Supervised AI Model Deployment

This tab is used to deploy an AI pipeline analysis on whole-slide MSI data.

Import Data

Click Select file… under Import MSI slide. A file dialog box will open, allowing you to browse and select the file. The supported data formats are the same as those described in the Data Import/Export section.

After the data is loaded, general information such as the number of pixels, number of ions, and the name and location of the file will appear in the text box. A TIC visualization will also appear in the viewer.

Import Model

To import a trained model, click Select file… under Import Model Pipeline and select the saved pickle file. Then click Import to load it.

After a successful load, details of the model—including model type, dataset, data split configuration, number and names of the classes, and the number and range of m/z features—will appear in the dialog box.

Preprocessing

In the Preprocessing section, users may apply data normalization, pixel aggregation, or spatial masking prior to classification.

Normalization

The available normalization approaches are the same as those described in the Normalization section of Dataset Preprocessing, with detailed formulations provided on the Functions and Formulations page.

Important

Please ensure that the normalization method matches the one used during dataset preparation for model training. Using a different normalization may lead to inconsistent or inaccurate results.

Pixel aggregation

The pixel aggregation methods are the same as those described in the Pixel Aggregation section of Dataset Preprocessing, with detailed formulations provided on the Functions and Formulations page.

Note

Unlike normalization, pixel aggregation during deployment does not need to match the settings used in dataset preparation for training. Users may adjust the aggregation level according to their requirements for spatial filtering.

Spatial masking

Users may limit model deployment to a specific region of the slide defined by a Mask.

To create a mask, first generate or select a visualization.
Click Create mask… to be redirected to the Segment Editor module.
Define a region of interest manually (similar to ROI Selection), or use other available 3D Slicer tools such as thresholding.
Use the module navigation arrow at the top of 3D Slicer to return to the MassVision module, where you can select the generated mask from the dropdown menu.

Deploy model

When satisfied with the settings, click the Deploy Model button to apply preprocessing and classification to the masked pixels of the imported MSI slide. The result will appear as a color-coded image, where each pixel is assigned a color corresponding to the predicted class from the training labels.