A fully automated PET radiomics framework

摘要 这项提议的总体目标是开发一个全自动化的PET放射组学框架，并评估 由该框架导出的PET放射组学特征(RF)在预测患者治疗反应中的有效性 与III期非小细胞肺癌(NSCLC)。放射组学在衍生生物标记物方面显示出令人兴奋的前景 治疗几种疾病。从正电子发射计算机断层扫描测量和评估放射学特征疗效的可能性 对于早期预测治疗反应是非常有影响的，因为PET探测了 肿瘤，与解剖学变化相比，这种变化表现得更早。然而，PET图像具有 高噪声和有限的分辨率，这导致不准确和不精确的射频测量，然后 临床价值有限。在此之前，我们已经开发了优化定量成像方法和 表明这些指标可以帮助评估更可靠的量化指标，从而提高预测能力 这些指标。在这些过去研究的基础上，通过结合成像物理、统计学 推理理论，深度学习，我们建议开发出准确和精确地估计RFS的方法 来自PET。这些方法将包括一种全自动的PET分割方法，使可靠的 使用一种实用的方法来划定肿瘤边界。接下来，进行非黄金标准(NGS)评估 将开发技术来优化RF量化协议。这项技术将为 精确测量PET图像中的RF值，无需获取地面真实RF值。这些方法将 在测量非小细胞肺癌患者的放射组学特征方面进行严格的验证 结合真实的模拟、物理模型研究和现有的患者数据。然后选择RFS 应用ACRIN 6697纵向临床资料预测疗效的回顾性评价 在III期非小细胞肺癌患者中的试验。为这一项目组建了一支强大的多学科团队， 由一名成像科学家、临床核医学放射科医生、具有以下专业知识的肿瘤内科医生组成 胸部恶性肿瘤的生物标记物发展和非小细胞肺癌生物学，生物统计学家和医学物理学家。 拟议的方法将对PET放射组学产生重大影响，因为它能够测量 精确和准确的RFS，并通过促进PET放射组学的临床翻译。冲击力增强 当我们研究PET RFS对III期NSCLC患者的预测能力时，NSCLC是主要的死亡原因 总体存活率低，需要及时改进个性化治疗方案。 此外，本项目中开发的方法是一般性的和潜在影响的精确医学方法。 对于其他癌症以及PET成像具有临床作用的其他疾病。

Summary The overall goal of this proposal is to develop a fully automated PET radiomics framework and evaluate the efficacy of PET radiomic features (RFs) derived from this framework in predicting therapy response in patients with stage III non-small cell lung cancer (NSCLC). Radiomics is showing exciting promise in deriving biomarkers for several diseases. The potential to measure and evaluate the efficacy of radiomic features derived from PET for early prediction of therapy response is highly impactful since PET probes the functional characteristics of the tumor, where changes are manifested sooner in comparison to anatomical changes. However, PET images have high noise and limited resolution, which leads to inaccurate and imprecise RF measurements that then have limited clinical value. Previously we have developed techniques to optimize quantitative imaging methods and shown that these can help estimate more reliable quantitative metrics leading to better predictive ability with these metrics. Building on these past studies and by combining concepts from imaging physics, statistical inference theory, deep learning, we propose to develop methods that accurately and precisely estimate RFs from PET. These methods will include a fully automated PET segmentation method that will enable reliable delineation of tumor boundaries using a practical approach. Next, a no-gold-standard (NGS) evaluation technique will be developed to optimize RF quantification protocols. This technique will provide a mechanism for precise measurement of RFs from PET images without access to the ground truth RF value. The methods will be rigorously validated in the context of measuring radiomics features in patients with NSCLC using a combination of realistic simulations, physical phantom studies and existing patient data. Select RFs will then be retrospectively evaluated on predicting therapy response using existing data the ACRIN 6697 longitudinal clinical trial in patients with stage III NSCLC. A strong multidisciplinary team has been assembled for this project, consisting of an imaging scientist, clinical nuclear-medicine radiologists, medical oncologist with expertise in biomarker development for thoracic malignancies and biology of NSCLC, biostatistician, and a medical physicist. The proposed methods are poised to have a strong impact on PET radiomics by enabling measurement of precise and accurate RFs, and by facilitating the clinical translation of PET radiomics. The impact is strengthened as we investigate the predictive ability of the PET RFs in patients with stage III NSCLC, a leading cause of death with low overall survival, and with an important and timely need for improved personalized therapy regimens. Further, the methods developed in this project are general and potentially impact precision-medicine approaches for other cancers as well as other diseases where PET imaging has a clinical role.