Skip to main content

Table 4 Performance of an i2b2-2014 model with custom embedding tested on 4 different datasets

From: Customization scenarios for de-identification of clinical notes

Test onPHI typeEmbedding
GloVeembed-mimicMatching embedding
physionetAll76.2/61.3/67.981.8/64.1/71.876.9/62.4/68.9 - embed-mimic-nursing
physionetName92.9/73.1/81.895.5/81.8/88.191.0/73.4/81.3 - embed-mimic-nursing
mimic-radiologyName92.9/85.7/89.197.2/85.9/9192.0/87.2/89.4 - embed-mimic-radiology
mimic-dischargeName92.5/85.4/88.893.4/89.8/91.692.1/86.3/89.1 - embed-mimic-discharge
mimic-echoName95.5/61.4/74.798.7/64.4/77.3too small to build embedding