointestinal diseases by artificial neural network (ANN) is based on fused features between VGG-16 and DenseNet-121 before and after high-dimensionality reduction by the principal component analysis (PCA). The third methodology is by ANN and is based on the fused features between VGG-16 and handcrafted features and features fused between DenseNet-121 and the handcrafted features. Herein, handcrafted features combine the features of gray level cooccurrence matrix (GLCM), discrete wavelet transform (DWT), fuzzy colour histogram (FCH), and local binary pattern (LBP) methods. All systems achieved promising results for diagnosing endoscopy images of the gastroenterology data set. The ANN network reached an accuracy, sensitivity, precision, specificity, and an AUC of 98.9%, 98.70%, 98.94%, 99.69%, and 99.51%, respectively, based on fused features of the VGG-16 and the handcrafted.

Negative Samples for Improving Object Detection-A Case Study in AI-Assisted Colonoscopy for Polyp Detection

Deep learning object-detection models are being successfully applied to develop computer-aided diagnosis systems for aiding polyp detection during colonoscopies. Here, we evidence the need to include negative samples for both (i) reducing false positives during the polyp-finding phase, by including images with artifacts that may confuse the detection models (e.g., medical instruments, water jets, feces, blood, excessive proximity of the camera to the colon wall, blurred images, etc.) that are usually not included in model development datasets, and (ii) correctly estimating a more realistic performance of the models. By retraining our previously developed YOLOv3-based detection model with a dataset that includes 15% of additional not-polyp images with a variety of artifacts, we were able to generally improve its F1 performance in our internal test datasets (from an average F1 of 0.869 to 0.893), which now include such type of images, as well as in four public datasets that include not-polyp images (from an average F1 of 0.695 to 0.722).

Diagnosis of Inflammatory Bowel Disease and Colorectal Cancer through Multi-View Stacked Generalization Applied on Gut Microbiome Data

Most of the microbiome studies suggest that using ensemble models such as Random Forest results in best predictive power. In this study, we empirically evaluate a more powerful ensemble learning algorithm, multi-view stacked generalization, on pediatric inflammatory bowel disease and adult colorectal cancer patients’ cohorts. We aim to check whether stacking would lead to better results compared to using a single best machine learning algorithm. Stacking achieves the best test set Average Precision (AP) on inflammatory bowel disease dataset reaching AP = 0.69, outperforming both the best base classifier (AP = 0.61) and the baseline meta learner built on top of base classifiers (AP = 0.63). On colorectal cancer dataset, the stacked classifier also outperforms (AP = 0.81) both the best base classifier (AP = 0.79) and the baseline meta learner (AP = 0.75). Stacking achieves best predictive performance on test set outperforming the best classifiers on both patient cohorts. Application of the stacking solves the issue of choosing the most appropriate machine learning algorithm by automating the model selection procedure. Clinical application of such a model is not limited to diagnosis task only, but it also can be extended to biomarker selection thanks to feature selection procedure.