AI Solutions in Healthcare for Medical Imaging

Many cancers start with changes so small that no human can detect them, even with current medical imaging technology. However, AI programs can be trained with deep learning to see the very earliest changes in cell structure that typically develop into cancerous cells. Advances in medical imaging technologies, including 3D and 4D capabilities, real-time analytics, and GPU-accelerated processing, give radiologists powerful tools to make faster and more accurate diagnoses and recommendations for care. In addition, AI can also help prevent radiologist burnout.

 

The hippocampus is a major component of the human brain. It plays an important role in the consolidation of information from short-term memory to long-term memory, and in spatial memory that enables navigation. In Alzheimer’s disease and other forms of dementia, the hippocampus is one of the first regions of the brain to suffer damage. Accurate identification of the hippocampus in magnetic resonance imaging (MRI) is an important step in the process of diagnosis, but this task of segmenting two neighboring small structures with high precision can be complex for radiologists and other doctors. Deep learning models can perform this work much faster and with higher accuracy, allowing medical practitioners to spend more time on patient diagnosis and care and less time examining images.

 

The data types in healthcare workloads vary—for example electronic health records, ultrasound, computed tomography (CT), MRI, and more. All this data contributes to different aspects of the healthcare services such as medical imaging, digital pathology, genomics, and other use cases. Model training requirements vary for distinct data types, and the goal is always to saturate the GPUs and provide the highest throughput at the lowest latency from the data storage.

 

In collaboration with NVIDIA, NetApp published a technical report titled “NetApp ONTAP AI Reference Architecture for Healthcare: Diagnostic Imaging.” The report offers directional guidance to healthcare providers to fast-track AI infrastructure initiatives specifically for diagnostic imaging practices in hospitals. It includes information about the high-level workflows used in developing deep learning (DL) data pipeline models for medical imaging, validation test cases and results, and sizing recommendations for deployments. The solution was validated with one NetApp® AFF A800, one NVIDIA DGX-2 system, and two Cisco Nexus 3232C 100Gb Ethernet (100GbE) switches.

 

NVIDIA Clara is a computational platform that enables developers to build, manage, and deploy intelligent medical imaging workflows. NVIDIA Clara Train SDK​ offers state-of-the art tools and technologies. This validation leverages Clara’s AI-assisted annotation to label a publicly available brain imaging dataset and train a hippocampus segmentation model based on ResNet-50 and AH-Net architectures.

 

For more information about NetApp AI solutions for healthcare, visit this page.

 

 

Mike McNamara

Mike McNamara is a senior leader of product and solution marketing at NetApp with 25 years of data management and data storage marketing experience. Before joining NetApp over 10 years ago, Mike worked at Adaptec, EMC and HP. Mike was a key team leader driving the launch of the industry’s first cloud-connected AI/ML solution (NetApp), unified scale-out and hybrid cloud storage system and software (NetApp), iSCSI and SAS storage system and software (Adaptec), and Fibre Channel storage system (EMC CLARiiON). In addition to his past role as marketing chairperson for the Fibre Channel Industry Association, he is a member of the Ethernet Technology Summit Conference Advisory Board, a member of the Ethernet Alliance, a regular contributor to industry journals, and a frequent speaker at events. Mike also published a book through FriesenPress titled "Scale-Out Storage - The Next Frontier in Enterprise Data Management", and was listed as a top 50 B2B product marketer to watch by Kapos.

Add comment