AI-POWERED PANCREATIC CANCER DIAGNOSIS SYSTEM

Abstract

Pancreatic cancer is one of the deadliest cancers and it is usually detected very late because early symptoms are hard to notice. To solve this problem, this project creates an Ai based system that uses CT scan images to detect pancreatic cancer early and accurately.

In this project , the method is quite different as it is using light weight and flexible Ai model built with Transfer learning, This means we take a pretrained CNN model like MobileNet and trained it on a properly organized dataset taken from Kaggle. This greatly reduces training time and computer resources while still giving excellent accuracy.

The model is designed to work well even when the dataset is small, making it suitable for hospitals or clinics with limited resources. We also use improved preprocessing and feature extraction methods to make the predictions more reliable and efficient.

The early results of the model are very promising: it gives 99% accuracy, 1.000 precision, 0.987 recall, and an F1-score of 0.993. These results show that the model is strong, fast, and highly accurate. The final system becomes a light, easy-to-use tool that supports radiologists in identifying pancreatic cancer, which can help doctors provide treatment earlier and improve patient survival.