Deep Learning and Medical Imaging for Automatic Detection of Pancreatic Cancer

Pancreatic cancer is one of the most difficult malignancies to detect early and is one of the leading causes of cancer death because early-stage disease almost always has no symptoms. As we know, the conventional imaging and biopsy-based diagnosis typically detects the disease at an advanced stage, when treatment options are restricted and survival rates are poor. Objectives: The objective of this study is to develop the deep learning based model for the early detection of pancreatic cancer referred to the medical imaging techniques, e.g. Computed Tomography (CT) and Magnetic resonance imaging (MRI). Feature extraction methods with Convolutional Neural Networks{CNNs} are employed by the proposed system to discover minutia and information within the medical scan that is too subtle to be noted by the human eye. Preprocessing, data augmentation and a clear definition of parameters for our CNN architecture contributes well in higher accuracy and reliability of result from the detector also. The model is trained using a diverse dataset of annotated pancreatic cancer, non-cancer images, which improves generalizability and minimizes false positives. Feature transfer, attention and ensemble learning are some of the methods that have been integrated in order to enhance classification/segmentation of tumor regions. Established performance metrics (Accuracy, Sensitivity, Specificity, and F1-score) have been implemented to evaluate the robustness of the model for further deployment in real time. This paper presents a web-based tool for automated and early diagnosis of pancreatic cancer to help radiologists and other healthcare professionals. The early diagnosis of the pancreatic cancer can lead to timely control and increased patient survival rates.