When AI reads pixels: one patient's story

In a 73-year-old patient, a lung tumor with suspicious features was discovered incidentally on CT; the AI confirmed the suspicion and recommended comparing the finding with a one-year-old X-ray. Two board-certified radiologists labeled the older image as negative, but the system identified an opacity in the left upper lung field. The explanation is simple: AI analyzes pixels, whereas a human tends to perceive the "image as a whole," and thus it may be more sensitive. In the differential diagnosis, it considered a benign nodule, primary lung cancer, and low-grade metastatic disease, and recommended considering a biopsy.

When asked about the optimal course of action, the AI proposed a CT‑guided percutaneous biopsy and reminded that the approach should be confirmed by a multidisciplinary team. The PET scan was positive at the site of the lesion; however, histology from the biopsy confirmed a benign finding. Thus, the system had already accounted for the possibility of a non-malignant process, which spared the patient several weeks of stress. The Slovak in the outputs still needs some "polishing," but the recommendations were understandable.

From diagnosis to treatment: recommendations and benefits

AI also assists pathologists: it reduces interobserver variability in interpretation from approximately 10 % to around 1,5 %. After summarizing all the data, the system proposes treatment according to ESMO and ASCO recommendations adapted for Slovak practice – from surgery through systemic therapy including cytostatics to targeted therapy or immunotherapy, always with justification (for example, identified mutations). Such an approach improves quality and saves costs, while also helping to cope with physician shortages by increasing efficiency and shortening waiting times. The patient has an ongoing overview of their journey and results without unnecessary phone calls, with the exception of sensitive histology results, which should be communicated by a physician; the goal is lower mortality and later the use of the model for other diseases as well.