A no-gold-standard framework to objectively evaluate quantitative imaging methods with patient data

点击翻译按钮获取中文摘要

Project Summary Quantitative imaging, where a numerical/statistical feature is computed from a patient image, is emerging as an important tool for diagnosis and therapy planning. Several new and improved quantitative imaging (QI) methods, which include reconstruction, analysis, and estimation methods are thus being developed. There is an important and timely need to optimize the QI methods on the underlying clinical quantitative task, as sub-optimal methods would yield quantitative values that are unreliable, and thus have limited clinical value. Performing this evaluation with patient imaging data is highly desirable, but the unreliability or unavailability of a gold standard for most patient studies makes evaluation impractical or impossible. To enable evaluation of imaging methods with patient data, several no-gold-standard evaluation (NGSE) techniques have been developed, but mostly in the context of detection tasks. More recently, similar NGSE techniques for quantitative tasks have been developed by us and others. We have demonstrated the efficacy of our NGSE technique in ranking segmentation methods for diffusion MR and reconstruction methods for quantitative SPECT. Our goal in this project is to take steps towards translating this mathematical concept to a clinical tool. Existing NGSE techniques make assumptions that may not hold in several QI applications, require large amounts of patient images that are often unavailable, and have been validated using only computational studies. To address these issues, we propose to develop and comprehensively validate a novel generalized Bayesian NGSE framework. This framework will be a generalized Bayesian approach that will reflect clinical scenarios accurately and not require multiple patient studies. The framework will be validated using new anthropomorphic physical phantom and patient data in addition to realistic and validated simulation studies. For clinical translation, it is also necessary to demonstrate the efficacy of the framework in answering an important clinical question. The clinical question we choose is that of using the NGSE framework to determine the optimal segmentation method to compute volumetric features from PET for early prediction of therapy response in patients with non-small cell lung cancer (NSCLC). Answering this question will help address a critical, urgent and unmet need for strategies to personalize the treatment of NSCLC, a disease with high morbidity and mortality rates. The proposed NGSE framework is well poised to accelerate the clinical translation of new and improved QI methods by enabling their evaluation with patient data. The framework will have multiple high-impact applications such as in determining the optimal QI method for measuring biomarkers to monitor cancer-treatment response, diagnose cardiac/neurodegenerative diseases, and conduct imaging- based dosimetry. Thus, developing this NGSE framework has the potential to significantly impact QI-based clinical decision making.