Towards the Practical Utility of Federated Learning in the Medical Domain [CHIL 2023]

Seongjun Yang*¹, Hyeonji Hwang*², Daeyoung Kim², Radhika Dua³, Jong-Yeup Kim⁴, Eunho Yang², Edward Choi² | Paper

¹KRAFTON, ²KAIST AI, ³Google Research, India, ⁴College of Medicine, Konyang University

Federated learning with eICU database

We evaluate FL methods on the eICU database. The database is a multi-center intensive care unit (ICU)database with high granularity data for over 200,000 admissions to ICUs monitored by eICU Programs across the United States. We implement six clinical prediction tasks from the eICU database.

You can see the details of FL with eICU here.

Federated learning with skin cancer images

We evaluate FL methods on skin cancer image datasets originating from different sources. We implement skin cancer image classification.

You can see the details of FL with skin cancer here.

Federated learning with Electrocardiogram (ECG)

We evaluate FL methods on the PhysioNet 2021 challenge dataset. The dataset contains ECG data originating from different sources. We implement the cardiac arrhythmia classification.

You can see the details of FL with ECG here.

The details of Experiment

We have attached all details of the experiments in the experiment_detail.

More experiment results

We attach the PRAUC results of FL methods with the eICU in the experiment_results.