Advances in the field of Artificial Intelligence (AI) are breaking into medicine with force. This powerful tool stands out especially in the analysis and processing of various medical images of X-rays, MRIs, CTs, microscopy … However, its usefulness goes much further and various AI technologies have shown excellent results when evaluating large quantities of data from medical records to make accurate diagnoses.
The successes achieved so far point to a not too distant future in which doctors, AI and robots work together and coordinated every day to apply a more precise and efficient medicine, thanks to the overcoming of human limitations when processing huge amounts of data. In this sense, an AI does not need to be perfect to be useful in medical practice, it just needs to be better and faster than doctors when offering a diagnosis or a prediction. It is estimated that around 15% of the diagnoses made by physicians are erroneous and AI could contribute to reducing this percentage of failures.
Although there is still much room for improvement in the general development of AIs, there are currently specific uses of this technology that have scientifically demonstrated greater precision and speed in different tasks normally performed by doctors. Google and IBM (with its Watson Health system) are two of the large companies that have given the greatest boost to this sector, as they are involved in multiple AI projects for medical purposes.
Below, these are several areas where AIs already outperform doctors when making diagnoses:
Artificial intelligence against cancer
Unsplash Artificial intelligence can offer high precision in diagnosing different types of cancer, making them promising aids for clinicians. In this sense, a system called Biomind AI, developed by a Chinese hospital and several universities, surpassed 15 prestigious doctors 3 years ago when it comes to diagnosing hematomas (accumulation of blood) and various brain tumors such as meningiomas (tumors in the meninges) or gliomas (tumors from glial cells). Specifically, its rapid diagnosis offered a percentage of success of around 90%, compared with 63-66% of the doctors. This milestone was possible thanks to the training of this AI for 10 years through the analysis of tens of thousands of images of diseases of the nervous system.
With regard to breast cancer, in 2019 the first results of an AI capable of detecting this tumor from mammograms were published. When comparing its performance with 101 radiologists, it was observed that it had a precision similar to the average of this group, although higher than 60% of these specialists.
In 2016, another AI was able to succeed where doctors had failed for months. In just 10 minutes, IBM's Watson technology was able to diagnose a rare type of leukemia (cancer of the white blood cells) in a Japanese woman. Artificial intelligence arrived at the diagnosis by analyzing the genetic data of the patient through its huge database, a task that would have taken weeks for specialists. Among the more than 1,000 mutations detected in tumor DNA, the AI was able to identify those that determined the type of cancer.
Also with childhood diseases
An AI developed in China, based on deep learning and natural language processing, has managed to diagnose multiple ailments (from mild to severe) in children with a precision similar to that of pediatricians. When their results were compared to less expert doctors, their reliability was superior. The researchers used data from the medical records of more than 500,000 Guangzhou patients to train the system. Among the strengths of this technology is a quick response and a greater possibility of success when diagnosing rare diseases, which are often difficult for doctors to recognize.
Studying the arrhythmias
Beyond cancer, the diagnosis of cardiac arrhythmias is a field in which AIs offer great reliability, since they are graphically reflected in electrocardiograms. There are already dozens of automatic systems that are capable of automatically and rapidly recognizing heart rhythm alterations, with high precision, to the point of being approved as diagnostic methods in various countries around the world, together with the supervision of cardiologists.
Artificial intelligence in eye diseases
The diagnosis of eye diseases is another area of application of artificial intelligence where successful results have been observed, especially to detect alterations of the retina through different images: fundus, optical coherence tomography (OCT) … For example , DeepMind from Google managed to automatically identify diabetic retinopathy (a disease of the retina caused by diabetes) in 2016 with a sensitivity (percentage of success in detecting the disease) of 97.5% and a specificity (percentage of success in detecting the disease). rule out disease) of 98.5%, through fundus images.
When validating the artificial intelligence algorithm with a sample of patients with ethnic differences, it again showed very high sensitivity and specificity (91% and 9% respectively). In comparison, ophthalmologists typically diagnose this retinal disease with a sensitivity of 73% and a specificity of 91%. In 2018, this tool once again demonstrated its potential: it was capable of detecting up to 50 eye diseases and problems with as high a precision as any specialist.
Inflammation of the uvea (uveitis), the intermediate layer of tissue that lines the eye, is another area in which AIs stand out for diagnosis. The Uvemaster AI, for example, has a diagnostic accuracy of 96.6%, higher than ophthalmologists. A success that has been achieved through deep learning through the analysis of vast amounts of clinical data. On the other hand, artificial intelligence is currently capable of diagnosing glaucoma (damage to the optic nerve due to too high pressure in the eye) with an accuracy greater than 90%.
