Update README.md

README.md

@@ -9,7 +9,7 @@ library_name: fastMONAI
 # Endometrial cancer segmentation
 This repository contains weights and exported learner (encapsulates both the model architecture and its trained parameters) for a deep learning model designed to automate the segmentation of endometrial cancer on MR images. 
 Our VIBE model utilizes a Residual U-Net architecture, trained on data derived from the study [Automated segmentation of endometrial cancer on MR images using deep learning](https://link.springer.com/content/pdf/10.1038/s41598-020-80068-9.pdf).
-The primary objective of this repository is to reproduce the results reported in the study. In addition, we have looked at improving the segmentation performance using multi-sequence MR images (T2w, VIBE, and ADC) as reported in the study [Automatic segmentation of uterine endometrial cancer on multi-sequence MRI using a convolutional neural network](https://www.nature.com/articles/s41598-021-93792-7).
+The primary objective of this repository is to reproduce the results reported in the study and to integrate this model into research PACS. In addition, we have looked at improving the segmentation performance using multi-sequence MR images (T2w, VIBE, and ADC) as reported in the study [Automatic segmentation of uterine endometrial cancer on multi-sequence MRI using a convolutional neural network](https://www.nature.com/articles/s41598-021-93792-7).
 
 ## Requirements
 Last checked and validated with fastMONAI version 0.3.9. Please ensure that you have the correct version of fastMONAI installed to guarantee the correct operation of the model.
@@ -19,8 +19,9 @@ The source code for training the model and running inference on your own data is
 Test our model live with the Gradio app for VIBE on [Hugging Face Spaces](https://skaliy-endometrial-cancer-segmentation-app.hf.space).
 
 ## Results for VIBE
-The box plot of the predictions on the validation set: 
-![](vibe_boxplot.png)
+Note that our results are not directly comparable with the results reported in [study](https://link.springer.com/content/pdf/10.1038/s41598-020-80068-9.pdf), as we opted to use the test set for validation to allocate more data to training. Unlike the approach detailed in the study, we refrained from post-processing steps, such as retaining only the largest object. 
+Our analysis indicated that this method could occasionally eliminate the tumor. As a reference, we've included the "n_components" column. 
+Below is the box plot showcasing predictions on the validation set:![](vibe_boxplot.png)
 
 The results from the validation set are also presented in the table below:
 
@@ -60,6 +61,9 @@ The results from the validation set are also presented in the table below:
 
 <b>Median DSC</b>: 0.8946, 0.8212, 0.779
 
+Prediction on a new subject in the research PACS: 
+![](research_pacs_predicition.png)
+
 ## Results for multi-sequence (T2, VIBE, and ADC)
 The box plot of the predictions on the validation set: 
 ![](t2_vibe_adc_boxplot.png)