Attention-Guided CNN with LSTM for Temporal Gastrointestinal Disease Progression Prediction in Endoscopic Video Sequences

The detection and tracking of changes in the progression of GI disease using endoscopic video analysis remains difficult due to temporal changes and image complexity. Proper models of prediction are critical in diagnosing individuals at an early stage and offering them treatment. In the work presented in this paper, we introduce novel Attention-Guided Convolutional Neural Network (AG-CNN) combined with LSTM networks to analyze temporal endoscopic video sequences. It uses spatial attention to catch necessary areas and temporal attention to capture assertive illness transformation in video frames. For disease progression prediction the proposed model yield 94.8% accuracy with 93.2% sensitivity and 95.6% specificity. The attention mechanism increased the feature extraction speed by 22 percent when compared to traditional CNN-LSTM classes. Independent validation checks confirmed that the two models produced a prediction concordance of 92.7% based on different patient samples. The overall performance of temporal GI disease progression prediction by the AG-CNN-LSTM framework shows potential usage in clinical diagnosis by providing automated disease progression monitoring and treatment planning.