A second opinion might not come from a doctor anymore; instead, patients might consult an artificial intelligence software.
Artificial intelligence (AI) is the simulation of human intelligence by machine learning algorithms. It is a technology that has been developed to mimic the way humans think and make decisions. The concept for AI was first described as early as 1950. However, it was not until the invention of deep learning that AI had practical applications. Now, AI has advanced to the point where its problem-solving capabilities are sometimes more advanced than humans’.
Healthcare tops the list of industries utilizing AI by a landslide, so it’s no surprise that recent advances in AI have been centered on improving healthcare—everything from administrative workflow to patient diagnosis and treatment.
Using machine learning algorithms, AI can see patterns and “learn” to diagnose based on these patterns. AI can be programmed to flag signs of abnormalities and symptoms of diseases in medical images like MRIs, x-rays, and CT scans. These types of software are composed of a neural network that evaluates medical images alongside genetic information. This data is used in AI’s decision-making process for high accuracy and insight into a patient’s condition.
It’s safe to say that AI is, by and large, successful at correctly diagnosing patients. It takes years of training for doctors to learn how to diagnose illnesses correctly. Even then, misdiagnosis is rampant; an estimated one out of 20 adult patients is misdiagnosed every year in the U.S.
Similar to the intense practice doctors must undergo, it takes thousands of examples for an algorithm to learn how to recognize illness. In fact, with a standard accuracy of 72.52%, AI diagnoses illness even more accurately than the average doctor, who, in the same study, diagnosed with 71.4% accuracy.
For AI to accurately diagnose patients with that level of accuracy, the examples need to be based on factual data. For that reason, AI is most useful in processes that involve digitized diagnostic information that doesn’t leave much room for guesswork or misinterpretation.
AI technologies that can interpret non-digital information, such as handwritten doctor’s notes, are currently developing. However, it may be some time before such technologies learn to recognize more complex, non-digital data.
In the past, AI was built to learn to diagnose following the same thought process doctors use. The original intention was for AI to mimic human behaviors as closely as possible. However, in many ways, AI’s cold, objective reliance on probabilities and factual data is advantageous. In some cases, machines are now more adept at diagnosing human illness than humans are.
Some illnesses are more complex to diagnose than others, whether for lack of resources, scarcity of qualified professionals, or human limitations. Below are several examples of common illnesses that are diagnosed with the help of AI.
Breast cancer is the most common cancer found in women, with about one in eight women afflicted in their lifetime. This type of cancer is detected using mammograms and ultrasounds. AI used in digital mammography has been found to improve the accuracy and efficiency of breast cancer screening.
According to a 2019 study published by the University of Oxford Press, the AI in digital mammography machines uses “deep learning convolutional neural networks, feature classifiers, and image analysis algorithms to detect calcifications and soft tissue lesions.” Based on these findings, the software assigns a score between one and ten. One represents the lowest level of suspicion that cancer is present, while 10 represents the greatest. To give the score, AI compares the patient’s data to thousands of ultrasounds images, both healthy and abnormal.
Lung cancer is the deadliest cancer globally, accounting for about 25% of all cancer-related deaths. AI technologies have been developed to recognize early signs of lung cancer not visible to doctors. New AI technology uses deep learning to find lung cancer early. Early diagnosis can make all the difference for cancer that kills 75% of patients within five years of diagnosis.
Rather than relying on what parameters a human programmer classifies as malignant, the machine analyzes thousands of CT scans and learns for itself what lung cancer looks like.
Before AI’s assistance, lung cancer was often not detected in time for effective treatment. This could be due to several things such as ignored symptoms or doctors’ limitations in detecting tumors with the naked eye–It’s easy for tiny, malignant tumors to be missed.
Tuberculosis remains the most deadly infectious disease. Now, it can be diagnosed with high accuracy using a smartphone camera.
Like the above examples, the AI used in this technology also applies a deep-learning model.
It was developed using a database of 250,044 chest x-rays. These x-rays, which did not show tuberculosis, were used to train the neural network to recognize abnormalities.
The model was then recalibrated using an augmented data set and an additional two-layer neural network.
The software can detect tuberculosis in smartphone photos of x-rays. This kind of AI is a critical tool in high populations of people with tuberculosis or limited resources. Traditional diagnosis by radiologists, on the other hand, has proven more expensive and less accurate. This particular AI features rapid reading and reporting–it determines whether an x-ray shows tuberculosis in under two minutes. This is critical for an infection like tuberculosis, as early detection leads to higher survival rates.
Earlier this week, Google announced the development of an AI-powered dermatology tool that can detect skin problems. Google says that every year they see nearly ten billion Google Searches related to skin, nail, and hair issues.
Google’s new application is web-based and requires the use of a smartphone camera. Using their camera, a user must take photographs of the problematic area from three different angles. The user is then asked to answer questions about their skin type and how long they’ve had symptoms. The application then draws from its knowledge of 288 conditions to provide users with a list of possible conditions they may be experiencing.
Google stresses that the tool is not to be used in place of a formal diagnosis. However, it may be conducive for those who have limited resources and cannot see a dermatologist. The tool shows information and an FAQ for the identified condition to give users information needed to make a treatment decision. As the AI continues to be developed, it may prove to be an invaluable tool for people unsure whether they need to see a doctor for their skin condition.
Diabetic Retinopathy is a condition that affects the eyes due to diabetes. There are a few technologies in development that use AI to screen patients for Diabetic Retinopathy.
One system, IDx-DR, has already been cleared by the FDA for the detection of diabetic Retinopathy. Two images are captured per eye. The images are then scanned for signs of diabetic Retinopathy within minutes.
Like the EyeArt AI Screening System, other systems use similar mechanisms but do not require substantial training to use. It also supports a dual diagnosis from both the AI and a human doctor, which opens up possibilities for a wholly novel diagnosis method.
AI can be used to flag abnormalities in medical imaging, predict outcomes of a stroke, and help patients manage chronic diseases. Increasingly, AI is being used to diagnose illnesses, such as cardiac arrhythmia, lung cancer, skin lesions, and Diabetic Retinopathy.
AI continues to have vast implications for the healthcare industry overall. This technology has become an essential tool in a field that is in constant demand of trained professionals. Over-demand and under-supply leave many physicians over-exerted and delay treatment. The medical profession has one of the most significant percentages of burnout; nearly 42% of practicing physicians report feeling burnout. Many doctors experience exceedingly high levels of stress before they even begin practicing.
With this in mind, we can see how AI can improve the experience of both patients and physicians and improve diagnostics, treatment outcomes, and, when early detection is critical, be the difference between life and death.
