PANCREATIC CANCER DETECTION USING RADIOMICS: A COMPARATIVE STUDY BETWEEN CNN ARCHITECTURE AND VISION TRANSFORMER

A CT scan or Computed Tomography scan image is a special type of medical diagnostic image that consists of a number of different Xray pictures taken from different directions around the body and then combined together by a computer to show a slice or cross-section of the region which is present inside the body. CT scans are usually painless, quick, comprehensive and they give valuable imaging information from multiple angles which can be helpful to doctors in making a comprehensive diagnosis..Thus we can say that the CT-scan images help the medical experts to treat the patient and track the patients health conditions by correctly identifying the nature of the disease. However, addressing the dynamic needs of a massive country like India, with a high clinical volume needs an alternative to manual human interpretation of diagnostic imaging . The manual process of analysing the CT scan image and determining whether there is cancer or not is a very tedious task that consumes a lot of time with a possibility that a mistake may occur in the process of predicting whether it was cancerous or not. Thus here for our work in order to simplify the process a bit we propose a pipeline for building a model based on neural networks which will try to accurately identify from the CT-scan image whether the image is cancerous or not. At first we build a CNN based architecture for image classification and later created a Vision transformer based architecture and found out that Vision transformer based models tend to perform better than CNN networks when it comes to image classification.