Update README.md

README.md

@@ -21,24 +21,22 @@ The model also predicts the chest X-ray view (AP, PA, lateral), patient age, and
 The [CheXpert](https://stanfordmlgroup.github.io/competitions/chexpert/) (small version) and [NIH Chest X-ray](https://nihcc.app.box.com/v/ChestXray-NIHCC) datasets were used to train the model.
 Segmentation masks were obtained from the CheXmask [dataset](https://physionet.org/content/chexmask-cxr-segmentation-data/0.4/) ([paper](https://www.nature.com/articles/s41597-024-03358-1)).
 The final dataset comprised 335,516 images from 96,385 patients and was split into 80% training/20% validation. A holdout test set was not used since minimal tuning was performed.
+The view classifier was trained only on CheXpert images (NIH images excluded from loss function), given that lateral radiographs are only present in CheXpert.
+This is to avoid unwanted bias in the model, which can occur if one class originates only from a single dataset.
 
 Validation performance as follows:
 ```
 Segmentation (Dice similarity coefficient):
-  Right Lung: 0.853
-   Left Lung: 0.844
-       Heart: 0.839
+  Right Lung: 0.957
+   Left Lung: 0.948
+       Heart: 0.943
 
 Age Prediction:
-  Mean Absolute Error: 5.42 years
+  Mean Absolute Error: 5.25 years
 
-Classification (AUC):
-  View:
-         AP: 0.999
-         PA: 0.998
-    Lateral: 1.000
-
-  Female: 0.999
+Classification:
+  View (AP, PA, lateral): 99.42% accuracy
+  Female: 0.999 AUC
 ```
 
 To use the model: