How Northwestern Medicine and Dell are bringing AI to medical imaging

Dell and Northwestern Medicine have stepped up their partnership to improve healthcare through the use of AI. This includes the creation and ongoing evaluation of a generative large language model (LLM) that can analyse patients’ chest X-rays.

Northwestern Medicine, which runs a number of hospitals in the Chicago area, started evaluating the use of generative AI in one of its emergency departments last year. It published a study that compared 500 AI-generated reports of chest X-rays with reports on the same images written by human radiologists based at the hospital. 

Five senior emergency physicians then rated the reports. The result? “AI and radiologist reports were not significantly different… the generative AI model produced reports of similar clinical accuracy and textual quality to radiologist reports,” the study’s authors wrote.

Northwestern’s chest X-ray model was trained in three weeks, using around one million chest X-rays. We also know the hardware platform uses Dell PowerEdge XE9680 servers with eight Nvidia H100 Tensor Core GPUs.

Draft reports?

Dell notes the LLM has been used to produce “draft x-ray reports”, which suggests a human radiologist is still expected to review the AI’s output.

Chest X-ray reports are used to identify pathologies, such as pneumonia or a collapsed lung, and the treatment a patient receives is likely to be governed at least in part by what the X-ray findings show. 

Northwestern and Dell don’t note whether doctors will be given access to the reports once they are generated by AI, which would happen almost immediately, or once they’ve been vetted by radiologists – which is likely to take far longer due to the sheer volume of imaging work they handle.

Recommended reading: Top companies disrupting healthcare with artificial intelligence

Not the only AI medical imaging project

Dell and Northwestern are now working on further AI projects together, including a “significantly improved” version of the radiology LLM. The healthcare provider has previously spoken of training the model to interpret other types of images, such as ultrasounds, CT and MRI scans. They are also building a predictive model for electronic medical records.

Northwestern Medicine has undertaken a number of other experimental AI projects in the past.

Last year, it announced it was working on an algorithm to detect early signs of heart failure by scanning both the structured and unstructured data in a patient’s record for symptoms and risk factors that can be suggestive of the disease. 22,000 patients were flagged by the system; the cases were then reviewed by clinicians, and around one-tenth of those cases were then followed up.

It has also partnered with Google in the past to develop an AI that could be used to spot signs of breast cancer in mammograms.

Dell and Northwestern Medicine AI partnership

Both Dell and Northwestern are enthusiastic about the ongoing partnership.

“Technology such as AI has the power to speed innovation that advances human progress,” said Jeff Boudreau, Chief AI Officer at Dell Technologies. “Healthcare is a prime example of where technology can make an impact and help save lives.”

He added: “By combining Dell’s technology with Northwestern Medicine’s expertise, we’re setting a new standard for AI-driven healthcare solutions.”

Dr Mozziyar Etemadi, Medical Director of Advanced Technologies at Northwestern Medicine was equally effusive. “When we think about what AI can do, we don’t just see the technology itself, we see the many patients and lives it will positively impact,” he said.

Worth a read

• Intel sets out its AI stall for enterprises at Intel Vision 2024: trust us with your data (not that other lot)
• Miao Luo, Director of Technology Strategy at Qt: “Your customer’s convenience is your gain”
• Intel claims Gaudi 3 AI accelerator is typically 40% faster than Nvidia H100