Hybrid features to classify lung tumor using machine learning

A lung tumor is an abnormal mass of cells inside a body. As a benign tumor is unproblematic, but a malignant tumor is cancerous because it can travel across the body and interfere with its surrounding tissue. Detecting these cancerous cells in the lung is important because delayed detection may hamper effective treatment options, leading to a lower survival rate. However, classifying tumor malignancy is highly dependent on the knowledge and experience of the radiologist. This study combines texture-based features extracted from lung Computed Tomography Scan (CT Scan) images such as Gray Level Co-occurrence Matrix (GLCM), Gray Level Run-length Matrix (GLRLM), Gray Level Size-zone Matrix (GLSZM), and Haralick Features aims to create a lung tumor classification system. This research contributes by creating an efficient and reliable system through Relief-F feature selection that uses features with the highest weight in rank that are able to differentiate classes of tumor malignancy and help medical professionals diagnose tumors more early in the treatment. As a comparison, several conventional machine learning classifiers, including SVM RBF, KNN, RF, DT, and XGBoost, were utilized to evaluate classifier performance. The result showed that the accuracy of the proposed hybrid features with a random forest classifier was the most performing approach with an evaluation score of accuracy of 99.55%, precision of 99.55%, recall of 99.55%, and F1-Score of 99.54%. Furthermore, accuracy among other classifiers was also higher than 90%. Proofing the selected features retain essential class information, demonstrating the study’s applicability in developing automated lung tumor classification systems from CT scans.