Artificial intelligence is increasingly entering hospitals and research laboratories, promising to transform the way tumors are detected, studied and treated. Algorithms capable of analyzing diagnostic images, interpreting genetic information and predicting possible complications of therapies are opening up new perspectives in oncology medicine, with the aim of making treatments more precise, timely and personalized. Yet, behind the potential of a technological revolution that could profoundly change healthcare, a fundamental question remains: to what extent is artificial intelligence already capable of concretely improving patients’ lives and which applications are still in the experimental phase?
The theme was also at the center of the conference Rethinking Oncology: innovation, multidisciplinarity and continuity of care, organized on 26 October 2026 at the University of Salerno, during which researchers and healthcare professionals explored the opportunities offered by new technologies, from precision medicine to the management of cancer pain, underlining the need to distinguish between promising results and actually demonstrated clinical benefits.
Artificial intelligence in oncology: how it changes the diagnosis of tumors
One of the areas in which artificial intelligence is showing particularly interesting results is that of diagnostics, especially in the analysis of radiological images and histological samples. Through machine learning systems, algorithms can recognize characteristics and anomalies present in mammograms, computed tomography scans and other tests, contributing to the work of specialists in identifying suspicious lesions.
A concrete example comes from the Swedish MASAI study, the results of which were published in January 2026 in the scientific journal The Lancet. The research, conducted on over 105 thousand women, compared mammography screening supported by artificial intelligence with the traditional double reading of exams by radiologists.
The results showed a greater sensitivity of the AI-assisted system, equal to 80.5% against 73.8% of the traditional method, maintaining a specificity of 98.5% in both groups. The number of cancers diagnosed in the interval between two screenings was also numerically lower in the group supported by artificial intelligence, without however demonstrating a statistically significant reduction.
This is important evidence because it shows how AI can help improve some diagnostic performances and lighten the workload of professionals, without replacing their judgement. It also remains necessary to verify the effects of large-scale adoption and, above all, to establish whether these improvements translate into a reduction in mortality over time.
Personalized cancer care: can AI help choose therapies?
Another frontier concerns precision medicine, an approach that aims to identify the biological characteristics of each tumor to guide the choice of the most appropriate treatments.
In fact, not all tumors behave in the same way, not even when they affect the same organ. Genetic and molecular differences can influence disease progression, response to drugs and the development of therapeutic resistance. Artificial intelligence allows us to process large quantities of information, relating genetic, protein and clinical data that would be difficult to interpret simultaneously using traditional methodologies alone.
Particularly promising is the integration between AI and multi-omics technologies, which allow analyzing different levels of cellular functioning. Research published in 2025 in the Journal of Translational Medicine examined the potential of these tools in identifying new therapeutic targets and in the possibility of using existing drugs to treat some female cancers.
Identifying a possible therapy through an algorithm, however, does not mean having proven that that therapy works. The hypotheses generated by artificial intelligence must be verified through experimental studies and clinical trials, which are essential to establish the safety and effectiveness of treatments.
The prospect is therefore that of supporting oncologists in the evaluation of increasingly complex information, not of entrusting the choice of treatments autonomously to a machine.
Chemotherapy and side effects: predict complications before they appear
Artificial intelligence could also play an important role in managing the side effects of oncology therapies, an aspect that has a profound impact on the quality of life of patients and, in some cases, on the possibility of continuing treatments according to the planned program.
Among the most problematic complications is chemotherapy-induced peripheral neuropathy, which can manifest itself with tingling, numbness, changes in sensitivity and pain, sometimes persisting even after the treatment has ended.
A study published in June 2026 in Supportive Care in Cancer analyzed data from 829 patients with colorectal cancer treated with oxaliplatin, developing an artificial intelligence model capable of estimating the risk of neuropathy using the available clinical information.
The results obtained were promising, but the research has important limitations, having been conducted retrospectively in only one center. Further studies will therefore be necessary to understand whether the system can work with the same reliability even on different populations.
The distinction is fundamental: predicting a side effect is not the same as preventing it. An algorithm could tell doctors which patients need more attention, but its real usefulness would depend on the possibility of translating that prediction into interventions capable of reducing complications without compromising the effectiveness of anti-cancer therapies.
Cancer pain: what algorithms can identify
The use of artificial intelligence is also being extended to the study of pain associated with tumors, a complex condition that does not depend exclusively on the extent of the disease, but can be influenced by personal autonomy, general health conditions and the effects of treatments.
A study published in April 2026 in the Journal of Anesthesia, Analgesia and Critical Care, called RUGGI, used machine learning techniques to analyze data from 86 patients with chronic cancer pain.
The researchers observed that impaired functional abilities, i.e. difficulty carrying out normal daily activities, was one of the factors most associated with pain intensity.
The result suggests that, to fully understand a patient’s condition, it is not sufficient to measure how much pain he feels, but it is also necessary to consider how much that pain interferes with his independence and quality of life.
Even in this case, however, the results must be interpreted with caution: the limited number of participants and the moderate performance of the models used do not yet allow them to be considered definitively validated clinical tools.
Patient monitoring and telemedicine: the role of AI after diagnosis
The opportunities offered by artificial intelligence do not end with diagnosis and the choice of treatments, but also concern the continuity of care, especially when the patient is at home and has to live with side effects, persistent symptoms or changes in their health conditions.
The integration between telemedicine, digital tools and predictive algorithms could allow information to be collected more regularly, identifying signals that require medical in-depth analysis and facilitating coordination between the professionals involved in the therapeutic process.
A system could, for example, help recognize a worsening of symptoms reported by the patient, helping healthcare personnel to establish which situations deserve more timely evaluation.
The benefit, however, also depends on the organization of health services: an automatic alert, however accurate, does not improve care if there is no professional in charge of evaluating it and intervening when necessary.
What are the risks of artificial intelligence in cancer treatment?
The growing diffusion of AI in medicine raises questions regarding the reliability of systems, the protection of personal data and the responsibility of clinical decisions.
Algorithms are developed and trained using large amounts of information, but the available data does not always adequately represent all categories of patients. A system that obtains satisfactory results in a certain hospital could, for example, prove to be less precise in another context, where population characteristics, diagnostic procedures or information collection methods change.
Added to this are the risks of tools based on generative artificial intelligence, which can produce apparently credible answers but contain errors, incomplete information or indications not supported by scientific evidence.
The European Society of Medical Oncology, ESMO, has published guidance dedicated to the use of large language models in clinical practice, recalling the importance of human supervision, protection of health data and validation of tools before their use.
The principle remains that of using technology to support the work of professionals, preventing inadequately verified tools from autonomously guiding decisions with potentially significant consequences for patients.
How to understand if an AI system is really useful to patients
The decisive question for hospitals and the National Health Service is to establish which technologies actually deserve to be introduced into care pathways.
To do this, it is not enough to know the precision of an algorithm or the speed with which it can analyze an exam. It is necessary to evaluate whether its use allows us to obtain more accurate diagnoses, reduce complications, improve the quality of life or make care more efficient without compromising safety and equity of care.
In this context, Health Technology Assessment takes on particular importance, an evaluation system that considers not only the clinical effectiveness of innovations, but also the costs, economic sustainability, organizational impact and consequences for patients.
Artificial intelligence therefore represents one of the most interesting prospects for the future of oncology, but its diffusion requires a balance between technological enthusiasm, scientific prudence and clinical responsibility.
The real transformation will not simply consist in having increasingly sophisticated algorithms within hospitals, but in demonstrating that these tools really help doctors make better decisions and patients receive more effective, safe and quality-of-life care.
