Ensemble model assisted classification of gastrointestinal bleeding using wireless capsule endoscopy

Wireless Capsule Endoscopy (WCE) is a useful method for imaging the intestines painlessly and looking into gastrointestinal tract diseases. The investigation of the enormous dataset produced by the patient's digestive tract WCE imaging takes a lot of time and a unique set of skills from a medical professional. Therefore, there is a strong need for effective analysis techniques that minimize examination times and increase diagnostic accuracy. To address the problem, the authors devise an approach that can automatically analyze WCE images to spot anomalies and help medical professionals make reliable diagnoses. This study adopts CNN based ensemble approach that combines the DenseNet201, MobileNetV2, and EfficientNetB7 model to classify WCE bleeding images. The CNN-based average ensemble model's performance is assessed using a dataset of 1309 bleeding and 1309 non-bleeding images generated by the wireless capsule endoscopy (WCE) tube. The suggested ensemble model achieved an accuracy of 98.74%, with precision, recall, and F1 score of 98.06%, 98.83%, and 98.44%, respectively. The proposed model is also compared with individual model and with custom built CNN model. The findings indicate that the proposed method offers an acceptable alternative and may prove beneficial for healthcare professionals.