update

MEME/index.html

@@ -255,7 +255,7 @@ <h1 class="title is-2" style="text-align: center; padding-bottom: 10px;">Experim
                 To test our new architecture, we performed an evaluation benchmark of the MEME model on various ED disposition and decompensation tasks. We asked the models in the benchmark to predict whether a patient should be admitted to the Emergency Room at the time of their visit. From the subset of patients who were admitted, we conducted a multilabel classification task where we asked the models to predict whether admitted patients would be discharged home during their admission, needed to be admitted to the intensive care unit, or would die during their stay. We showcase the prevalences of these tasks in Table 1 below, where each subsequent task is deemed to be more difficult than the previous one.
 
               <div class="column" style="text-align: center;">
-                <img src="resources/prevalance.png" alt="Image description" width="60%">
+                <img src="resources/prevalance.png" alt="Image description" width="80%">
             </div>
 
             In our study we compared our method to three traditional machine learning models defined in a previous ED benchmark (logistic regression, random forest, and MLP), two EHR-foundation models (EHR-shot, and GenHPF), and a GPT3.5-turbo generative model. We evaluated all methods on the F1 score, the Area under the Reciever Operating Characteristic (AUROC), and the Area under the Precision Recall Curve (AUPRC).