Machine Learning in Medical Diagnosis

JENNIFER JOHN BRITTO – With the current concerning global shortage of physicians, medical diagnosis has become a prominent issue. The standard number of physicians in a hospital is definitely not adequate enough to support all the patients; according to a Harvard medical study, 70-77% of patients faced serious medical problems due to misdiagnosis and negligence by doctors. In the medical field, time and accuracy are very valuable—especially when regarding diagnoses, because a life can be saved or lost within a blink of an eye. Thankfully, innovation in technology has created a solution: machine learning.

Machine learning for medical diagnosis is the use of data and models to identify an ailment using the test results and denoted symptoms. In a study published by Artificial Intelligence in Medicine, researchers found that machine learning required a lower number of tests to diagnose a patient than doctors; this greatly saved the amount of time and effort required by patients and hospital staff. It is also important to note that utilizing machine learning has another positive consequence: fewer tests means lower expenses and equipment used. In a study published by BMJ Open Quality, researchers concluded that machine learning increases the efficiency of hospitals by reducing readmissions due to better diagnoses and predictions. This factor is especially important during the pandemic, as many hospitals do not have enough resources for every patient in need of immediate care.

Similarly, machine learning is a great asset to developed countries, particularly with present concerns of going to the hospital during the pandemic. Because so many developing countries’ medical offices are understaffed, the use of machine learning significantly decreases the workload for overworked physicians. Machine learning screening is also considerably cheaper than regular screening tools, those techniques are much more affordable for patients. Furthermore, machine learning diagnostics for certain conditions can be done virtually, and allows patients to identify their symptoms; this not only eases physicians’ fears that patients are ignoring their symptoms, but also limits interpersonal contact. Based on the identified symptoms, physicians will be able to see possible diagnoses and flag medical issues that need urgent care. 

As with most innovations, machine learning has a few concerns in addition to its many benefits— one major concern regarding machine learning is the possibility of it replacing the jobs of physicians. Machine learning has consistently outperformed physicians in studies, as it is capable of  predicting future risks for patients. For example, in a study published in the 2019 Nature Medicine, a machine learning algorithm was compared to the work of six radiologists in identifying lung cancer. Researchers determined that the algorithm’s diagnosing capability for lung cancer was significantly more accurate than all of the radiologists. On the contrary, many studies have emphasized that at the moment it is not possible for machine learning to completely replace physicians. One major issue is that not every doctor has the same perspective or opinion on a medical issue, and that is actually beneficial. For many medical issues, professionals have not reached a general consensus on what the “correct” diagnosis is, so artificial intelligence may make a misdiagnosis. 

Overall, both physicians and machine learning have their imperfections. However, when both work together, one accounts for the other’s shortcomings. As a result, errors in the diagnostic processes are minimized, physicians save time, and treatment costs are reduced. It is far from flawless, but research in machine learning applications to medicine is rapidly developing, and hopefully within the next few decades it can be completely integrated into and accepted by the medical industry.

Copy Editor: Alina Baiju