Image analysis

Daher said medical image analysis is "really creating the excitement right now" and has been the "first and foremost" use case for AI. AI in medical imaging results in automation of manual tasks, as well as acceleration and augmentation of image analysis, Daher explained.

I think we all agree medical imaging AI is here. It's no longer a question of whether it's going to happen; it is happening.

"It's essentially applying all kinds of image analysis tools to our medical images to do anything from automating the heavy manual-intensive tasks that today are done with eye-balling or a lot of clicking and a lot of time from expensive resources, such as radiologists," Daher said. The application of AI in this scenario, Daher said, accelerates "the way we analyze and interpret images" and augments the analysis and quantification of images.

An analysis of 93 medical imaging AI companies revealed the largest three image modality focuses were computed tomography, MRI and X-ray imaging. Other companies focused on image modalities such as mammography, 3D breast tomosynthesis and ultrasound echocardiography for cardiac conditions. Another area that has gained prominence is fundus imaging due mainly to interest from companies like Google and Amazon, Shah said.

In terms of organ imaging, an analysis of 88 medical imaging AI companies revealed focus on a single organ, which Shah said is "reflective of a cautious approach to gain sufficient expertise in one area first." Shah said the brain is the most targeted organ, followed by the lungs and breast, but the heart is a close fourth.

"In the coming few years, we expect [the heart] to leapfrog over the others to gain a higher spot in this ranking owing to an increasing number of companies looking at developing solutions for the cardiac area," Shah said.

Cognitive computing

Though at a less mature stage than medical image analysis, Daher said cognitive computing is another progressive opportunity for AI in medical imaging. AI use cases, or smart medical imaging processes, are starting to develop across every step of the image workflow, from ordering and scheduling an image to interpreting and following up, he said.

Daher noted that clinical guidelines for the imaging process that take years to be curated and validated are "set to be disrupted by AI."

"Suddenly, we have a view of many exams and ultimate outcomes of the patient that underwent this exam, and we can start using this to develop new appropriateness in ordering guidelines," Daher said.

Embedded AI

AI in medical imaging machines is also in the early stages, but Daher said "we can safely say today that AI, deep learning especially, is going to be the driver of the development of the next generation of medical imaging equipment. It is really setting out to address the big technology limitations of our imaging modalities today."

For MRI images, a major challenge is the time it takes to process the data, Daher said. Early moves applying deep learning to image reconstruction in MRI is where stakeholders can start seeing the next generation of MRI "getting us a step closer to the holy grail of real-time MRI in 10 or 20 years."

"But we're not there yet," Daher said. "We think that smart medical imaging machines in the short- to mid-terms are going to do a lot of smart things that we're starting to notice today with applying AI at the edge." 

Daher said he is starting to see features of smart equipment, such as an AI algorithm that can spot hemorrhaging in the brain, being implemented in CT machines on board ambulances. Using the preliminary image findings can allow paramedics to transport a patient to the right place and "save precious minutes."