Over 90 Total Lots Up For Auction at One Location - WA 04/08

Five key takeaways about AI from RSNA

by John W. Mitchell, Senior Correspondent | November 28, 2018
Artificial Intelligence
In a lively session that featured candid insights from Dr. Paul Chang, vice chair of radiology informatics at the University of Chicago, and two other panelists, an audience of radiologists, data scientists, informatics experts, and vendors learned several fundamental AI truths.

The RSNA session titled “Deep Learning & Machine Intelligence in Radiology” also included featured panelists Dr. Luciano Prevedello, chief of imaging informatics and a neuroradiologist at Wexner Medical Center and Abdul Hamid Halabi, global business development lead for Nvidia.

Here are the five main takeaway points from the discussion:

1 – Here's what artificial intelligence actually means
AI is achieved when a computer mimics the activity of the human brain and eyes, learning not through human coding but rather through vast amounts of data review, according to Prevedello. AI scientists, including radiologists, walk computers through deep learning with progressive iterations. The result is error rates that are increasingly better than what humans can achieve. This makes the AI especially adaptable for finding anomalies, such as cancer lesions, strokes, or fractures sooner. AI also can eliminate variability in diagnosis from one well-trained radiologist to another.

2 – AI is neither new nor spooky
Deep learning that drives AI is a form of linear regression, a statistical tool that has been around for a long time. It doesn’t so much require cleverness as it does a lot of data to teach algorithms. Chang asserted that AI is not, “new or spooky.” However, the trend in AI machine learning is akin to a roller coaster ride and currently we are approaching the top. The climb to the peak is marked by nervousness and even fear – but there is also a lot of excitement. He predicted that many early vendors will not succeed. But he said these failures are necessary for the imaging AI sector to learn from its mistakes and advance.

3 – Taking a lesson from self-driving cars
Halabi said rapid advancements in the hardware used by gamers enabled AI development. He also played a video of an Nvidia self-driving car. It featured the “view” through the computer as it recognized other vehicles (data) and how this data behaved. The video illustrated that while most of what the self-driving car saw was normal traffic, it instantly recognized anomalies – just as imaging AI must.

The goal for Nvidia is to train their self-driving vehicle for the day that two deer jump into the road from opposite directions. The algorithm must navigate such a scenario with a perfect outcome to avoid a crash, property damage and death or injury to the passengers. For imaging, AI must detect even the rarest and most unlikely diseases and injuries.

You Must Be Logged In To Post A Comment