Multimodal radiomics using MRI and amino acid PET in neuro-oncology

点击翻译按钮获取中文摘要

Metabolic imaging techniques such as positron emission tomography (PET) characterize specific metabolic and cellular features of gliomas which may provide clinically relevant information beyond structural magnetic resonance imaging (MRI). In the last years, PET using radiolabeled amino acids has become an important diagnostic tool to overcome some of the shortcomings of conventional MRI. The value of radiomics has been assessed for various clinical applications in patients with brain tumors such as the prediction of response to neuro-oncological treatment or the noninvasive assessment of important molecular markers. At present, most of these studies are based on MRI alone. In contrast, amino acid PET radiomics studies remain scarce. Nevertheless, several studies in the last years demonstrated that amino acid PET radiomics provides valuable additional diagnostic information. The first goal of the planned work program is to develop and evaluate an automated and objective assessment of treatment response based on amino acid PET/MRI and radiomics in patients with brain tumors. Currently, response assessment in brain tumor patients is mainly based on the manual evaluation of contrast-enhanced MRI. The addition of amino acid PET seems especially helpful in clinically challenging situations such as the evaluation of non-enhancing lesions as well as the differentiation of treatment-related changes from tumor progression. So far, the potential of integrating radiomics into response assessment has not been investigated. Therefore, a fully automated and objective assessment of treatment response including amino acid PET/MRI and radiomics would be of high clinical relevance. The second goal of the proposed translational research project is to establish the first open database of amino acid PET/MRI of patients with brain tumors. The performance of radiomics benefits from large amounts of imaging data, histomolecular, and clinical information. In Neuro-Oncology, the number of available datasets is low, which makes the development of robust and reliable diagnostic, predictive and prognostic radiomics models especially challenging. A common way to overcome this limitation is to use open databases such as The Cancer Imaging Archive (TCIA). Up to now, such repository does not yet exist for amino acid PET, especially not for hybrid PET/MRI. The third focus of our project is to correlate local radiomics features with the underlying tumor biology. Here, local histomolecular data from image-guided stereotactic serial biopsies will be spatially linked with preoperative amino acid PET/MRI data and local radiomics feature maps. In case meaningful correlations can be assessed, the results would provide further insights into biological meaning and potentially aid towards a broader acceptance and a successful clinical translation of radiomics.