Shared Resource Core 2: Clinical Artificial Intelligence Core

点击翻译按钮获取中文摘要

PROJECT SUMMARY Artificial intelligence (AI) algorithms have the potential to fundamentally change medicine through their ability to recognize complex patterns in medical data. The Clinical Artificial Intelligence and Imaging Core (AI Core) is an essential shared resource that will support the Aims of the Harvard/UCSF ROBIN Research Projects to enable large-scale analysis of granular clinical data, allowing non-invasive characterization of tumoral and patient heterogeneity and a path towards clinical translation. This will be achieved through the following Specific Aims: i) retrieve, curate, and annotate digitized clinical data to support quantitative analyses and AI/informatics pipelines for the ROBIN Molecular Characterization Trial and Research Projects, which will produce one of the most comprehensive datasets for DMG and neuroblastoma patients in existence for AI- based data analysis, ii) develop and evaluate task-specific AI pipelines using our well-established data preprocessing, AI-derived imaging biomarkers, and natural language processing (NLP) platforms for tumor heterogeneity, radiation resistance/response, and toxicity characterization in accordance with the Research Projects and Data Science Core, and iii) standardize and release AI/informatics methods across data types and applications in ways that ensure transparency, reproducibility, and access to advance scientific knowledge within the wider research field, as well as accelerate clinical translation to the pediatric radiation oncology clinic. Achieving these aims will be possible through synergy with the molecular mechanistic analyses in the Data Science Core, as well as with the ROBIN-NEST Cross-Training Core and Administrative Core to disseminate our methods and provide training to the greater ROBIN Network and the scientific community. This Core is led by pioneers in the field of AI analysis of medical imaging (PI: Aerts) and clinical text (PI: Savova), with significant experience building open access platforms for medical AI applications. For imaging analysis, we developed and maintain PyRadiomics, one of the world’s most widely used and highly cited radiomics pipelines, developed with support of NCI’s investments in infrastructure and data, including the Informatics Technology for Cancer Research (ITCR), Imaging Data Commons (IDC), and Quantitative Imaging Network (QIN) programs. For clinical text, we have developed Apache cTakes(™), a leading open access natural language processing platform for extracting medical, grammatical, and semantic information from clinical texts, and DeepPhe, an open-source software for cancer clinical phenotyping, also supported by the NCI’s ITCR program (PI: Savova). We will use and build on our open access methods and state-of-the art AI- based phenotyping methods developed in these NCI projects to support the Harvard/UCSF ROBIN investigators to incorporate fundamental clinical -omics data into their investigation of intratumoral heterogeneity and predictors of radiation response and late effects.