Live Science · September 13, 2026

AI trained to 'think' like human pathologists may be better at spotting cancer

MedicineHealth AdministrationComputer Science

THE AI ANGLE

Mimicking pathologists' search and zooming behaviors to screen tissue slides for cancer

Researchers developed a training approach called Pathology-CoT that teaches AI to mimic how human pathologists actively scan, pan, and zoom across tissue slides rather than analyzing static patches. Built into a tool named Pathology-o3, this method detected cancer-positive lymph node slides with up to 100% sensitivity, outperforming general vision-language models like OpenAI's o3. For faculty across medicine, computer science, and health administration, the study demonstrates that capturing clinical search behaviors—not just final diagnostic labels—can significantly enhance diagnostic AI performance.

THE TEACHING ANGLE

Students can evaluate the administrative and clinical trade-offs of deploying an intentionally high-sensitivity triage AI that yields between 15.5% and 37.1% false positives to ensure no malignancies are overlooked.

Read the original at livescience.com Generate teaching or study materials

Instructors get discussion guides, assignments, and mini-cases. Students and readers get a plain summary, class prep, and an exercise. All built from the full article. Three are free with an account.

AI trained to 'think' like human pathologists may be better at spotting cancer

More in Medicine