These are the days of miracle and wonder…
The way the camera follows us in slo-mo
The way we look to us all
from “The Boy in the Bubble” by Paul Simon
The term artificial intelligence or AI is at once exciting, mysterious, and even a little scary. As each of us goes through our day, we probably have no idea how much of our activities are touched by AI. Examples include your smartphone calculating the speed of your commute to work, the spam filter in your email, and online shopping recommendations that pop up when you buy a particular nutritional supplement on Amazon.
Then there’s machine learning or ML, which is one form of AI. According to Wikipedia, this is “a field of computer science that uses statistical techniques to give computer systems the ability to ‘learn’ (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed.” The key word is data. The more data provided, the more accurate the results.
This brings me to deep learning, a branch of ML that “trains a computer to perform human-like tasks, such as recognizing speech, identifying images or making predictions. Instead of organizing data to run through predefined equations, deep learning sets up basic parameters about the data and trains the computer to learn on its own by recognizing patterns using many layers of processing.”[i] Deep learning draws upon biological models such as the way nerve cells in our brains network sensory input, memory and information flow.
Similar to a human brain, deep learning is able to use examples and prior knowledge to build on itself. The implications of this are beyond astonishing, and there are already applications in finance, art, automotive technology, and medical science.
Deep learning in diagnosis
One of the most promising areas for deep learning is medicine, especially in diagnosing disease. Efficient and accurate identification of a potentially life-threatening condition in a person’s body is essential for getting them the right treatment in time. It’s equally important for not over-diagnosing a condition, and then possibly clobbering it in a way that cures the disease but leaves a person with poor quality of life.
We clearly see examples of these problems in prostate cancer (PCa), where a pressing question is how to identify which PCa cell lines are potentially lethal. When a pathologist (person who analyzes tissue samples) looks at a tiny specimen from a biopsy needle, he or she determines if a cluster of misshapen cells is a Gleason pattern 3 or 4, a judgment that will influence which treatment options the patient will be offered.
Great strides are being made in applying deep learning to the challenge of PCa diagnosis. I came across an exciting Google Research Blog called “An Augmented Reality Microscope for Cancer Detection.”[ii] Researchers in the Google Brain Team devised a way to improve an ordinary microscope used in pathology labs with “eyes” (camera) and a “brain” (computer). Remember above when I said the key word is data? The “brain” was educated to recognize, and build on, patterns from thousands of actual cancer specimen slides. As the pathologist is looking through the eyepiece of the microscope, the computer camera sees in real time what the human sees. The computer immediately augments (adds its intelligence to) objects in the field of vision by highlighting cell features that it recognizes as cancer. This helps eliminate doubt about ambiguous abnormalities that might or might not indicate potentially lethal PCa. In other words, the computer quickly cuts through what we humans often experience as a judgment call.
The benefits of AI in medicine are in rapid development. Though AI, ML and deep learning cannot replace the doctor-patient relationship and dialogue, these tools can quickly equip doctors with the information they need to share with the patient. IBM’s supercomputer, Watson, can sift through mountains of patient and disease data to swiftly filter and present what is most crucial for a doctor to know. For instance, in a proof-of-concept genomics analysis, Watson” took 10 minutes to come up with conclusions that were similar to those reached by the team of experts after 160 hours of analysis.”[iii]
The day is fast approaching when novel devices like the Augmented Reality Microscope enhance the efficiency and accuracy of diagnosing prostate (and other) cancers, and increase our confidence that their patients are receiving treatment and care tailored for each of their needs.