Cancer

In the Cancer tutorial, we will use KaleCancer to combine clinical variables and pathology slide features and explore cancer research questions. The examples follow the same workflow: load the inputs, prepare a model, train it, then evaluate and interpret its predictions.

Choose an example

Example

What it demonstrates

Classification

Predict a category, compare the contribution of different inputs, and explore what the model uses.

Survival Analysis

Analyse event risk with incomplete follow-up, evaluate survival predictions, and explore individual explanations.

Exploring HANCOCK

Hands-on: answer research questions on a real head and neck cancer cohort by changing a line or two.

Classification and Survival Analysis use synthetic clinical data and generated pathology patch features. Exploring HANCOCK uses a reduced copy of the public HANCOCK cohort (CC BY 4.0), downloaded automatically by the notebook. None of these results are evidence of clinical performance.

From data to interpretation

Cancer workflow: clinical tables and pathology patch features are loaded, combined during model training, then evaluated and interpreted.

The arrows trace the existing workflow from data loading through model training to evaluation and interpretation. Motion is decorative; the complete diagram remains readable without animation and when reduced motion is requested.

What you will do

  1. Prepare and load synthetic clinical data and pathology patch features.

  2. Build and train a model for classification or survival analysis.

  3. Evaluate the model’s predictions.

  4. Interpret the results and explore how the inputs contribute.

For execution options, see Quick Start.

Use subject-level separation, fit preprocessing on training data and examine uncertainty. A successful synthetic run establishes that the workflow executes, not that it generalises to clinical use. Keep patient and confidential data off public execution services.