Urinary tract infections are typically diagnosed through urine culture, a process that involves examining urine samples for bacterial growth under a microscope for 24 to 48 hours. Astonishingly, more than two-thirds of these samples yield negative results after this lengthy analysis. The development of AI by the Austrian and Italian research team has the potential to detect negative samples more accurately than previous methods, leading to a 16 percent reduction in laboratory workload.
Transparent and interpretable AI
What sets this AI apart is its transparency and interpretability, which are crucial in the field of medicine. The AI explains to doctors why it categorizes a sample as negative. The researchers employed decision trees as a form of interpretable AI. Decision trees function similarly to human thought processes, asking and answering sequential questions to reach a judgment. The AI's decision criteria align closely with those of medical professionals.
Enhanced diagnosis and workload reduction
The AI algorithm takes into account seven parameters, with sensitivity reaching 95 percent. This comprehensive approach significantly outperforms previous methods, reducing the burden on laboratories and expediting diagnosis times. The researchers published the complete decision tree algorithm in the "American Journal of Clinical Pathology," enabling other institutions using the same flow cytometry device to implement this methodology immediately.
While this study demonstrates immense potential, it has thus far only been conducted in a single hospital. The research team is actively seeking collaboration partners interested in conducting similar studies in their respective healthcare facilities. Variations in sampling procedures and dietary influences necessitate further evaluation in diverse hospital settings.
MEDICA-tradefair.com; Fraunhofer Austria Research GmbH