Predictive algorithms analyze millions of images in seconds. The technological revolution that challenges traditional medicine is already a reality, that’s where.
In a silent corridor of Humanitas Mater Domini in Castellanza or in the rooms of the Breast Unit in Milan, a server processes thousands of mammograms in a few seconds. It identifies a nearly imperceptible cluster of pixels, a variation in breast density that the human eye might dismiss as simple glandular tissue. It is not a simulation: from January 2026, the integration of software artificial intelligence latest generation in digital mammography with tomosynthesis has become an operational reality in these centers.
The adoption of AI in Italian diagnostics is no longer an experiment, but a clinical protocol. Software based on neural networkslike those validated by GISMa (Italian Mammographic Screening Group) and used in projects such as BREASTnegative.aiare trained on massive databases — over 7 million images — to recognize tumor patterns with specificity that touches the 90%. The system does not limit itself to looking for cancer: it objectively evaluates breast density, one of the main risk factors, acting before the pathology becomes visible to traditional tests.
The AI challenge between Humanitas and centers of excellence
The dilemma, however, shifts from the technical to the temporal level. If a predictive algorithm, such as those being tested at the Polytechnic of Milan within the European network UNIQUEidentifies a high probability of malignant development in the following 24 months, the doctor finds himself at a crossroads. Intervening immediately means risking the overtreatment of a lesion that could remain silent; waiting means ignoring a technological alert that could save a life.
In Italy, the legislative framework – stuck at Gelli-Bianco law — struggles to follow the speed of the silicon. The responsibility for the final diagnosis remains with the doctor, but the “invisible diagnosis” creates a legal short circuit. If the software reports a critical issue in a hospital such as Kauno Klinikos or in the centers Humanitas Medical Care and the radiologist decides not to proceed, his position becomes vulnerable in case of subsequent appearance of the tumor.
From the “Black Box” to shared medical responsibility
The concrete risk is the birth of one technological defensive medicine: The professional might ratify every suggestion of the machine just to avoid legal repercussions. There “black box” of the algorithm also makes the informed consent: how to explain to a patient that he has to operate on the basis of a probabilistic calculation that not even the surgeon can explain down to the smallest logical details?
The challenge of 2026 lies in the creation of a chain of shared responsibility. The model is not the replacement, but the “augmented desk”where AI acts as a second mandatory reader to reduce the false negatives (which still weigh on public health). The success of this revolution will depend on the ability of hospitals and universities to manage the paradox of prediction: learning to trust a machine that sees the invisible, while maintaining the humanity — and burden — of the final clinical decision.
