Table 1 Summary of EHR features incorporated during pretraining. We pretrain multimodal MSN using each of the EHR features summarized below.
From: Multimodal masked siamese network improves chest X-ray representation learning
Group | Feature | Type | Values | Dimension |
|---|---|---|---|---|
Demographic (\(x_{D}\)) | Age | Numeric | \(\{ x \in \mathbb {N} | 18 \le x \le 100 \}\) | \(x_{age} \in \mathbb {R}^{1}\) |
Sex | Binary | \(\{Male, Female\}\) | \(x_{sex} \in \mathbb {R}^{2}\) | |
Scan medatadata (\(x_{SM}\)) | View | Multiclass | \(\{AP, L, PA, LL\}\) | \(x_{view} \in \mathbb {R}^{4}\) |
Position | Binary | \(\{Erect, Recumbent\}\) | \(x_{pos} \in \mathbb {R}^{2}\) | |
Inpatient stay information (\(x_{SI}\)) | ICU admission | Binary | \(\{Negative, Positive\}\) | \(x_{icu} \in \mathbb {R}^{2}\) |
In-hospital mortality | Binary | \(\{Negative, Positive\}\) | \(x_{mort} \in \mathbb {R}^{2}\) |
