No one is born knowing the names of all the objects around them. We learn them as parents, older siblings and others show us the object and say its name, or point to things in picture books and repeat the word that matches it. We learn words like cup, ball, shoe, bunny, blanket for items in our daily lives. We learn parts of our faces: Where’s your nose? Where’s your chin? All these labels, and more, are eagerly grasped by our brains in a seemingly unquenchable thirst for learning.
Artificial Intelligence (AI) is modeled on the human brain and how it learns. It isn’t born knowing. It must be trained by humans in order to recognize, reason logically, and anticipate or predict outcomes.
When properly trained AI is placed at the service of radiology, it can expedite disease detection and diagnosis, adding speed and accuracy to the interpretation of clinical photos (as in skin lesions) and scans performed on MRI, ultrasound, X-rays, CT, and PET/CT. Precision labeling is, of course, the key. AI is only as accurate as the dataset it’s trained on. As the saying goes, “garbage in, garbage out.”
A creative approach to training
A company called Centaur Labs has developed a creative way to label 150,000 training images per day. They utilize “a network of ‘tens of thousands’ of medical students across 140 countries.”[i] Medical students who enroll as labelers have a smartphone app by which they receive images to label. For example, they might receive a photo of what appears to be a skin cancer. They select from a multiple-choice list of diagnostic labels, and submit their response. Although they’re basically volunteering their time, it’s competitive and offers incentives. The app—which also uses AI—”judges these users on their work, rewarding those who excel with cash prizes, while also collecting opinions on each case.”[ii]
This is a process of collective human intelligence, a sort of group mind which Centaur’s website tells us can combine “multiple opinions into a more accurate result.” Collective intelligence is powerful, since the whole is greater than the sum of the parts. In fact, Mother Nature has endowed species like insects, birds and fish with group-mind abilities. Take the case of a swarm of honeybees that must move into a new hive. Scientists call the swarm’s discernment and decision-making activity the hive mind, and it’s analogous to the activity of our own brains by which a “committee” of neurons achieves consensus and eliminates the outliers.
Centaur Labs developed a brilliant way to amass brainpower, tapping into an enormous medical group mind at the service of image labeling that’s practically free (except for the cash prizes). According to Centaur’s CEO, Erik Duhaime, “AI learns like humans—by example—and to train an algorithm it takes thousands or even millions of examples. It is difficult to curate large medical datasets, and nearly impossible to source accurate labels from those with medical knowledge and specialized training.” It appears they have accomplished a terrific way around the obstacles. Their website claims, “Using collective intelligence and performance-based incentives, we’re more accurate than medical experts, and more scalable – we label tens of thousands of cases every day.”
In our practice, we have already been employing the benefits of AI in our prostate imaging. We were early adopters, but we’re now joined by an increasing number of radiology centers. As the demand for accurate AI in radiology grows, it raises the problem of supply. We can expect more companies like Centaur Labs to generate solutions. Seen from a distance, it’s starting to look like the entire medical world is becoming one giant group mind.
NOTE: This content is solely for purposes of information and does not substitute for diagnostic or medical advice. Talk to your doctor if you are experiencing pelvic pain, or have any other health concerns or questions of a personal medical nature.
References
[i] Stempniak, Marty. “Lab startup specializing in labeling medical images for artificial intelligence raises $15M.” Radiology Business, Sep. 3, 2021. https://www.radiologybusiness.com/topics/artificial-intelligence/centaur-labs startup-labeling-medical-images-ai?utm_source=newsletter&utm_medium=rb_news
[ii] Ibid.
About Dr. Dan Sperling
Dan Sperling, MD, DABR, is a board certified radiologist who is globally recognized as a leader in multiparametric MRI for the detection and diagnosis of a range of disease conditions. As Medical Director of the Sperling Prostate Center, Sperling Medical Group and Sperling Neurosurgery Associates, he and his team are on the leading edge of significant change in medical practice. He is the co-author of the new patient book Redefining Prostate Cancer, and is a contributing author on over 25 published studies. For more information, contact the Sperling Prostate Center.
