As the technology and research programs keep developing new technologies for image processing, it is essential to integrate these core technologies into a complete pipeline that can manage image input/output, pre- and post-processing, user interfaces, and appropriate computational resources. Prompt deployment is a key for testing, applications, and obtain feedback from application scientists. We have designed a platform to efficiently develop new pipelines based on Python, which digests xml-based pipeline design and makes the development of highly complex pipelines efficient, readable, and reliable. We have also developed a cloud-based architecture (MRIcloud.org, see “Software and Databases” page) that offers users an instant access to new technologies, as well as supercomputing resources.
(i) Development and deployment of new pipelines: The figure below shows an example of deployment of a new pediatric brain atlas library with detailed cerebellar parcellation and a segmentation tool. As a new atlas library is developed, they can be integrated into a segmentation pipeline and quickly deployed in the cloud interface as. We are now deploying the first pipelines for cerebrovascular reactivity (CVR) mapping, and cerebrovascular reactivity measurement using resting-state data (RS-CVR). We also have pipelines for quantitative magnetic susceptibility mapping (QSM). A new family of deep-learning based segmentation and lesion detection tools will also be deployed in the new cloud platform.
(ii) Development of a new platform to perform research using clinical data Due to the advancement of machine learning technologies in recent years, the demands for large data have increased drastically. Consequently, more and more researchers are interested in clinical data, which can readily offer tens of thousands of datasets. On the other hand, the regulations about data handling are becoming stricter and it is becoming increasingly difficult to move clinical data outside the clinical firewall. To address this, we are creating a platform inside the clinical data center that can process data and create knowledge without taking the raw data outside the firewall. Currently 4 studies have been approved by IRB based on this new platform, which is now considered as an example of “Good Practice” by Johns Hopkins University.
(iii) Delivering new biomarkers to bedside The deployment of an image analysis platform within the clinical data center enables us to integrate new pipelines for MRI-based biomarkers into clinical workflows. Our prototype is now installed in a mock PACS system, in which physicians can simply select a new biomarker service in the PACS interface and the result automatically returns to the interface.