| Victor Aderanti |
Early cancer detection is vital for improving patient outcomes worldwide. This study focuses on advancing Region Of Interest (ROI) segmentation and feature extraction in medical imaging through Radiomics techniques, employing tools such as 3D Slicer, Pyradiomics, and Python. Various dimensionality reduction methods, including PCA, K-means, t-SNE, ISOMAP, and Hierarchical Clustering, were utilized to manage high-dimensional features, aiming to improve both interpretability and efficiency. The effectiveness of the reduced feature sets in predicting T-staging, a crucial aspect of the TNM cancer staging system, was evaluated. Multinomial logistic regression models were created and assessed using metrics such as MSE, AIC, BIC, and the Deviance Test. The dataset comprised CT and PET-CT DICOM images from 131 lung cancer patients. Results indicated that PCA identified 14 features, Hierarchical Clustering found 17, t-SNE discovered 58, and ISOMAP identified 40 features, with texture-based features being particularly significant. This research underscores the promise of combining Radiomics with unsupervised learning techniques to enhance cancer prediction capabilities from medical images. |