Knowing the actual condition that a patient is suffering from is sometimes the most technical and expensive stage in the whole process of treating a disease. Doctors call that, disease diagnosis. Since time in memorial, efforts have been there to involve artificial intelligence in disease diagnosis -to combat the hustle in this level of medical care.
In fact, it’s literary the most crucial part of treatment, but unfortunately it’s also the stage where things can go wrong when a misdiagnosis takes place.
AI Offers Urgent Solution
Now, using machine learning and artificial intelligence, scientists at Shiley Eye Institute in conjunction with the California University, at San Diego Medical School, have developed a new AI-powered tool that screens patients with retinal diseases. Findings of this project were availed in the February 22 issue of the Cell.
The tool uses computational techniques to screen victims with common but potentially blinding retinal diseases and as a result, the whole process of diagnosing as well as treating the conditions has speeded up. Ideally, that means a very great achievement in medicine; it translates to efficiency, better use of resources and time-saving.
The Untapped Potential of Artificial Intelligence in Medicine
Since around 1959, scientists have always cited that, if discovered how AI has very great potential to practically transform disease diagnosis. Funny enough is that those saying that, were not even close to what we currently see this technology doing.
With the eyes of today, professors are still saying the same, “Artificial intelligence can revolutionize not only disease management but more so diagnosis, only that it demands vast amounts of data,” said Kang Zhang, the senior author, and Ph.D., MD holder, also a professor at Shiley Eye Institute.
The Biggest Hurdle
Current computational approaches are very demanding, involving and very costly. Training an AI system might require feeding in millions of images and that involves spending a lot of time. In that line, there are also concerns about the capability of a machine to handle the task with relation to memory capacity and processing power.
However, in their research, Zhang and colleagues took a slightly different and better approach. They employed an AI-powered convolutional neural network, which as scientist say, it can fasten the training because it uses a special brain.
The network was able to review over 200,000 eye scans, all which were conducted with Optical Coherence Tomography (OCT). Basically, OCT is a noninvasive tech that uses light to create 2D and 3D imageries of tissue for observational analysis.
The team then took a shortcut using a technique called transfer learning. This simply entails using the knowledge gained while solving one condition, stored in a machine, to solve another problem but of related characteristics.
For farther understanding, a good example is using an AI neural network that has been optimized to master anatomical structures of the eye like the cornea, optic nerve or the retina to examine the whole eye for unusual occurrences. This practically allows the AI platform to learn effectively with a much lesser dataset than the traditional methods.
After that, there was another process involved: occlusion testing; where the computer points out the areas in the imagery that generate the greatest interest and draws a conclusion. This is ideally, what the doctors would use as the summary of the diagnosis.
Commenting on that, Dr. Zhang said, “Although machine learning is complex like the black box and hard to know what is happening, but with the additional procedure of occlusion testing, the computer is able to translate to us where it is focusing on, on the image and how it arrived at a diagnosis. That’s what is making the system trustworthy.”
The Targeted Eye Conditions
As is obvious in any research, the focal point of this study was based on two common causes of irremediable blindness: diabetic macular and macular degeneration. Well, it’s worth mentioning that these conditions can be treated if detected early enough.
Confirming that the system was accurate, five ophthalmologists came up to review the scans that were used by the machine, to write their own detailed conclusions separately. These were later compared to the system’s report, and it was excellent.
Besides giving a comprehensive diagnosis, the AI platform was able to offer referral as well as treatment recommendation, something that seemed mind-blowing because that was never done in earlier studies.
Although with a relatively a simple training, the machine performed similar to a smartly trained ophthalmologist, and on top of that, it could decide whether the affected should be referred for treatment immediately -all with a 95% accuracy. That is simply an A+ if we were to grade the machine.
The scientist did the same to diagnose childhood pneumonia, which involved feeding the system with chest X-ray images. It was reported that the machine could accurately tell the difference between bacterial pneumonia and viral pneumonia.
In summary, with more data and advancement of AI, doctors will be able to diagnose patients faster, in the near future.