講者姓名：許鈞南  Chun-Nan Hsu

講者機構：University of California, San Diego

講者網頁：https://profiles.ucsd.edu/chun-nan.hsu

演講時間:  2026/01/13 上午 10:30

演講地點:  大仁樓三樓301教室

英文摘要:

Many AI models developed for healthcare applications–such as disease diagnosis, risk and prognosis prediction, and treatment outcome prediction–are based on data derived from electronic health records (EHR). However, many of these models have failed to generalize across different hospitals or even over time within the same hospital. This failure has led to significant hesitancy regarding AI adoption in healthcare. This talk will outline the challenges of building AI models that generalize across different hospitals and will describe potential solutions. These points will be illustrated by three example applications: chest X-ray interpretation, prediction of the risk of opioid use disorder, and outcome prediction of symptomatic COVID-19 patients.