nph-prediction.md

NPH Prediction

NPH Prediction using 3D UNET

Overview

This module is for segmenting a CT scan of a brain into 3 parts: lateral ventricles, cerebral matter, and subarachnoid space using a 3D UNet. A Support Vector Machine is then used to determine if the subject has possible Normal Pressure Hydrocephalus (NPH).

How to Run NPH Prediction Module:

STEP 1. Login

From the BisQue Homepage, the top right-hand corner has the Sign In button.

If you do not have an account, you can create one. If you already have an account, login using your credentials.

STEP 2. Upload Image(s)

BisQue supports many of the popular medical imaging file formats, i.e. NIFTI, DICOM. We will cover two of them here.

Upload NIFTI

Once logged in successfully, you will see Upload in the top menu bar. Click on Upload and feel free to Drag-and-Drop files or Select Choose Files or Choose Directory. Navigate to the NIFTI files on your computer and select the ones you want to upload, either individual files or an entire folder/directory.

{% hint style="info" %} File Formats

The supported NIFTI file types should end with .nii or .nii.gz {% endhint %}

Example

CT35000.nii.gz
CT36000.nii
YOUR-FILENAME.nii.gz

Upload DICOM

Once logged in successfully, you will see Upload in the top menu bar. Click on Upload and feel free to Drag-and-Drop files or Select Choose Files or Choose Directory.

{% hint style="warning" %} WARNING: DICOM FILES MUST BE ZIPPED!

Before uploading your folder of DICOM files, make sure they are compressed/zipped. On Mac, two-finger click on the file and hit Compress YOUR-FOLDERNAME. This will zip the folder and allow you to upload that single zipped file to BisQue. We currently do not support uploading the raw directory to BisQue. {% endhint %}

Example

Folder of Raw DICOM.

IM-0001-0001-0001.dcm
IM-0001-0001-0002.dcm
IM-0002-0001.dcm
IM-0002-0002.dcm
IM-0002-0003.dcm
IM-0002-0004.dcm
IM-0002-0005.dcm
IM-0002-0006.dcm
IM-0002-0007.dcm
IM-0002-0008.dcm
IM-0002-0009.dcm
IM-0002-0010.dcm
IM-0002-0011.dcm
IM-0002-0012.dcm
IM-0002-0013.dcm
[...]

Zip the folder containing these files and you are good to go!

Select the option that says Several DICOM files in a package (unpack and inspect). BisQue will correctly parse the DICOM files and upload them as dataset if there are multiple images. After its uploaded, you can click on the Blue Text to be taken to your uploaded files.

STEP 3. Go To NPH Prediction Module Homepage

Once you are logged in successfully, you can access the NPH Prediction module by Clicking Here or by using the Menu bar at the top of the homepage.

From the Menu Bar. Using the Menu bar at the top of the screen, go to Analyze --> NPH Prediction. You might have to scroll down a little bit since we are adding more modules.

STEP 4. Run NPH Prediction Module

Once on the NPH Prediction homepage,

• Select an image you uploaded.

  • Click on Select an Image Button
  • It should navigate you to Resource Browser, where you should be able to select an image that is uploaded.
  • You should be able to see the selected image on the module page.

  

• Select a PyTorch Model, which is the magic box that is trained to predict the presence of NPH by looking at the input scan.

  • Click on Select file button.
  • Cick on Gear Icon on top right corner of Resource Browser. Tick the box that corresponds to Show public data.

  

  • In the search bar, type filename:*_epoch200.pt and hit Enter. This should filter out the required pytorch files. Now, select the first model in the list and hit Select.

  

• Hit RUN Button.

• It is expected to take around 60 seconds to 300 seconds of runtime, depending upon the input file size, cluster compute availability, and various other factors.

• Visualize the Results

  

• Instead of an image, one can also run the module on dataset of images.

  • Users just need to upload the images to dataset.
  • That dataset can be used as an input.
  • Users need to just click on Select a set of images intead of Select an image button.

Mask Editing

BisQue can also let users make corrections to masks predicted by the NPH model.

In this user manual, we present a toy example illustrating how one can make edits to the predicted mask.

Step A: Open Predicted Mask

• Go to Bique home page and click on images

• Select the mask that was generated recently.

  

• You should see a mask editor button at the top.

  

• Click on that button. This should enable you to edit the mask.

• As a fun example, you can go to slice-3 and draw “three” on it. You can save the edited mask.

  

• Click on “Save and Quit” Button. The edited mask should be saved on bisque (with a name that is different from the original mask). Users might have to wait for 20 seconds and edited mask (with modified name) should be displayed on the screen.

• One can click on the 3D-Viewer and verify the edits (in this case).

  

• Happy Editing!