Multiclass Anomaly Detection in GI Endoscopic Images using Optimized Deep One-class Classification in an Imbalanced Dataset

Wireless Capsule Endoscopy helps physicians examine the gastrointestinal (GI) tract noninvasively, with the cost of generating many images. Many available datasets, such as KID2 and Kvasir, suffer from imbalance issue which make it difficult to train an effective artificial intelligence (AI) system. Moreover, increasing number of classes makes the problem worse. In this study, an ensemble of one-class classifiers is used for detecting anomaly. This method focuses on learning single models using samples from only one class, and ensemble all models for multiclass classification. A total of 1,778 normal, 227 inflammation, 303 vascular diseases, and 44 polyp images have been used from the KID2 dataset. In the first step, deep features are extracted based on an autoencoder architecture from the preprocessed images. Then, these features are oversampled using Synthetic Minority Over-sampling Technique and clustered using Ordering Points to Identify the Clustering Structure. To create one-class classification model, the Support Vector Data Descriptions are trained on each cluster with the help of Ant Colony Optimization, which is also used for tuning clustering parameters for improving F1-score. This process is applied on each classes and ensemble of final models used for multiclass classification. The entire algorithm ran 5 times and obtained F1-score 96.3 +- 0.2% and macro-average F1-score 85.0 +- 0.4%, for anomaly detection and multiclass classification, respectively. The results are compared with GoogleNet, AlexNet, Resnet50, VGG16 and other published algorithms, and demonstrate that the proposed method is a competitive choice for multiclass class anomaly detection in GI images.