Deep MIB - Predict tab

This tab contains parameters used for efficient prediction (inference) of images and generation of semantic segmentation models.

Back to Index --> User Guide --> Menu --> Tools Menu --> Deep learning segmentation

 

collapse all on page

Contents

How to start the prediction (inference) process

Prediction (inference) requires a pretrained network, if you do not have the pretrained network, you neet to train it.
The pretrained networks can be loaded to DeepMIB and used for prediction of new datasets

To start with prediction:

• select a file with the pretrained network in the Network filename... editbox of the Network panel. Upon loading, the corresponding fields of the Train panel will be updated with the settings used for training of the loaded network
  Alternatively, a config file can be loaded using Options tab->Config files->Load
• make sure that directory with the images for prediction is correct
  Directories and Preprocessing tab->Directory with images for prediction
• make sure that the output directory for results is correct:
  Directories and Preprocessing tab->Directory with resulting images
• if needed, do file preprocessing (typically it is not needed):
  
  • select Preprocess for: Prediction ▼ in the Directories and Preprocessing tab
  • press the Preprocess button to perform data preprocessing
• finally switch back to the Predict tab and press the Predict button

 

collapse all

Settings sections

The Settings section is used to specify the main parameters using for the prediction (inference) process

Overlapping vs non-overlapping mode

Same padding in the non-overlapping mode may have vertical and horizontal artefacts, that are eliminated when the overlapping mode is used.

Dynamic masking settings and preview

• , press the button to specify parameters for dynamic masking.
  
• Masking method ▼, specify whether to keep blocks with average intensity below or above the specified in the Intensity threshold value... editbox
• Intensity threshold value... is used to specify the treshold value; patches with average intensity above or below this value will be predicted
• Inclusion threshold (0-1)... is used to specify a fraction of pixels that should be above or below the tresholding value to keep the tile for prediction
• The Eye button, hit to see the effect of the specified dynamic masking settings on a portion of the image that is currently shown in the Image View panel of MIB

Example of patch-wise segmentation with dynamic masking

Snapshot showing result of 2D patch-wise segmentation of nuclei.

• Green color patches indicate predicted locations of nuclei
• Red color patches indicate predicted locations of background
• Uncolored areas indicate patches that were skipped due to dynamic masking

List of available image formats for the model files

• MIB Model format ▼, standard formal for models in MIB. The model files have *.model extension and can be read directly to MATLAB using model = load('filename.model', '-mat'); command
• TIF compressed format ▼, a standard TIF LZW compressed file, where each pixel encodes the predicted class as 1, 2, 3, etc...
• TIF uncompressed format ▼, a standard TIF uncompressed file, where each pixel encodes the predicted class as 1, 2, 3, etc...

List of available image formats for the score files

• Do not generate ▼, skip generation of score files improving performance and minimizing disk usage
• Use AM format ▼, AmiraMesh format, compatible with MIB, Fiji or Amira
• Use Matlab non-compressed format ▼, resulting score files are generated in MATLAB uncompressed format with *.mibImg extension. The score files can be loaded to MIB or to MATLAB using model = load('filename.mibImg', '-mat'); command
• Use Matlab compressed format ▼, resulting score files are generated in MATLAB compressed format with *.mibImg extension. The score files can be loaded to MIB or to MATLAB using model = load('filename.mibImg', '-mat'); command
• Use Matlab non-compressed format (range 0-1) ▼, the score files are generated in MATLAB non-compressed format without scaling, i.e. in the range from 0 to 1. The file extension is *.mibImg and these files can not be opened in MIB. The score files can be loaded to MATLAB using model = load('filename.mat'); command

 

collapse all

Explore activations

Activations explorer brings the possibility for detailed evaliation of the network.


Here is the description of the options:

• Image ▼ has a list of all preprocessed images for prediction. Selection of an image in this list will load a patch, which is equal to the input size of the selected network
  The arrows on the right side of the dropdown can be used to load previous or next image in this list
• Layer ▼ contains a list of all layers of the selected network. Selection of a layer, starts prediction and acquiry of activation images
• Z1..., X1..., Y1..., these spinners make it possible to shift the patch across the image. Shifting of the patch does not automatically update the activation image. To update the activation image press the Update button
• Patch Z..., change the z value within the loaded activation patch, it is used only for 3D networks
• Filter Id..., change of this spinner brings various activation layers into the view
• Update press to calculate the activation images for the currently displayed patch
• Collage press to make a collage image of the current network layer activations

Snapshot with the generated collage image

 

collapse all

Preview results section

• Load images and models press this button after finishing prediction to open original images and the resulting segmentations in the currently active buffer of MIB
• Load models press this button after finishing prediction to load the resulting segmentations over already loaded image in MIB
• Load prediction scores press to load the resulting score images (probabilities) into the currently active buffer of MIB
• Evaluate segmentation. When the datasets for prediction are accompanied with ground truth labels ( requires a label file(s) in the Labels directory under Prediction images directory; it is important that the model materials names match those for the training data!).

Details of the Evaluate segmentation operation

• Press the button to calculate various precision metrics
  
• As result of the evaluation a table with the confusion matrix will be shown. The confusion matrix displays how well the predicted classes are matching classes defined in the ground truth labels. The values are scaled from 0 (bad) to 100 (excelent).
  In addition, the calculated class metrics (Accuracy, IoU, MeanBGScore) as well as the global dataset metrics are shown.
  
• In addition, it is possible to calculate occurrence of labels and Sørensen-Dice similarity coefficient in the generated and ground truth labels. These options are available from a dropdown located in the right-bottom corner of the Evaluation results window:
  
• The evaluation results can be exported to MATLAB or saved in MATLAB, Excel or CSV formats to 3_Results\PredictionImages\ResultsModels directory, see more in the Directories and Preprocessing section.
  For details of the metrics refer to MATLAB documentation for evaluatesemanticsegmentation function

Back to Index --> User Guide --> Menu --> Tools Menu --> Deep learning segmentation

Published with MATLAB® R2024a and subsequently processed by prettify_MATLAB_html V6.8b2