Pancreatic cancer, although rare, is highly deadly, often diagnosed in advanced stages. Surgical intervention as well as a better prognosis are possible thanks to early diagnosis of neoplastic tumors including pancreatic intraepithelial neoplastic and mucinous cystic neoplasms. In our research, we focused on using non-invasive approaches and artificial intelligence (AI) to predict pancreatic cancer by identifying early warning signs. We developed a Blended Deep Learning Model (BDLM) that combines Convolutional Neural Network (CNN), Particle Swarm Optimization (PSO), and Recurrent Neural Network (RNN) for pancreatic tumor identification. To enhance CT images of pancreatic tumors, we used Particle Swarm Optimization with Split and Merge Segmentation and applied image processing techniques like Gaussian filter. To address spatial discrepancy segmentation, we employed RNN to refine segmentation and improve smoothness and shape based on CNN findings. We evaluated the training and configuration objectives of the model in terms of pancreatic tumor image segmentation and classification performance. Through numerous simulations, we validated the enhanced performance of our proposed approach, which outperformed current techniques, as revealed through thorough comparative data analysis.