Radiogenomics framework for non-invasive personalized medicine

点击翻译按钮获取中文摘要

Project summary Radiogenomics, is a burgeoning area of research that aims to link medical imaging with multi-omics molecular profiles of the same patients. Radiogenoimcs has shown its potential through its ability to predict clinical outcomes e.g. prognosis, and through predicting actionable molecular properties of tumors, e.g. the activity of EGFR, a major drug target in many cancers. The typical imaging genomics workflow consists of the following steps: (1) Identify the tumor through segmentation. This is often also defined as identifying Regions Of Interests or ROIs through a manual process with a radiologist or using computer vision algorithms. (2) Feature extraction, often also known as radiomics, whereby 100s of features are identified that capture the shape, the texture and the intensity distributions of lesions in 2D or 3D. (3) Supervised machine learning to predict clinical outcomes such as prognosis, overall survival or response to treatment, or predicting molecular profiles such as gene expression patterns or metagenes, or individual molecular properties such as the mutation status of a gene (e.g. EGFR). This workflow has been demonstrated in several cancers including lung cancer, brain tumors, hepatocellular carcinoma, breast cancer etc. Current radiogenomics applications are limited to study associations between imaging and molecular data, and predicting long term outcomes. However, no actionable information is gained from radiogenomics maps. In this renewal, we propose to develop a radiogenomics framework to support treatment response, treatment allocation and treatment monitoring: (1) we will develop informatics algorithms that integrate radiogenomic data for treatment response, (2) algorithms that allow combining radiogenomic data during treatment follow-up, and (3) algorithms that use the radiogenomic map to suggest novel drugs and predict drug target activities. Combining these complementary data sources in a radiogenomics framework for data fusion can have profound contributions toward predicting treatment outcomes by uncovering unknown synergies and relationships. More specifically, developing computational models integrating quantitative image features and molecular data to develop radiogenomics signatures, holds the potential to translate in benefit to tumor patients by investigating biomarkers that accurately predict therapy response of tumors. Readily, because medical imaging is part of the routine diagnostic work-up of cancer patients and molecular data of human tumors is increasingly being used in clinical workflows, therefore if reliable radiogenomic signatures can be found reflecting treatment response, translation to the clinical applications is feasible.