Coding Phase 1.1

Hello everyone!

I am back after two long weeks of end-semester examinations. Since I was unable to work during this period, it was important to get started immediately to make up for the lost time. So, I set upon the task of binarizing the originally greyscale segmented cellular images as well as extending the dataset. As opposed to 600 images previously, the dataset now contains 915 cellular images with corresponding (binarized) segmentations. 

Steps to converting an image to its 8-bit mask: 

1. Open the image using Fiji is Just ImageJ.

Segmented image before binarization

2. Click on Image → Type → 8-bit. 

3. Once the image is converted to an 8-bit format, click on Image → Adjust → Threshold.

4. A dialog box for adjusting threshold opens up. Click on the Dark Background option and set the threshold value (or, enable auto setting). Now, click Apply.

5. Finally, an 8-bit binary image is obtained as the mask for the corresponding cellular image.

After a segmented image has been binarized, it can be used to create an SBML file using the XitoSBML source code. The steps can be outlined as:

1. Run the XitoSBML source code on Eclipse IDE. (Main class: Spatial_Img_SBML.java)

2. A pop-up for Fiji is Just ImageJ and a dialog box called DomainType Namer opens up.

3. Click on File → Open to get the segmented image.

4. Choose the required domain on the DomainType Namer. In this case, the cytosol is selected.

Log window

5.  Save the resulting domain hierarchy diagram and SBML document.

Segmented image and corresponding domain diagram

Resulting SBML document

To add XitoSBML as a plug-in to Fiji is Just ImageJ, steps to be followed:

1. Open the Fiji application. Click on Help → Update. 

2. Once the update is completed, click on the Manage Update Sites option and select XitoSBML from the list. 

3. Finally, click on Apply Changes option.

With all this covered,  I am now set to start the implementation of deep learning model UNet in Java using deeplearning4j. For reference: 

1. https://github.com/eclipse/deeplearning4j/blob/master/deeplearning4j/deeplearning4j-zoo/src/main/java/org/deeplearning4j/zoo/model/UNet.java

2. https://deeplearning4j.konduit.ai/getting-started/tutorials/convolutional-networks

3. https://github.com/montardon/deeplearning4j/blob/master/src/com/tad/TrainUnetWithMembraneImages.java

4. https://github.com/montardon/unetdl4j

5. https://github.com/zhixuhao/unet

6. https://gitter.im/deeplearning4j/deeplearning4j/archives/2019/07/10

Until the next time

Ta-Ta!