Quantitative, multimodal, imaging-based assessment of hypoxia in Non-Small Cell Lung Cancer

肺癌是英国癌症死亡的主要原因，非小细胞肺癌(NSCLC)是最常见的肺癌类型。肺癌肿瘤中的缺氧，即缺氧，已被证明会导致更差的患者预后，缺氧肿瘤需要的放射治疗时间是非缺氧肿瘤的三倍。牛津大学最近的一项研究表明，抗疟疾药物阿托瓦酮在减少非小细胞肺癌患者肿瘤缺氧方面有效。该药物已经获得批准，因此很容易进入临床实践。目前使用[18F]-氟咪唑正电子发射断层扫描和计算机断层扫描(FMISO-PET/CT)来测量缺氧，其中肿瘤和降主动脉由专家临床医生手动分割以计算缺氧体积。这是一个耗时的过程，自动化这个过程将加快临床试验期间测量缺氧的时间。FMISO PET扫描也很昂贵，仅用于研究，因此，为了识别癌症过程中的缺氧肿瘤患者，需要从常规收集的或更便宜的模型中测量缺氧。在这个项目中，我们的目标是使用多模式成像数据自动测量肺癌肿瘤中的缺氧过程。利用人工智能(AI)包括深度学习技术，将开发从FMISO-PET/CT扫描等多模式成像数据中分割肿瘤的新方法。此外，还将研究深度放射组学，以提取可以测量肿瘤体积内缺氧的特征。然后，深度学习和深度放射组学都将被用来看看是否可以从其他常规收集的或更便宜的方式中找到任何低氧特征，这些方式可以很容易地整合到患者的癌症路径中。深度学习在许多医学成像任务中都取得了成功，包括图像分割。放射组学是从放射图像中提取纹理特征的一个活跃的研究领域，最近的研究表明，卷积神经网络可以捕获纹理信息，因此可以开发先进的深层放射组学技术来从不同模式提取低氧特征。该项目本质上是高度合作的风险和合作，包括牛津大学肿瘤科和牛津大学医院NHS基金会信托基金。这项合作将提供对阿托瓦酮作为肿瘤缺氧修改剂(ATOM)数据集的访问，该数据集包含非小细胞肺癌患者，每个患者都有许多成像和非成像模式。总之，自动化从FMISO-PET/CT扫描测量缺氧的耗时任务将加快这一过程，并允许更多患者纳入未来的临床试验。此外，能够从常规收集的或更便宜的方式测量低氧，将允许将低氧测量整合到患者的癌症途径中。这意味着可以为缺氧肿瘤患者提供替代治疗，以改善患者的预后。该项目属于EPSRC医疗保健技术研究主题，以及医学成像和人工智能技术研究领域。

Lung Cancer is the leading cause of cancer death in the UK and non-small cell lung cancer (NSCLC) is the most common type of lung cancer. Hypoxia, a lack of oxygen, in lung cancer tumours has been shown to lead to worse patient outcomes and hypoxic tumours require up to three times as much radiation therapy as non-hypoxic tumours. A recent study at the University of Oxford showed that the anti-malarial drug atovaquone is effective at reducing hypoxia in tumours of non-small cell lung cancer patients. The drug is already approved, and it would therefore be easy to adopt into clinical practice. Currently hypoxia is measured using [18F]-fluoromisonidazole Positron Emission Tomography and Computed Tomography (FMISO-PET/CT) scans, where the tumour and descending aorta are segmented manually by an expert clinician to calculate the hypoxic volume. This is a time-consuming process and automating the process would speed up the time taken to measure hypoxia during clinical trials. The FMISO PET scans are also expensive and only used in research, therefore, to identify patients with hypoxic tumours during their cancer pathway, hypoxia needs to be measured from routinely collected or cheaper modalities.In this project we aim to automate the process of measuring hypoxia in lung cancer tumours using multimodal imaging data. Using Artificial Intelligence (AI) including deep learning techniques, novel methods for tumour segmentation from multimodal imaging data such as FMISO-PET/CT scans will be developed. Furthermore, deep radiomics will then be investigated to extract features that could measure of hypoxia within the tumour volume. Both deep learning and deep radiomics will then be used to see if any hypoxic features can be found from other modalities that are routinely collected or cheaper that could easily be integrated into a patient's cancer pathway. Deep Learning has been successful in many medical imaging tasks, including image segmentation. Radiomics is an active area of research for extracting textural features from radiological images, and recent studies have shown that convolutional neural networks can capture textural information, therefore advanced deep radiomic techniques could be developed for extracting hypoxic features from different modalities.The project by its nature is high-collaborative venture and collaboration includes Department of Oncology in Oxford, and Oxford University Hospitals NHS Foundation Trust. The collaboration will provide access to the Atovaquone as Tumour HypOxia Modifier (ATOM) dataset, containing non-small cell lung cancer patients, each with many imaging and non-imaging modalities.In summary, automating the time-consuming task of measuring hypoxia from FMISO-PET/CT scans will speed up the process and allow more patients to be included in future clinical trials. Additionally, being able to measure hypoxia from routinely collected or cheaper modalities will allow the measurement of hypoxia to be integrated into the patient cancer pathway. This would mean that patients with hypoxic tumours could be offered alternative treatment to improve patient outcomes.The project falls within the EPSRC Healthcare Technologies research theme and the Medical Imaging and AI Technologies research areas.