Pancreatic cancer is a lethal form of malignancy with a high mortality rate. The poor survival rate is likely due to its aggressive nature and its early detection challenges. The current confirmative diagnosis of pancreatic cancer is through manual examination of tissue biopsy by pathologists under a microscope. This method is time consuming and often results in late or deferred treatment. The lack of pathologists specialised in hepatobiliary in Malaysia contributes to further delay of treatment especially in rural areas. This study confers a classification of pancreatic cancer cells extracted from whole slides images to aid pathologists to reach a diagnosis faster. A total of 21 pancreatic fine needle aspirate Papanicolaou (PAP) smears and cell blocks (CB) was obtained from the Pathology Department, Hospital Selayang, digitised using a whole slide scanner and segmented into single cell images for annotation. Annotated single cell images of cancerous and normal cell types were used to train an artificial intelligence (AI)-model that is able to differentiate pancreatic cancer cells from non-cancerous normal cells. From 21 whole slide images (WSIs), 4149 single cell images were extracted for establishment of cell database (3210 cancerous pancreatic cells, 939 normal pancreatic cells). DenseNet21 model was used for the training of pancreatic cancer detection. A total of 3320 single cell images was used for the training and validation, while the remaining 829 cell images were used for the testing of the developed PAP and CB models. The models were able to detect and classify the test dataset up to 97.8% accuracy for CB model and 91.04% for PAP model. This automated pancreatic cancer algorithm can be implemented together with an online digital pathology platform for the uses in hospitals across Malaysia. This can hasten diagnosis and aid rural areas with a scarce number of specialised pathologists to assist in the diagnosis of pancreatic cancer.