DASS Good: Explainable Data Mining of Spatial Cohort Data

Area of Science:

• Clinical informatics
• Machine learning in healthcare
• Radiotherapy research

Background:

• Clinical machine learning models are challenging to develop with spatial data, such as radiation dose distributions.
• Predicting long-term toxicities from radiotherapy requires integrating complex spatial information.

Purpose of the Study:

• To describe the co-design of a hybrid human-machine modeling system, DASS.
• To support the development and validation of predictive models for radiotherapy-induced toxicities.
• To augment domain knowledge with data mining for oncology applications.

Main Methods:

• DASS system co-designed with oncology and data mining experts.
• Incorporates human-in-the-loop visual steering and spatial data.
• Utilizes explainable AI to combine domain knowledge with automatic data mining.

Main Results:

• Demonstrated DASS with two clinical stratification models for head and neck cancer.
• Successfully integrated spatial data and human expertise in model development.
• Received positive feedback from domain experts on the system's utility.

Conclusions:

• DASS facilitates the creation of applicable clinical machine learning models with spatial data.
• Hybrid human-machine approach enhances predictive model development for radiotherapy.
• Design lessons learned offer insights for future collaborative AI system development in medicine.