WEDNESDAY, Aug. 9, 2023 (HealthDay News) -- Incorporating human preferences into an image-based diagnostic algorithm can improve artificial intelligence (AI)-based decision support, according to a study published online July 27 in Nature Medicine.

Catarina Barata, Ph.D., from the Instituto Superior Técnico in Lisbon, Portugal, and colleagues examined whether human preferences can improve diagnostic AI-based decision support using the example of skin cancer diagnosis. Nonuniform rewards and penalties based on expert-generated tables were utilized, balancing the benefits and harms of diagnostic errors; these were applied using reinforcement learning to find a strategy that maximized cumulative rewards while considering both clinician and patient preferences.

The researchers found that the reinforcement learning model improved the sensitivity for melanoma from 61.4 to 79.5 percent and for basal cell carcinoma from 79.4 to 87.1 percent compared with supervised learning. There was a reduction noted in AI overconfidence, while accuracy was maintained. The rate of correct diagnoses made by dermatologists increased by 12.0 percent with reinforcement learning; the rate of optimal management decisions increased from 57.4 to 65.3 percent.

"The AI learned to take into account not only image-based features, but also consequences of misdiagnosis in the assessment of benign and malignant skin manifestations," lead author Harald Kittler, M.D., from the Medical University of Vienna, said in a statement. "This, in turn, helps physicians make more accurate decisions tailored to individual patients in complex medical scenarios."

Several authors disclosed financial ties to medical technology companies.