This article introduces mdx, a platform designed for supporting data science. mdx is a computing cluster that officially started operation in May 2023. mdx uses virtualization technology, allowing multiple virtual clusters on top of a physical cluster. As of July 2023, thereʼre more than 100 research projects using mdx. This article introduces mdx features and an example of its use in a research project.
Analysis of a large amount of medical image data collected from a nationwide scale is an issue in the progress of ICT use in the medical field. This paper presents the medical image bigdata cloud platform, which is developed and operated by National Institute of Informatics to solve this issue. The cloud platform collects medical image data from all over the country every day. More than 420 million images have been accumulated so far and used for researches of AI image analysis. Stable development and operation of the platform that collects, accumulates, and analyzes medical image data enable not only research and development of new technology using past data, but also quickly starting research and development of medical image processing methodologies in response to emergencies of public health. This paper also introduces one example, CT image analysis for COVID-19 pneumonia, utilizing the cloud platform.
With the development of medical imaging technology, various techniques have been developed and used to visually understand the inside of the living body. However, these technologies can only directly obtain images and videos, and diagnosis is still performed by human hands, such as physicians. There are great expectations for software that can reduce such labor, and an increasing number of technologies are already being used in the field of medicine, but the target is limited because they require knowledge and skills in both medicine (medical imaging) and computing technology. Therefore, in this study, researchers in the medical imaging and high-performance computing fields are collaborating to accelerate and scale up the image reconstruction of PET. This paper describes the details of this effort and the results obtained so far.
In this paper, we introduced the development of computer-aided detection (CAD) software using deep learning (DL) on a supercomputer. We described our DL training environment based on asynchronous parallel execution Bayesian optimization was constructed on a supercomputer. As an actual example of the DL learning environment, we indicate an example of a hyperparameter search for lung nodule detection in chest CT images using 3D U-Net. The constructed environment enabled us to train deep learning with hyperparameter tuning in a short time.
Deep learning has been applied to medical image processing as well as image reconstruction. In this paper, we focus on PET (Positron emission tomography) image reconstruction using deep learning, discussing the history and state-of-the-art of these techniques and future perspectives.