@article {Zhang1280, author = {Jun Zhang and Richard Jacko and Trung Vu and David Poon and Preethi Subramanian and Timothy Sbory and Shivangi Vora and Sydney Cearlock and Max Melchioris and Elizabeth Barmash and Matthew Orabella and Chris Markeson and Michael Knopp}, title = {Artificial Intelligence Integration in Cloud-based Real-time Data Quality Assurance for Multi-Institutional Clinical Trials}, volume = {59}, number = {supplement 1}, pages = {1280--1280}, year = {2018}, publisher = {Society of Nuclear Medicine}, abstract = {1280Objectives: To introduce why, what and how to integrate artificial intelligence (AI) approaches into cloud-based data quality assurance (QA) for multi-institutional imaging clinical trials to advance the workflow and turnaround time for QA feedback. To integrate AI technology into the next generation workflow for the NCI NCTN Imaging and Radiation Oncology Core (IROC) that we serve. Methods: Over 10 years of multi-institutional medical imaging clinical trial data at IROC Ohio (previous CALGB/Alliance/SWOG Imaging Corelab) are being used to develop and train AI approaches. Dedicated medical imaging and protocol driven QC requirements were structured into algorithms. Imaging modality adapted QC programs were developed using in-house software tools. Subsequently, a new concept of virtualized, cloud and machine deep learning based AI driven QA methodology was introduced to leverage the extensive computational capabilities of today{\textquoteright}s virtualized environments to enable automatization for real time DICOM header and image based QA assessment and reporting. Results: QA data and process workflow from 12 imaging clinical trials with different nuclear medicine and PET / CT imaging modalities and different imaging protocols were utilized and summarized. A protocol adaptive QA matrix methodology was demonstrated and combined with the previously established QA heatmapping data compliance approach (green (within range), yellow (beyond range but acceptable) and red (out of range, not acceptable). Using predefined master templates consisting of different QA modules enabled a flexible adjustment and management of protocol requirements. The need of an imaging QA designer became apparent who would implement the adaptive QA matrix for a specific trial, validate the performance of the AI based protocol QA implementation and on-going review potential learning based add-on implementations during live clinical trials. The AI driven, adaptive QA workflow could substantial improve QA accuracy and turnaround time to real time by reducing user effort and enabling direct feedback to performing sites and trial management, however also requires new expertise in designing and overseeing the AI driven workflow. Conclusions: Introducing AI based QA processes based on deep learning of prior clinical trials appears to be a feasible approach to advance QA performance in multi-institutional clinical trials. However, new processes and skills are required to implement an AI driven QA approach for a specific trial. Our current observation is that AI and big data driven approaches will be able to even further improve QA performance and turnaround times, but may require equivalent or even more staffing effort in setup and maintenance making it more practical for larger trials.}, issn = {0161-5505}, URL = {https://jnm.snmjournals.org/content/59/supplement_1/1280}, eprint = {https://jnm.snmjournals.org/content}, journal = {Journal of Nuclear Medicine} }