The history of medical diagnosis is a march through careful remark. Doctors in ancient Egypt first diagnosed urinary tract infections by observing patterns in patients' urine. To diagnose heart and lung diseases, medieval doctors added core elements of the physical examination: pulse, palpation and percussion. Laboratory tests were added within the twentieth century, and complicated imaging and genetics within the twenty first century.
Despite all of the advances, nevertheless, diagnosis remains to be largely a human task. Doctors depend on what are often known as disease scripts – collections of signs, symptoms and diagnostic findings which might be hallmarks of a disease. Medical students spend years memorizing such scripts, training to identify, for instance, the submillimeter-sized deviations in ECG wave measurements that might alert them to a heart attack.
But after all, people make mistakes too. Sometimes misdiagnoses occur because a physician misses something – when the disease patterns fit the script, however the script is misinterpreted. This is estimated to occur in 15 to twenty percent of all doctor visits. In other cases, misdiagnoses occur since the disease has features that don’t fit known patterns – they don’t fit the script, similar to when a heart attack occurs without telltale symptoms or EKG findings.
Artificial intelligence may help solve these two fundamental problems – whether it is sufficiently funded and used appropriately.
First, AI is less prone to the same old aspects that lead doctors to make diagnostic errors: fatigue, lack of time and cognitive bandwidth when treating many patients, gaps in knowledge, and reliance on mental shortcuts. Even when diseases follow patterns, computers are sometimes higher than humans at spotting details hidden in voluminous health data.
Using AI to enhance the accuracy and timeliness with which doctors detect disease can mean the difference between life and death. An ischemic stroke, for instance, is a life-threatening emergency wherein a blocked artery impedes blood flow to the brain. Brain imaging confirms the diagnosis, but that imaging have to be performed and interpreted quickly and accurately by a radiologist. Studies show that through superhuman pattern recognition skills, AI can discover strokes seconds after imaging is performed—dozens of minutes ahead of often-busy radiologists. Similar capabilities have been demonstrated in diagnosing sepsis, pneumonia, blood clots within the lungs (pulmonary embolism), acute kidney failure, and other conditions.
Subtle patterns
Second, computers will be helpful for diseases for which we haven't yet developed proper scripts. Artificial intelligence can diagnose diseases based on latest patterns which might be too subtle for humans. Consider, for instance, hypertrophic cardiomyopathy, a rare genetic disorder wherein the center muscle grows larger than it should, eventually resulting in heart failure and sometimes death. Experts estimate that only 20 percent of those affected are diagnosed, a process that requires a consultation with a cardiologist, a cardiac ultrasound, and sometimes genetic testing. What, then, concerning the remaining 80 percent?
Researchers across the country, including on the Mayo Clinic and UC San Francisco, have demonstrated that artificial intelligence can recognize complex, previously unknown patterns and discover patients prone to have hypertrophic cardiomyopathy. This means AI-driven algorithms can screen for the condition in routine EKGs.
AI was in a position to discover these patterns after examining the ECGs of many individuals with and without the disease. The rapid growth of healthcare data – including detailed electronic health records, imaging, genomic data, biometrics and behavioral data – combined with advances in artificial intelligence has created an enormous opportunity. Because of its unique ability to discover patterns in the information, AI has helped radiologists find hidden cancers, pathologists characterize liver fibrosis and ophthalmologists detect retinal disease.
One challenge is that AI is pricey. It requires large amounts of knowledge to coach computer algorithms and the technology to support it. As these resources grow to be more ubiquitous, it may well grow to be difficult to guard the associated mental property, discouraging private investment in these products. In general, diagnostics have long been considered unattractive investments. Unlike their therapeutic counterparts, which receive around $300 billion in research and development annually, diagnostics receive a modest $10 billion in private funding.
Who pays for AI?
Then there may be the query of who specifically pays for the usage of AI-based tools in medicine. Some applications, similar to stroke detection, save insurers money (by stopping expensive ICU stays and subsequent rehabilitation), and these technologies are inclined to be reimbursed more quickly. However, other AI solutions, similar to hypertrophic cardiomyopathy detection, can result in higher spending on expensive follow-up therapies to treat newly diagnosed chronic diseases. Although the usage of AI in such cases can improve the standard of care and long-term outcomes, reimbursement and thus adoption will be slow without financial incentives for insurers.
It is normally difficult to boost the funds needed for brand spanking new medical developments, however the National Academies of Medicine have estimated that improving medical diagnostics could save tens of billions of dollars and countless lives.
Artificial intelligence offers a option to get there. It should complement, not replace, the human expertise that already saves so many lives. The way forward for medical diagnostics will not be about handing over the keys to AI, but moderately about leveraging what it may well do and what we will't. This could possibly be a special moment for diagnostics if we invest enough and get it right.
image credit : www.mercurynews.com