Integrating Radiomics and Unsupervised Machine Learning Techniques for T-stage on PET Lung Cancer Images

Location

D.P. Culp Center Ballroom

Start Date

4-5-2024 9:00 AM

End Date

4-5-2024 11:30 AM

Poster Number

17

Name of Project's Faculty Sponsor

Mostafa Zahed

Faculty Sponsor's Department

Mathematics and Statistics

Classification of First Author

Graduate Student-Master’s

Competition Type

Competitive

Type

Poster Presentation

Presentation Category

Health

Abstract or Artist's Statement

Cancer remains a leading cause of death worldwide, with timely detection and diagnosis critical to treatment success. This research explores how integrating Radiomics techniques—specifically leveraging 3D Slicer, Pyradiomics, and Python—can refine and optimize region of interest (ROI) segmentation and feature extraction processes for medical image analysis. Dimension reduction methods including PCA (Principal Component Analysis), K-means, t-SNE (t-distributed Stochastic Neighbor Embedding), ISOMAP (Isometric Feature Mapping), and Hierarchical Clustering are applied to high-dimensional extracted features to obtain reduced feature representations for enhanced interpretability and analytical efficiency. T-staging, short for tumor staging, is a critical component of cancer diagnosis and treatment planning. T-staging, a component of the TNM (Tumor Node Metastasis) system, categorizes cancer based on the size and extent of the primary tumor, ranging from T0 to T4. Multinomial logistic regression models are developed to evaluate the ability of selected features to predict T-stage as an outcome, with model performance assessed using the Likelihood Ratio Test, Wald Test, Hosmer-Lemeshow Test, AIC, Deviance Test, and Score Test. The analysis primarily focuses on datasets which collected by Huiping Han, Funing Yang, and Rui Wang of Harbin from Medical University in Harbin in China in year 2022. This dataset consists of CT and PET-CT DICOM images of 150 patients with lung cancer, utilizing machine learning techniques including PCA, Hierarchical Clustering, K-means, t-SNE and ISOMAP implemented in SAS and R. Thresholds identify the most significant features for analysis. Overall, this work provides insight into how Radiomics and unsupervised learning were leveraged to enhance cancer prediction from medical images. The results suggested that texture-based features were the most essential in the analysis.

This document is currently not available here.

Share

COinS
 
Apr 5th, 9:00 AM Apr 5th, 11:30 AM

Integrating Radiomics and Unsupervised Machine Learning Techniques for T-stage on PET Lung Cancer Images

D.P. Culp Center Ballroom

Cancer remains a leading cause of death worldwide, with timely detection and diagnosis critical to treatment success. This research explores how integrating Radiomics techniques—specifically leveraging 3D Slicer, Pyradiomics, and Python—can refine and optimize region of interest (ROI) segmentation and feature extraction processes for medical image analysis. Dimension reduction methods including PCA (Principal Component Analysis), K-means, t-SNE (t-distributed Stochastic Neighbor Embedding), ISOMAP (Isometric Feature Mapping), and Hierarchical Clustering are applied to high-dimensional extracted features to obtain reduced feature representations for enhanced interpretability and analytical efficiency. T-staging, short for tumor staging, is a critical component of cancer diagnosis and treatment planning. T-staging, a component of the TNM (Tumor Node Metastasis) system, categorizes cancer based on the size and extent of the primary tumor, ranging from T0 to T4. Multinomial logistic regression models are developed to evaluate the ability of selected features to predict T-stage as an outcome, with model performance assessed using the Likelihood Ratio Test, Wald Test, Hosmer-Lemeshow Test, AIC, Deviance Test, and Score Test. The analysis primarily focuses on datasets which collected by Huiping Han, Funing Yang, and Rui Wang of Harbin from Medical University in Harbin in China in year 2022. This dataset consists of CT and PET-CT DICOM images of 150 patients with lung cancer, utilizing machine learning techniques including PCA, Hierarchical Clustering, K-means, t-SNE and ISOMAP implemented in SAS and R. Thresholds identify the most significant features for analysis. Overall, this work provides insight into how Radiomics and unsupervised learning were leveraged to enhance cancer prediction from medical images. The results suggested that texture-based features were the most essential in the analysis.