Performance Estimation of ML Techniques for Pancreatic Tumor Classification in PET/CT Images

Abstract

AbstractPancreatic cancer is one of the deadliest diseases and detecting the cancer at an early stage is the difficult task and the reason that ‘cancer’ symptoms happen only at a problematical period. The reliable screening tool to spot high-risk patients is lacking. Machine Learning (ML) algorithms were utilized in Positron Emission Tomography (PET)/Computed Tomography (CT) pancreatic images. Combination of PET/CT is a current mechanism, which combines both the functional information of ‘PET’ with the anatomic features of ‘CT’. Incorporated ‘PET/CT’ devices show ‘PET’ with contrast medium enhanced ‘CT’ images of the complete body in single condition. Classification techniques have a significant function in ‘Computer Aided Diagnostic (CAD)’ systems in identifying the malicious (cancerous) area from any kind of medical images. Supervised ML techniques, for example K-Nearest Neighbor (KNN), Decision Tree (DT) and Support Vector Machine (SVM) have been accepted as strong ideal in classifying features for medical images. Classification using SVM achieved 80.7% accuracy; KNN achieved 90.8% accuracy and decision tree achieved 89.9% accuracy in detecting pancreatic cancer. From the above deliberations, this paper attempts to evaluate the latest ‘ML’ techniques for ‘Pancreatic Cancer’ classification using second order statistical features from ‘PET/CT’ images. The results indicate that KNN can be achieved the maximum accuracy than SVM and DT algorithms.KeywordsComputed tomographyGrey level run length matrix methodK-nearest neighborMachine learning algorithmsPancreatic cancer