Enhanced Pancreatic Cancer Detection through Deep Learning Based Hybrid Feature Fusion and LSTM Classifier

Because of its subtle and diverse symptoms, pancreatic cancer detection in medical imaging remains a serious issue.Recent developments in image processing and deep learning open up new possibilities for creating reliable detection techniques.This research presents a detailed methodology for pancreatic cancer detection integrating deep learning and image processing techniques.The process entails obtaining high-resolution CT scan images from the NIH-Pancreas-CT and MSD dataset with competent annotations, and then augmenting the data to enhance the variety of the training dataset.A convolutional Neural Network (CNN) is used to extract deep features following total variation L1 (TV-L1) norm-based pre-processing.For precise region partitioning, the segmentation procedure makes use of 2D Rnyi entropy optimized via Sine-Cosine Optimization.Texture features are extracted using the gray-level size zone matrix (GLSZM) approach and combined with CNN-based features to create a hybrid representation.An LSTM classifier is used for classification.Results reveal the hybrid approach's superiority, achieving 97.67% accuracy in three-stage classification and 98.53% in two-stage classification.This comprehensive methodology demonstrates the efficacy of integrating multiple techniques for robust pancreatic cancer detection, with potential implications for improving clinical outcomes.