Integrating Quantitative MRI and Artificial Intelligence to Improve Prostate Cancer Classification

点击翻译按钮获取中文摘要

PROJECT SUMMARY Prostate cancer (PCa) develops in sixteen percent of males and is the second leading cause of cancer-related death in men in the United States. While incidence is high, PCa presents with a wide range of aggressiveness and in many cases does not develop into life-threatening aggressive cancer. Current diagnostic strategies may fail to detect all instances of clinically significant PCa and have limited ability to accurately distinguish clinically significant from indolent PCa due to incomplete and inconsistent information. This not only subjects patients to detrimental co-morbidities including overtreatment and undertreatment, but also exacerbates already significant healthcare costs. Consequently, there is an urgent clinical need to achieve accurate detection and classification of clinically significant PCa and determine the appropriate management strategy. Multi-parametric MRI (mp-MRI), consisting of T2-weighted, diffusion-weighted, and dynamic contrast-enhanced imaging, has emerged as the preferred imaging technique for non-invasive detection and grading of PCa. However, the current standardized scoring system for mp-MRI, Prostate Imaging Reporting and Data System (PI-RADS) v2, has limited ability to distinguish between indolent and clinically significant PCa, with sensitivity and specificity in the range of 60-85%. This suboptimal accuracy and considerable variation in performance is mainly due to the fact that current PI-RADS scoring is based on qualitative analysis and subjective interpretation of mp-MRI, confounded by scanner- and patient-specific variations, including B1+ inhomogeneity, arterial input function, and susceptibility and eddy current effects. This proposal aims to overcome these critical limitations of current mp-MRI by establishing a new MRI-based artificial intelligence based on two synergistic innovations: 1) new quantitative dynamic contrast-enhanced MRI analysis techniques and diffusion-weighted MRI acquisition methods that minimize scanner- and patient-specific variations, and 2) novel multi-class deep learning models that can fully integrate the multi-labeled quantitative mp- MRI information. By leveraging the synergy between existing mp-MRI data and to-be-acquired quantitative mp- MRI data with subsequent mapping of all lesions at whole-mount histopathology, the proposed MRI-based deep learning model will be evaluated for detection and classification of clinically significant PCa, compared with the current standard-of-care, PI-RADS v2. Completion of this project will lead to the creation, clinical deployment, and pivotal validation of a new MRI-based artificial intelligence that achieves unprecedented accuracy for detection and classification of clinically significant PCa, thereby increasing confidence in separating indolent PCa from significant PCa and reducing unnecessary biopsies, undertreatment, and overtreatment.