🩺 Artificial intelligence in radiology brings both significant benefits and important risks. Language models support the analysis of medical documentation, while convolutional networks can achieve very high accuracy in precisely defined tasks such as image interpretation, detection, and assessment of pathological changes. 🧬 As a result, AI reduces variability between radiologists’ assessments, improves the reproducibility of analyses, and increases work efficiency by relieving specialists of routine tasks.
⚠️ At the same time, significant limitations and risks remain. Deep learning models perform poorly outside the scope of the data on which they were trained, which limits their ability to generalize in atypical cases. They are vulnerable to various types of attacks, may “hallucinate” by generating synthetic images, and language models can produce incorrect yet convincing responses. Legal liability in the event of errors and the need for continuous physician oversight also remain major concerns.
👉 Therefore, in the foreseeable future, AI will primarily serve as a tool supporting radiologists, automating technical tasks, while key clinical decisions, ethical judgment, and responsibility will remain in the hands of specialists.
✒️ We encourage you to read the publication discussing these issues, prepared in collaboration with researchers from the Institute of Electronics of Lodz University of Technology, Jagiellonian University, AGH University of Krakow, and Silesian University of Technology:
Obuchowicz R., Piórkowski A., Nurzyńska K., Obuchowicz B., Strzelecki M., Bielecka M. "Will AI Replace Physicians in the Near Future? AI Adoption Barriers in Medicine" Diagnostics. 2026; 16(3):396. https://doi.org/10.3390/diagnostics16030396