Pancreatic Tumor Detection using Image Processing

Since pancreatic cancer is known to be one of the most dangerous and deadly disease, its survival rates are very poor since it usually does not manifest the presence until the disease has already reached an advanced stage. The anatomy complexity and early symptoms of the pancreas make proper and timely diagnosis even more complicated and delays treatment, which is sufficient to fatally affect the result of the treatment. The paper reviews the image processing techniques and AI-driven methodologies for the early detection of pancreatic tumors and the accurate segmentations. Precisely, this paper tackles CAD systems primarily developed based on CNNs along with U-Net-based architectures in the context of a medical analysis, especially involving complex imaging modalities such as CT and MRI scans. These methods, primarily developed for biomedical image segmentation, address key diagnostic metrics, such as Dice Similarity Coefficient, sensitivity, and specificity, which are critical for determining the accuracy for tumor delineation and overall diagnostic reliability. These models based on CNN can identify tumours by differentiating it from adjacent tissue through enhanced and pre-processed images, thus limiting the limitations and the element of subjectivity behind standard imaging by radiologists. The architecture of U-Net was specially designed towards segmentation of medical images, and the results from this architectural design clearly differentiate the subtle tissue texture in the pancreas, which eventually has been a positive feature for their early detection. This model processes imaging data in a way such that effects of misdiagnosis and delayed diagnosis become much less burden to health care providers while precision levels in diagnoses increase.