How is Artificial Intelligence (AI) transforming the landscape in Oncology – and why we need to embrace it?

Artificial intelligence (AI) and its implementation in Oncology marks a new chapter in prevention, diagnosis, and treatment in cancer care, holding even greater potential to overcome current hurdles and shape the future. Although there is, in general, a well-structured consensus of guidelines and great strides of progress have been made throughout the years, several problems regarding misdiagnoses, treatment inadequacies, and managing high-volume data and its translation in clinical practice continue to arise every day^[1].

AI's most prominent role lies in its ability to analyze vast pathological and radiological patient datasets. Of note, more than 80% of the US Food and Drug Administration (FDA)-approved AI devices in 2021 were correlated with cancer care and diagnostics. Machine learning and radiological predictive AI tools have many times proven to be superior in terms of early detection, for instance, of lung and breast malignant lesions, when compared to the human eye^[2]. Classification of large-scale patient data, precise characterization of features, and monitoring of lesions have become feasible via AI. In pathology, AI-aided deep learning algorithms maximized the accuracy of classifying and diagnosing cancerous and premalignant specimens, for example, melanoma samples, when compared to certified experts^[3]. Furthermore, radiomics data have been widely exploited in predicting response to chemotherapy and immunotherapy among various tumor types^[2].

Precision oncology, drug development, and genome-based interventions are other fields greatly facilitated by AI applications. Large-scale multi-omics analyses unlocked a new level of personalized care by identifying unique alterations that can be potentially targeted with specific drugs. Integrating these various types of data can also predict patient responses, offering holistic, patient-tailored therapeutic approaches^[4]. Moreover, several assistance-decision systems have shown equivalent concordance with multidisciplinary medical teams in relation to treatment planning, enhancing accuracy and personalized care^[2].

The field of cancer research has also been substantially aided by AI applications. Basic and translational research as well as drug development are only a few areas where AI is being applied with impressive outcomes, while minimizing time lapses and human errors have accelerated clinical research at remarkably lower costs. In addition, optimizing parameters in patient selection has led to very important improvements regarding clinical trial strategies^[2].

Some of the aforementioned applications are already implemented in everyday clinical practice, heralding a new era in the field of medical oncology. However, there are many reported cases when the algorithm breaks, raising critical concerns regarding bias, robustness, and data security. Health-care bias occurs in two main parameters: data bias and inherent algorithmic bias. For instance, inaccuracies occur in special patient populations, such as older patients, melanoma patients with darker skin, etc., which are often underrepresented in clinical trials. This plainly means that AI can often be overestimated when it comes to processing real-world data, and ensuring diverse features are included in machine learning training techniques is critical for its canonical function^[5]. Over and above, data safety should be a primary concern when it comes to incorporating AI in clinical practice; sensitive data is often prone to cyberattacks, and the typical anonymization techniques have only proven inadequate. Strategies known as homomorphic encryption and noise adding to database are currently developed to mitigate those percentages, but, to date, are not widely implemented in large-scale networks. The clinician's role in such cases remains undoubtedly critical; he/she must maintain the insight to question AI outputs while ensuring that its role is supplementary rather than substitutional to the physician's judgment^[6].

It is evident that AI is revolutionizing the health-care model in Oncology in ways impossible to imagine up until very recently. It is critical to exploit its abilities properly to augment researchers, clinicians, and, above all, patients. The future is undoubtedly multimodal and, nevertheless, artificial^[4].

In the present issue, there is a very interesting article about the AI-oriented public health model and its implementation in cancer care. There are also other articles about many tumor types, covering a wide area in the field of medical oncology.