Maximizing pancreatic carcinoma classification performance using parrot optimized vision transformer

Pancreatic cancer is a rare kind of cancer that is detected during the final stages. This is because the symptoms are very common and also do not show up in the starting phase. Hence an automated system for identification and classification of pancreatic cancer becomes essential. This becomes possible with the help of artificial intelligence and machine learning. The aim of this research is to develop a model that classifies pancreatic cancer using Pancreatic CT image dataset involving 1411 images from Kaggle website. The input images are augmented for increasing the dataset quality and preprocessed using Gabor filter. Segmentation is performed using UNet and features are extracted using YOLOv11 model. Pancreatic carcinoma classification is achieved using a modern deep learning-based classifier called the Vision Transformer. The classified results are optimized with the help of Parrot metaheuristic optimization algorithm. The proposed model produced an accuracy of 99%, precision of 98.5%, recall value of 97.7%, F1-Score of 96.4% and Matthew's correlation coefficient value of 97.3% in addition to true positive and false positive rates of 96.1% and 0.07%. These results are considered phenomenal and superior when compared to existing models of Random Forest, Convolutional Neural Network, Deep belief networks, and Support Vector Machine.