Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Health Technology and Informatics | en_US |
| dc.contributor.advisor | Cai, Jing (HTI) | en_US |
| dc.creator | Ma, Zongrui | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13860 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Towards reliable radiomics modeling : a multi-institutional multi-modality feature repeatability study on head and neck cancer patients | en_US |
| dcterms.abstract | Background: Radiomics has shown promise in cancer diagnosis and treatment decision making. However, the reliability of radiomic features and models remains a critical challenge. While feature repeatability has been extensively studied, the relationship between feature stability and model reliability is not well understood. Furthermore, understanding feature repeatability across imaging and data modalities and its relationship with image characteristics is essential for developing robust clinical prediction tools. Current research lacks comprehensive evaluation of feature repeatability across different imaging scenarios, its systematic correlation with image properties, and how feature reliability translates to model stability. | en_US |
| dcterms.abstract | Purpose: This thesis systematically investigates feature repeatability and its impact on radiomics modeling through multi-institutional multi-modality analysis in head and neck cancer. The research encompasses three primary objectives: 1) to systematically quantify and compare feature repeatability across CT and MRI modalities in nasopharyngeal carcinoma, 2) to compare the feature repeatability in CT radiomics and dosiomics features and elucidate the relationships with image characteristics, 3) to validate the beneficial impact of feature repeatability on model performance and reliability through multi-institutional analysis. | en_US |
| dcterms.abstract | Methods and Materials: A multi-institutional investigation of radiomics feature repeatability was conducted across three retrospective cohorts totaling 2,053 patients and nine institutions. Three imaging modalities were utilized including computed tomography (CT), magnetic resonance imaging (MRI), and radiation dose maps. The study collected pre-treatment CT images of head-and-neck cancer patients from seven institutions via The Cancer Imaging Archive (TCIA), CT and MRI data of nasopharyngeal carcinoma patients from Queen Elizabeth Hospital (2012-2016), and planning CT and dose distributions of cervical cancer patients (2012-2022) from Peking University Third Hospital. A comprehensive perturbation framework was implemented to evaluate feature robustness, incorporating geometric transformations (rotation: ±20°, translation: 0-0.8 pixels) and contour randomization through deformation vector fields. Radiomics features were extracted using PyRadiomics, encompassing first-order statistics, morphological metrics, and texture characteristics derived from original, Laplacian-of-Gaussian, and wavelet-decomposed images. Feature stability was quantified using intraclass correlation coefficients (ICC), with stratified thresholds (0-0.9) for repeatability assessment. The impact of feature reliability on model performance was evaluated through internal cross-validation and external institutional validation using Cox proportional hazards regression. Model discrimination was assessed via concordance indices (C-index), while risk stratification significance was determined through Kaplan-Meier survival analysis. For the cervical cancer cohort, comparative analyses between radiomics and dosiomics feature stability were conducted across multiple regions of interest, providing insights into data modality-specific feature robustness. All feature extraction algorithms were implemented using standardized computational frameworks adherent to the Image Biomarker Standardization Initiative. | en_US |
| dcterms.abstract | Results: Quantitative assessment of feature stability across imaging modalities demonstrated superior repeatability in shape-based features (mean ICC: 0.92, 95% CI: 0.89-0.94), with MRI-derived radiomic features exhibiting significantly higher stability compared to CT-derived features (86.8% vs 42.3% features achieving ICC>0.9, P<0.001). In the comparative analysis of CT-radiomics and dosiomics features in cervical cancer specimens, CT radiomic features demonstrated superior repeatability metrics (mean ICC: 0.81, 95% CI: 0.78-0.84) compared to dosiomics features (mean ICC: 0.67, 95% CI: 0.63-0.71), particularly in features extracted from rectum and femoral ROIs (mean ICC: 0.85, 95% CI: 0.82-0.88). Feature repeatability demonstrated strong correlations with image characteristics, specifically entropy (r=0.76, P<0.001), uniformity (r=-0.72, P<0.001), and variance (r=0.74, P<0.001) across all modalities. The integration of highly repeatable features (ICC ≥ 0.9) consistently enhanced prognostic model performance across different head and neck cancer datasets, demonstrating improved validation metrics (ΔC-index: +0.02 to +0.05, P<0.01) and enhanced model generalizability. Notably, stringent repeatability criteria effectively mitigated performance degradation in heterogeneous datasets, suggesting that feature stability is crucial for robust model development. | en_US |
| dcterms.abstract | Conclusions: Our work presents a comprehensive investigation of radiomic feature reliability across multiple imaging and data modalities and institutions in head and neck cancer patients. Through our rigorous multi-institutional analyses and systematic evaluation of feature stability patterns, we characterized the intrinsic relationships between image characteristics and feature repeatability and developed a robust framework for reliable radiomics modeling. Our findings demonstrate that prescreening and incorporation of high-reliable features significantly enhances model performance and generalizability, advancing the theoretical and practical foundations of radiomics. Our work establishes a methodological framework for developing more reliable radiomics models and facilitates their translation into clinical practice. | en_US |
| dcterms.extent | 103 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2025 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13860