By Adele Calello
In the medical field, the availability of data has witnessed exponential growth, encompassing diverse sources. Alongside traditional "structured" data, unstructured data (such as texts, images, sounds, etc.) represents about 80% of the data generated daily. Artificial intelligence algorithms in the medical field play a crucial role in interpreting this huge amount of data and identifying possible cause-and-effect relationships between the data itself and a patient's health conditions.
The field where significant progress has been achieved in leveraging artificial intelligence to aid doctors is diagnostics. There exists substantial scientific evidence supporting their reliability, notably in oncology, respiratory, and cardiology domains. By training machines to interpret images derived from X-rays, ultrasounds, CT scans, electrocardiograms, and biological tissue analyses, it becomes feasible to detect tumor, cardiovascular, dermatological, and respiratory pathologies with a high degree of accuracy.
The medical imaging sector stands out as one of the fields in medicine greatly impacted by artificial intelligence. Over recent years, advanced AI models have demonstrated significant success in detecting abnormalities on X-rays and CT scans, as well as diagnosing various dermatological conditions.
Currently, AI is being explored for new applications in pathology diagnosis, including the interpretation of ECGs and histopathological slides. In traditional radiodiagnostics, AI integration is seamless due to digitized workflows and standardized image storage. These established AI algorithms can even predict clinical outcomes with reasonable accuracy using CT images of traumatic brain injuries or tumors.
Another significant area of focus involves prediction systems capable of identifying potential pathologies before they manifest. For instance, by analyzing electrocardiograms and a patient's medical history, it is possible to predict the risk of developing cardiovascular conditions like atrial fibrillation or heart failure. Similarly, these tools enable accurate prediction of lung cancer development up to six years in advance.
Of particular interest are the artificial intelligence systems designed to assist doctors, as they can recommend the optimal management or treatment approach for a patient's condition from a pharmacological perspective. These AI-driven suggestions rely on existing guidelines, scientific evidence from publications in international medical journals, the outcomes of patients with similar conditions, and the medical history of the individual being treated.
One of the critical risks associated with the use of artificial intelligence in medicine is the insufficient testing and lack of scientific evidence supporting the systems in use. Methodologically rigorous clinical trials involving multiple centers, hospitals, and institutes should be conducted to systematically assess the effects on a representative sample of the population studied, from the outset to the conclusion of the study.
Artificial intelligence systems must be properly trained to mitigate evaluation distortions, known as bias in technical terms. Literature documents cases of AI tools failing to provide answers to certain questions (diagnostic, prognostic, predictive) because the patients for whom the answers were sought were not adequately represented in the sample used to train the system.
However, it's crucial not to envision artificial intelligence as a replacement for doctors. While the tools may be intelligent, the ultimate decisions still rest with the specialist due to ethical, and deontological issues, and responsibility. Finally, these tools must be regulated at an institutional level in compliance with the new European legislation on medical devices, to which these tools largely belong. This should involve implementing stricter rules regarding safety and efficacy testing for their approval and market introduction.
Source