| Author: | Zhang, Jiang |
| Title: | Radiotherapy data analysis and reporting (RADAR) toolkit : an end-to-end artificial intelligence development solution for precision medicine |
| Advisors: | Cai, Jing (HTI) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Cancer -- Radiotherapy -- Planning Cancer -- Radiotherapy -- Data processing Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Health Technology and Informatics |
| Pages: | 195 pages : color illustrations |
| Language: | English |
| Abstract: | Radiotherapy is one of the mainstream treatment modalities for cancer. A large amount of structured data, including image, dose, and structure delineations, is produced during treatment planning. Technological advancement, especially artificial intelligence, facilitates the development of more sophisticated quantitative biomarkers from radiotherapy data for improved performance in precision medicine. Nevertheless, challenges in low data processing efficiency, incomplete data usage, and lack of reliability assessment hinder the development and bench-to-bedside translation. This thesis aims to develop an end-to-end and integrated RAdiotherapy Data Analysis and Reporting (RADAR) toolkit for efficient, comprehensive, and reliable quantitative biomarker developments and to evaluate its utility and performance in multiple clinical applications. RADAR is composed of GUI-equipped semi-independent modules for data curation, feature extraction, and model development. During the development of RADAR, we embedded a new multi-model feature set with innovative designs of anatomical features based on structure delineations and implemented perturbation-based repeatability assessment algorithm. By using the RADAR platform, we investigated radiomic feature (RF) repeatability and its agreements across imaging modalities and head-and-neck cancer subtypes via image perturbations, attempting to provide a direct perceptivity in RF pre-selection for robust model construction. We retrospectively collected contrast-enhanced computed tomography (CECT), contrast-enhanced T1-weight (CET1-w), T2-weight (T2-w) magnetic resonance (MR) images of 231 nasopharyngeal carcinoma (NPC) patients from Queen Elizabeth Hospital (QEH), and CECT images of 399 oropharyngeal carcinoma (OPC) patients from online database. Randomized translation and rotation were implemented to the images for mimicking scanning position stochasticity. The intra-class correlation coefficient (ICC) was calculated for each RF to assess its repeatability and quantitatively compared to evaluate the repeatability agreement. We also investigated the impact of RF repeatability on generalizable model development on Nasopharyngeal Carcinoma (NPC) cases using CET1-w MR images of 286 NPC patients from QEH for training and 183 from Queen Mary Hospital for external validation. Two separate survival models were developed using high-repeatable and low-repeatable RFs exclusively and compared on their prognostic performance in the validation set. In addition to the two technical studies, we developed two quantitative biomarkers based on anatomical and radiomic features for prognosis and treatment efficacy predictions of NPC patients. Based on the same NPC cohorts, we identified independent prognostic factors from anatomical features of lymph node tumor and constructed a prognostic index with N stage. In the last study, we identified single predictive radiomic feature extracted from primary gross tumor volume for patients receiving concurrent chemoradiotherapy with/without addition of adjuvant chemotherapy (ACT). We further investigated the predictive value of its voxel-wise feature mapping for feature explanation. We have successfully developed RADAR for efficient, comprehensive, and reliable radiotherapy data analysis for clinical biomarker development, With the help of RADAR, we discovered that more than half of the wavelet-filtered RFs, especially texture features, were highly susceptible to scanning position variations, irrespective of image modalities or HNC subtypes. It was more prominent when a smaller discretization bin number was used. Using high-repeatable RFs for model development yielded a significantly higher concordance-index (0.63) in the validation cohort than when only low-repeatable RFs were used (0.57, p-value= 0.024), suggesting higher model generalizability. For the two developed biomarkers, the anatomy-based prognostic index demonstrated superior cross-institutional performance in disease-free survival (DFS) than the clinical baseline N stage. The predictive radiomic feature, gldm_DependenceVariance in 3mm-sigma LoG filtered image, was discovered, and the high-risk patients who received additional adjuvant chemotherapy achieved a 3-year DFS rate of 90% versus 57% for low-risk patients. The predictive value can also be generalized to the highlighted subvolume of the feature map. In conclusion, RADAR has been demonstrated as a highly useful tool for efficient analysis of radiotherapy data and effective development of biomarkers for precision medicine. We urge caution when handling wavelet-filtered RFs and advise taking initiatives to exclude low-repeatable RFs during feature pre-selection for generalizable model construction. By using the newly designed anatomical features, the spatial characterization of lymph node tumor anatomy improved the existing N-stage in NPC prognosis. The radiomic signature with its voxel-wise mapping could be a reliable and explainable ACT decision-making tool in clinical practice. |
| Rights: | All rights reserved |
| Access: | open access |
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/200/12551