Diagnosis of Hemorrhage in Wireless CapsuleEndoscopy Images using a Hybrid Convolutional Neural Network -Long Short Term Memory (CNN –LSTM) Model

The non-invasive nature of Wireless Capsule Endoscopy (the WCE) along with its high-definition imaging capabilities make it an essential diagnostic procedure for gastrointestinal illnesses. Wireless capsule endoscopy is considered a non-invasive method. It involves swallowing a small capsule containing a camera and other components, which then travels through the gastrointestinal tract, capturing images along the way. Since the capsule is ingested orally and does not require the insertion of instruments or devices into the body through incisions, it is classified as a non-invasive procedure. This approach eliminates the need for sedation and reduces discomfort compared to traditional endoscopic procedures, which involve the insertion of a flexible tube (endoscope) through the mouth or anus. However, the large amount of data generated presents challenges in analysis and interpretation. Healthcare practitioners rely on automated segmentation of WCE images into clinically relevant categories, such as detecting bleeding, for prompt diagnosis and treatment. In order to categorize WCE images into two groups—bleeding and normal—this study presents a hybrid model that combines the CNN and LSTM (Long Short-Term Memory) architectures. The CNN component extracts spatial correlations within images, while the LSTM component handles temporal dynamics. Using transfer learning techniques, the model is trained on a huge dataset of annotated WCE photos. Extensive testing on a diverse dataset of WCE images from various gastrointestinal patients demonstrates that the hybrid CNN-LSTM model outperforms standalone CNN or LSTM models in classification accuracy. This approach is resilient against common illumination and image quality variations in clinical environments.