Region Prediction

Region Prediction#

The Tonotopic Region Prediction feature in VASCilia identifies the cochlear region (APEX, MIDDLE, BASE) for labeled volumes using a deep learning model. This tool streamlines tonotopic regional classification by analyzing a subset of images and providing a majority-vote-based prediction.

Key Features#

  1. Deep Learning-Based Classification:

    • Utilizes a ResNet-50 model fine-tuned for cochlear region prediction.

    • Predicts one of three regions: APEX, MIDDLE, or BASE.

  2. Subset Analysis:

    • Analyzes a subset of images from the middle of the dataset for efficient and accurate region prediction.

    • Default configuration processes 13 images for majority voting.

  3. Majority Vote Decision:

    • Combines predictions from selected images to determine the most likely region.

    • Outputs the final prediction based on the most common class.

Usage Instructions#

  1. Run Region Prediction:

    • Trigger the Region Prediction feature to start the analysis.

  2. View Results:

    • The predicted region (APEX, MIDDLE, or BASE) will be displayed in a pop-up message.

Model Details#

  • Architecture: ResNet-50

  • Custom Output: Adjusted fully connected layer for three classes.

  • Input Preprocessing:

    • Resize to 256x256 pixels.

    • Center crop to 224x224 pixels.

    • Normalize using ImageNet mean and standard deviation.

Example Output#

A pop-up message shows the predicted cochlear region:

Region predicted as MIDDLE

Practical Considerations#

  • Ensure that the model file (model_region_prediction) is correctly loaded in the plugin.

  • The default configuration processes 13 images but can be adjusted as needed.

  • The method assumes that the dataset represents a continuous stack of images from a cochlea.

Extending the Functionality#

To add or modify functionality, edit the following file:

  • predict_tonotopic_region.py

—

Tonotopic region prediction Action Preprocessing Example