An improved machine learning approach for diagnosis of pancreatic cancer using multi modal clinical data

Abstract

Clinical diagnosis is a challenging task since data sources are newlineheterogeneous. Owing to the advancements in the healthcare field, the newlinebenefits of diagnosis have proved to exhibit an improvement. Among clinical newlineexperts, there is a positivity and hope that the use of a computer-aided newlinediagnosis system, will aid radiologists in interpreting images and better newlinedecision making for diagnosis. Human beings have always been susceptible to newlinemany kinds of life-threatening diseases. Several diseases pose a serious threat newlineto humans out of which, Pancreatic Adenocarcinoma (PDAC) is common, newlineaffecting people over 45 years of age. PDAC ranks 4th in the world among cancers. Generally, the conventional approach of examining individuals to diagnose the disease are newlineCT, MRI, PET scan, and Ultrasound (US). This provides a pathway for newlinebuilding an efficient pancreatic cancer diagnosis system from multi-modal newline(Structured EHRs, Jaundiced Eye images, and pancreatic CTs) clinical data to newlinepredict the risk level of individuals and detect the signs of pancreatic disease newlinefor immediate surgical planning. Extracting patterns from multi-modal healthcare data is challenging given the nature of the pancreas, as it exhibits variable shape and volume, its newlinelocation in the abdomen also makes diagnosis challenging, problems with newlinenoisy data, and many other risk factors associated with the disease. In this newlinepresent situation, extracting useful patterns of disease is a difficult task. The newlinedetection of tumors from jaundice eye images and pancreatic CT images is a newlinecomplex process with the current segmentation methods. newline This research aims to extract meaningful disease patterns from newlineclinical data, taking into account the different sources of medical data and the newlinelack of a common framework for diagnosing pancreatic disease from newlinemultimodal clinical data. The thesis intends to present an efficient mining framework to newlineclassify the severity level of the patients affected by pancreatic cancer for newlineimmediate treatment planning. This thesis discusses jaundiced eye sclera newlinesegmentation, feature selection, pancreatic CT image segmentation, and risk newlinelevel categorization for clinical decision support. Based on the literature review, there are various methods implemented to segment the pancreatic tumors and detect lesions from the newlinepancreatic CT image for severity level measurement newline newline