AI for rare disease diagnosis with NVIDIA DGX
Thanks to the collaboration among the Friuli Centrale Health Authority, the University of Udine, AREA Science Park and SISSA – International School for Advanced Studies, the MIRA project (“Malattie Rare e Intelligenza Artificiale” o “Medical Innovation for Rare using Artificial Intelligence”) was launched with the aim of using AI to improve the timing of diagnosis, treatment and management of rare diseases.
CUSTOMER NEEDS
The diagnosis, treatment and management of rare diseases suffer from significant diagnostic delays (7-10 years on average) and there is no treatment available in 95% of cases.
As a contribution to improving this unacceptable situation in 2025, the MIRA project was created with the aim of identifying patients at risk of rare diseases by analysing clinical data and images using artificial intelligence algorithms. The project also aims to develop new data analysis technologies (natural language processing and machine learning) to extract data in standard terminology in full compliance with privacy regulations.
One of the objectives was to enhance the computational capabilities of the region by providing an infrastructure capable of supporting advanced activities in the field of artificial intelligence applied to health, with particular reference to the early diagnosis of rare genetic diseases.
A proven architecture for large-scale artificial intelligence, based on NVIDIA DGX BasePOD, designed to simplify the management of the most complex computational workloads.
From architectural design to component selection, from hardware/software integration to final commissioning of the solution, beanTech has accompanied the promoters of the MIRA project through all stages of infrastructure implementation.
This is housed within the “Ancona Hydroelectric Power Plant” in Udine, in a technical space provided by CAFC S.p.A. and equipped to ensure reliability, dedicated air conditioning and operational continuity. The entire system is powered by the hydroelectric turbine connected to the Ledra canal, with the possibility of backup via the local distributor’s network.
The NVIDIA DGX BasePOD platform represents a tested architectural model for AI computing applied to science: a combination of specialized hardware, enterprise-grade software and optimized design, enabling research teams to focus on scientific activities while reducing infrastructural complexity.
Configuration includes:
- 2 NVIDIA DGX H100 systems, each with 8 NVIDIA Hopper GPUs, engineered for high-performance AI workloads;
- NVIDIA Quantum 2 InfiniBand QM9700 and NVIDIA Spectrum SN2100 Ethernet networking for low-latency, high-bandwidth connectivity;
- NVIDIA Base Command software with AI Enterprise, for orchestrating workloads, monitoring the infrastructure, and accessing pre-trained AI models.
Thanks to this infrastructure, project partners now have:
A shared, high-performance system for AI applied to genetics and biomedical diagnostics.
A sustainable infrastructure powered by renewable energy.
A scalable foundation for future scientific projects and new users.
A virtuous model of collaboration between public institutions and local entities, replicable in other contexts.
Sustainable infrastructure in the heart of Friuli
Discover the project that combines technology, strategic vision and promotion of the region, strengthening Friuli-Venezia Giulia's position as a benchmark for scientific innovation and sustainability applied to high-performance computing.
There are patients aged 50 or 60 who have received incorrect diagnoses. The aim, with this methodology, is to accelerate the diagnostic and treatment process. It will assist physicians, without replacing healthcare professionals.
The acquisition of the computing platform installed at the Ancona Hydroelectric Plant aims to create a specialized center for the development and execution of Machine and Deep Learning algorithms, capable of enabling real-time processing of complex machine learning algorithms.